Note that Sony clarified the confusion over their backward compatibility - www.kotaku.com.au/2020/03/sony-clarifies-ps5-backwards-compatibility-says-we-believe-most-ps4-games-will-be-playable/
Watch this video Series X destroying a RTX 2080Ti at while PC running1440P and Series X running at 4K using Microsoft DirectX 12 API: What witch craft is this ?! | Developer Details and shows Insane Power: //ruclips.net/video/VatBO3CknGQ/видео.html Original video here: ruclips.net/video/0sJ_g-aWriQ/видео.html
Another thing to consider is that in terms of expandable storage the Series X is using a proprietary storage format for playing Series X games within that expanded storage while the PS5 will let the user use an NVMe off the shelf (granted that it meets the requirements but still). Sony went with a proprietary storage format with their PS Vita and it did not go well to say the least. Seems like Sony learned that lesson at least.
The co-processors in the I/O die is very important to accessing data off the SSD. If the co processors can manage the data on the SSD, then the cpu/gpu won't have to do unnecessary arithmetic/operations on bits of data. It woud just need to manage what's already available. Beware of the next gen Zen 3/4 with this technology built into I/O dies ( *there is a patent on this from AMD already* )
Microsoft has DX12, what does Sony have? Can anyone clarify whether the current Sony consoles 'have' any flavor of DX and if not, why doesn't it seem to be a problem with current gen? Something else of note: the current premium consoles from Sony and MS have the same 1.8TF difference that the next gen consoles are reported to have. I feel this likely means the difference in performance between Xbox series x and PS5 will be much less noticeable for next gen. Looking through some of Digital Foundry's comparisons, the differences seem to largely boil down to a slightly higher detail setting on Xbox One X, with native 4k output vs 1800p checkerboarded to 4k on the PS4 Pro.
After all this time, I finally care about an upcoming console generation. All the nerd talk from Sony that most console gamers don't care about made me happy
That SSD is nuts, it'll definitely be interesting to see the types of game worlds that come from all that access and whether they'll use it as a form of expandable RAM.
All this tech and it's news made me very happy... except for the fact that apparently AMD's gpu source code got leaked/stolen. That's not good AT ALL!!!!
SSaini But even though it’s more powerful. Most developers would just code for the weaker console. So it doesn’t really matter. Unless you are only playing first-party games or sponsored games
SSaini I don’t get why you are saying last gen is last gen. I’m clearly talking about the next gem consoles. As for higher specs, nothing is stopping a refresh mid way through the console’s generation. We could see a PS5 pro, or Xbox series X extreme or smth. As for higher specs, it doesn’t mean much, again look at all the previous generations for examples, games run perfectly fine on the weakest ones, the higher specs will just allow “special” features. We are yet to see what they are, but an example from last generation was 4K 60 “native” on the Xbox one X
sadly nvida and intel quit the console game in 2001 (original xbox) Correction: nvidia is still in the game (Nintendo switch) - thx dogen nvidia is the joint developer of the PS3 GPU - Thx Pooya Ansary
That Zen2 goodness. AMD's new leadership has definitely been on a hot streak, and I saw RDNA like Zen: Showed promising potential, took a new generation to make it shine. RDNA2 I feel will be the Zen2 of GPUs. I've been seeing some insanity going on with AMD's designs, and it's got me pretty excited. Their engineers are scary when given the budget
They got rid of Raja thats what happen. It also helped they have the funds to put into RDNA 2 thanks to Sony and Microsoft. Consoles saved AMD now console helping them with gpu.
@@ZeroFallout1 Microsoft not really.. They work witu amd before and disnt give them credit and amd are demanding more from them.. So amd showing their hand for notoriety
@@elitebeing21 yeah Microsoft choice of apu helped keep the lights on while AMD was working on Zen. If not for the 2 consoles we might not of had AMD anymore. They were that close to shutting the doors. I'm just glad they have some money to work on gpu side of business now.
Not true, NAND flash suffers from Read Disturb errors. Constant reads of memory rows can affect cells in different rows of the same block. This can shift the threshold voltage of nearby cells corrupting data and potentially damaging endurance. Due to the way multiple bit per cell flash works, higher bit counts are going to be n^2 more susceptible to Read Disturb related damage. MLC is susceptible enough for it to be a noticable phenomenon under certain workloads, QLC based SSD's as used in the PS5 and XSX are exponentially more vulnerable. www.delkin.com/blog/a-closer-look-at-read-disturb-errors/ arxiv.org/abs/1805.03283
@@williamforbes6919 So-called "read disturb" is a transient problem. You have to read the same row over and over tens of thousands of times. All flash controllers mitigate this problem away automatically, by rewriting a block before it becomes susceptible to "read disturb". Streaming in a lot of assets stored at different locations of the SSD isn't going to trigger a "read disturb". Pretty much no normal operation will do it. You're not going to wear out an SSD by streaming data from it constantly.
@@TrueThanny The two modes of potential damage I'm aware of include invalid (out of spec) voltage states in erased rows of memory occuring when they are finally written to, and automatic block refreshes due to file system rules or bit error correction eating into endurance. As far as I'm aware, these drives are Samsung QLC based, and that there is going to be an extremely high volume of reads across the drive. QLC has a low number of programming cycles and requires many more block refreshes (35000 vs 1000 P/E cycles, and ~100,000 vs ~25000 between block refresh). In this specific scenario, QLC would (using magical numbers I made an educated guess on) have around 140 times the read related endurance penalties as experienced by Samsung TLC, this is combined with a hell of a lot more random reads than I personally imagine you would get in almost any other workload. Classical workloads which will run on a platter drive probably wouldn't ever experience measurable drive endurance loss due to read accesses. I'm not claiming it's a showstopper or anything that can't be engineered around, I'm extremely confident this is something they have already solved with their filesystem and drive controller. I do however think it is something that should have a tangible impact on drive endurance in this very specific scenario, and requires at least minimal efforts for mitigation.
@@TrueThanny This is assuming that they actually use it as an extension of vram and aren't just BS'ing us like they usually do. If they are embellishing their use of the drive, then it probably won't even be a measurable effect.
@@williamforbes6919 Dude I've been running an MX 500 for YEARS. That's for games, work, Hyper-V VMs etc. If anyone would have worn out an ssd in home use it's me. It runs as fast as the day I got it. Its 512 gigs and cost me $180.00 new. That's how long I've had it. Please stop spreading your FUD. SSDs have come a long way in reliability and longevity. A console won't change that.
While i agree with the assesment it really comes down to how much the PC prices will drop after prices have been announced on the consoles, as well as the actual price on consoles. This is for sure a much more interesting console generation than the last one and it will more than likely be by far the best bang for buck for gaming.
@@gismo3564 The Xbox Series X is going to be faster than a 2080 Super. Depending on a few factors, it could even be as fast as a 2080 Ti in some cases, given the latter's memory bottleneck. You will not be able to build a PC as fast for anywhere near the same amount of money. That will probably remain true after big Navi is released on desktop, unless AMD and nVidia are finally forced to cut prices somehow.
Make some room on that boat. Bought an Xbox one a little after release just to play Destiny with a friend... it's been gathering dust since about 3 months after that. After realizing how much of a cut down game Destiny was, I finally talked him into a gaming pc with some spare parts I had to help him. Good to see that consoles are finally stepping up in hardware to current pc levels instead of using bargain bin pc tech. Looking forward to see if developers actually having current tech will finally push better optimization in pc games as well. Won't hold my breath, but maybe cross my fingers.
In terms of hardware consoles and PCs are the same field of interest right now. Mark Cerny himself said they build the GPU of PS5 together with AMD and that a lot of the stuff they developed for PS5 will also end up on PC GPUs at the end of this year.
I don't care about consoles but I do care about hardware and how it progresses and in this case, consoles matter for PC gamers as the hardware stack is almost the same now.
AMD dabbling toes in high CU count with Microsoft XboX, and dabbling toes in high clock speeds with Sony PS5. Next up: AMD using what they've learned for a high-CU, high-clock desktop Navi2... and win!
@@BudgetGamerz So what? It's AMDs IP and I'm 100% sure they have a two-way licencing so that any design improvement proposed by Sony or Microsoft is owned back by AMD again.
16:50 Glad someone has picked up on this. Seen a ton of PC Gaming talking-heads talk about fast SSDs as this completely ubiquitous feature in PC gaming, when really we only have one part of the puzzle: shorter load screens. Genuinely interested to see how this develops on the PS5.
I was also worried if the jump to 16GB RAM was too little for next-gen but that alleviates a lot of the worries. It is also interesting how much further Mark Cerny wanted to push the speed of PS5 SSD because even with a regular SSD you still get a massive boost over the hard drives.
Yeah I'm very curious how its gonna be handled on the Windows side when devs start building games that require SSD speeds equal to at least xbox series x. I wouldn't be surprised if we see SSD storage and maybe even I/O speeds being listed in system requirements. Yea SSD's are nothing new on PC but like you said it's dif on the console side where the dev can count on every single user on the new consoles having SSD storage and count on exact I/O specs.
Just to clear something up: Cerny in his presentation talked about the top 100 PS4 games being able to run in boost mode on the PS5 at launch. The boost mode enhances the graphics so that the games look better or run with more FPS. Every PS4 game will be playable at launch of the PS5, it‘s just some games aren‘t able to handle the boost mode. Sony already cleared this up after the presentation.
21st he was talking about boost mode. The top 100 PS4 games will be playable in boost mode but every PS4 game will be playable in normal mode at launch. Check the statement they made after the press conference.
Enhanced graphics as in full resolution,no dynamic resolution,to get improved graphics a la Xbox one x they'd have to bake that into the silicon,they haven't ,so therefore.....they fooled you
Where? Not even that rectifying blog post is this clear. It makes sense, but I wont take anything for granted in this confusing situation. But if so, that's great! Why do they have so much communication difficulties?
@@pmpmpmpmpmpmpm Sony cleared it up, Google it it's been out for a few days already. Almost all of the 4000+ games will be available. Spin that into a negative.
Sony's higher clocks would've been fine if the difference was between 52 vs 46 CU's or something like that. 36 is just such a massive difference no ammount of clock speed would be able to level the performance.
Mateusz Wojtkiewicz I don’t agree with you at all the gpus are actually a lot closer to each other than last gen you cant just look at CUs and just say series x is the more capable machine it’s more than just tflops you guys still don’t get it the clock speed of the CUs matter and the ps5 gpu is significantly clocked higher than series x Sony went for 36CUs at 2.23ghz Microsoft went for 52CUs at 1.8ghz Sony’s gpu can outperform the series x gpu and that’s what people don’t understand
This the most honest breakdown of both consoles I’ve seen. You didn’t sugarcoat anything or tried to make things seem bigger than what they actually are and you explained everything perfectly. Great job
Hi Jim! Nice video as always, but there are just a couple things I'd like to discuss about. The short one is about the SSD: NAND doesn't wear out with reads only, at least not in a tangible way. So the idea of loading assets on the go is quite clever if they manage to translate it in a lower load for CPU and GPU. The longer one is about the node: according to TMSC there's no other node other than N7 and N7P on DUV, which are the same density and IP combatible with each other. According to Dr. Ian Cutress they have 90% of the same steps and machinery in common. Also according to Dr. Cutress, Zen2 mobile CPUs (the closest thing to next-gen console CPUs) are still on N7, which means that Series X/PS5 CPUs are about 10% more efficient at the same clocks, on average, even compared to mobile chips. Infinity Fabric on 7nm is also smaller and much more efficient as stated by AMD and all of this translates to a quite large power budget available for the GPU section only. RDNA1 already is on N7P, this means the efficiency gains all come from the architecture itself: I'd call it the "Maxwell step" much needed by AMD and I suppose it's been enabled by further distancing from the GCN structure, leaving the bare minimum for backwards compatibility. I'm pretty sure they decided to make the SoCs on N7P because of the already mentioned IP compatibility with both Zen2 and RDNA1/2 and the decreasing demand of non-EUV products from now to the next couple years, bringing the cost down by a lot at least on the AMD side. Moreover, TSMC announced "N6", an EUV node which is IP compatible with N7 and N7P which I'm sure will power the future mid gen consoles or "slim" ones. I've also done some calculations, based on the info given by Microsoft and Sony, about power consumption and die space occupied by the single sections of the SoC. RDNA2 CUs look like ~10% larger than RDNA1 ones and the 50% efficiency gain can also be confirmed quite easily with Series X power supply data. I've used the number to write a video for a friend of mine if you're interested: it's in Italian but most of the numbers are shown on screen.
@@jonathanhensley4883 I omitted all of the calculations for obvious reasons, but they're pretty straightforward once you account for every information given from Microsoft, Sony and AMD What did you find harder to understand?
@@_Agosto_ Just the tech speak in general. Went way over my head, but it was still interesting. Much better than the fanboy pissing contests just arguing about who's is better.
@@jonathanhensley4883 I just used console data to get a glimpse into next gen AMD cards, since I don't really care about either xbox or ps5. The explanation is in another video, even though it's also not 100% complete for time and convenience reasons
The assumption that NAND flash doesn't suffer from Read only related wear is incorrect. Read Disturb errors are caused by flash memory reads causing a gate bias charge degrading the threshold voltage of other unread rows of flash cells in the same block. This affect is exponentially greater in QLC memory than in MLC, where is already considered to be an issue. Because of this, read wear levelling is a commonly used technology in flash controllers, but it does require the equivalent of defragmentation to be run on a drive, making sure related data is stored in the same blocks and that no specific memory block sees a much greater number of reads than any other. arxiv.org/abs/1805.03283
@21st Games sell consoles simple as that. For the end-user first or third party is irrelevant, of course, GTA V sell more consoles than any exclusive from any of the console companies but at the same time, Marvel's Spider-man on PS4 was the best selling superhero game of all time surpassing even all the multiplat superhero games so we can see that people just go where the games are regardless of the platform.
@@weaverquest spiderman brah I'm not a child anymore their single play survival games you play one you've played all of them they keep milking the same concept over and over again
It is all about the exclusives which is why I personally am always on PC and Nintendo Switch these days because the Switch has such a huge variety of exclusives that when paired with a good gaming PC you really have access to many games. Plus I've always been a handheld enthusiast and have a bunch of different handheld consoles going all the way back to the GBC. I've had a PS4 since it came out but haven't used it much the last couple of years. Graphics are nice but the gaming experience as a whole is more important, to me anyway. As long as you're having fun and enjoying your purchases is what matters at the end of the day.
@@TerraWare You should check out more stuff on your PS4 library. There a ton of amazing games to play for very cheap at this point and it is certainly no coincidence you see Sony games at GOTY awards each year. My favourites are God of War, Horizon Zero Dawn, Bloodborne, Gravity Rush 2, Shadow of the Colossus, Persona 5 Royal, Marvel's Spider-Man, Death Stranding, Dreams, and also the upcoming Ghost of Tsushima looks super promising.
PC users can already see the benefits of an SSD in asset streaming firsthand: Get a GPU with limited vRAM (Say, 1GB or so) but a powerful enough core to play with loads of texture mods and run it on a Hard Drive, you'll likely wind up in a situation where you'll be at a decent FPS...until something spawns or you turn around too fast or the like, at which point you'll hear the ticking of a HDD head seeking and have a pause until it finishes. Move that install to even a slow SSD (I have an 840Evo) and it'll be much more fluid, you'll still get small hiccups here and there but it'll still be playable. Sims 3 also runs far faster on an SSD than a HDD, it's actually the only game I know of that generally is storage bottlenecked to the point where setting up a synced RAMDisk for certain bits of data (eg. Objects, Saves) is actually something I've considered at times when playing it...
I have a fast NVME, also an EVO 900 series SSD and a bunch of others. I put all my fave games on NVME as its massively faster but honestly I can't notice any difference. They load about the same speed and big ass games with streaming textures seem to work seamlessly regardless of what kind of solid state storage you're using. I believe the new Doom really has texture streaming down to a fine art and looks amazing coming straight from any old SSD eliminating load times completely. I'm fascinated but also dubious of Sony's claims that it can do more with faster storage than is already being done. Its gonna be fun watching it all unfold.
PREY (Bethesda game not older one) stuttered like crazy. A constant stutter/judder as I moved. One or more per second. Then I move the game to an SSD and it all went away. Turns out it was asset streaming constantly which is odd considering the game was fairly modern and I had 8GB of VRAM.
@@ClayMann You really have to game the setup to show the benefits of an SSD because by default the games usually are made to run off a slow HDD: That's why I noticed it when I was modding Skyrim on a GTX 275. The only one I've noticed that comes from the factory as HDD bottlenecked is Sims 3 but you quickly run into a single-thread CPU speed bottleneck there too.
NVMe drives, the good ones are beast, I upgraded from a 840 Evo to a Firecuda 510 and games are so smooth that I enjoy playing them. My 840 was so little in capacity, I had to use games on my 2TB HDD and games where near enough unpleasant and i didn't enjoy them as I had micro/largestuttering and hitching.
@@ClayMann Well PC got way more overhead to the hard-drives. What Sony has done is using on-die custom controllers. Fascinating if it works and if it can work, a custom eco-system is the best spot.
"...though it most likely went over the heads of most console gamers." Yeah, shortly after this presentation I was reading in various RUclips comments sections about how Cerny is a liar and how he promised this and that with the PS4, how he spent all that time talking about nothing, how SSDs don't really matter, blah blah blah. But for people who pay attention to what he actually says, it's clear that he chooses his words carefully and puts a lot of thought into the hardware design criteria. It's really interesting to see how this next generation is starting out. We've gotten unprecedented access not only to hardware specs and design criteria, but Microsoft even let DF disassemble the damn thing. This kind of transparency is great, and I think it's interesting to see how both companies are attempting to innovate their hardware in very different ways. I personally think that Sony is going to have an advantage with that I/O throughput, and I love the fact that even though the included SSD is way too small, the PS5 will support third-party NVMe drives, which will make upgrading cheaper. On the other hand, I kind of like how Microsoft's SSD design allows for easy swapping, and if they are affordable enough, I could envision just having several on hand and swapping them whenever necessary rather than having to attach slow, bulky external storage. Great analysis, Jim. I agree with most of it. I'm more excited for this generation of console than I've been in a LONG time. The technology is becoming a lot more interesting, and these machines very well could make a strong argument for going console versus a budget gaming PC.
"UPDATE: A quick update on backward compatibility - With all of the amazing games in PS4’s catalog, we’ve devoted significant efforts to enable our fans to play their favorites on PS5. We believe that the overwhelming majority of the 4,000+ PS4 titles will be playable on PS5."
Power wars aside, I really want to know more about the knock on effects of the fast SSD they seem to be alluding to. Does it save CPU cycles, and perhaps avoid the usual memory allocation steps in some cases etc etc.
@@sydius4248 Yes it seems to me as well that it's blurring the line between ram and virtual ram. It's true that there's a large bulk of data on DDR ram that isn't often used, but needs to be memory allocated because of the way a PC works. Offloading all that onto an ultra fast SSD and only load on demand would certainly free up a big chunk of that 16GB of ram which will have a noticeable effect. For instance the texture limitations might not be as big of an issue if the SSD is so fast. I think I'm beginning to piece together how this change in PC paradigm comes together on PS5. I need more detail though. Far more.
As a programmer that spent lot's of time with codecs using AVX2, I can tell you for a fact that running AVX code on, at least intel CPU, causes the CPU to limit it's clock. AVX2 doesn't run has fast has some other stuff. And that just the clock speed limit, using AVX to get things done quick and consistant requires lot of hard work, and pretty much require hand writing assembly. It's a lot of pain. Probably XBox Series X ditched that hardware, to get better clock and smaller packages. Opting to use co-processors/GPU for the gaming use case of AVX2/AVX512. I wouldn't be surprised if both Sony and microsoft ditch most of the ZEN2 fixed hardware functions. I remember a cpp talk on youtube about how much of the size of a CPU is fixed function hardware. I mean each and every core needs AVX, crypto and tons of other instruction. But most of a program/most program only plays with interger or basic floating point, so in a fast majority of cases, most of the CPU is actually idle... In console, where hardware backward compatibility can be limited, you don't need SHA-1 fixed function, opting for SHA-2 only. 3Dnow might be skipped (or partly skipped)... So microsoft probably removed some of the complex instructions that forced the CPU to be more complex and running slower, probably opting for a mix of custom instruction, co-processors and "they won't need it". Sony probably got more conservative and kept more instruction in the CPU, thus limiting the clock becomes required, based on what extension is used, and due to a more complex scheduler. But doing so probably cost a lot less, not having to overly customize the hardware has much; nor compiler stack.
I know AVX2 lowers clock speeds on Intel but Mark Papermaster said a long time ago that it wouldn't affect clocks on Zen 2. My real point though was that I just don't see why it should be an issue as AVX2 is use how often? With that said, I did consider your point on Microsoft disabling it (my thoughts being, if it were an issue then that would be the smart option). I think disabling stuff like AVX2 would actually cost more as they are working with IP blocks which already has the tech. I found it interesting that Sony didn't mention 4-bit and 8-bit support though as well.
That's not entirely accurate. For example, SHA and AES block you'd utilize for fast hashes needed in probabilistic and associative structures (hashmaps). And some cpus, especially those customized for the Super 8, maintain clocks even in AVX modes (first and most famous is the E3-2676v3). Could be the case here as well. And although you might write AVX seldom, you might use it frequently through libraries. They benefit tremendously from fast memory, hence on consumer Zen (for example) you'd often run in memory bottlenecks; which is what might contribute to ditching it or go with a mixed algorithm (64b prescan, 128b+ compute on matches). The consoles wouldn't have that, hence I expect the opposite: devs actually using AVX more! - Neither does removing fixed functions contribute to a faster cpu (pipeline deals with any such latency mismatch) nor to a cooler one (units are put asleep if not used). Indeed removing is more work ($$) as you'd need to revisit power domains.
@@EditioCastigata Ok well put it this way - I think AVX2 is a convenient excuse for Sony lacking (at least) a hundred MHz, when the far more likely reason is simply that MS spent more on cooling.
@@adoredtv I see another gamble at play here, essentially answering the question “when do enough devs learn using the new computing paradigm:” Sony might've done what's been done in the past while betting on devs are not ready yet, analyzed what's been used currently, and hence gone for increased clocks. Microsoft is betting on devs picking up the skills now (and/or they provide a rich library of examples and best-practices as offset, like Intel does!) and hence have gone wide yet remaining in the performance-power sweet spot.
Well about the load on the fly from SSD... you have to take in to consideration that there is LOD as well. So if you have say 5 LOD levels. You can dump it one or two level for the stuff that is behind you . Then read in the full LOD JIT. If the textures is really smartly configured, the same area of memory can contain all LOD, so its just a matter of how much you read of the texters to get the full load. If you read the full texture you get the full load. If you read a quarter, you get the second LOD. If you read 1/16, you get the third.. and so on. Even if you manage to turn so fast that the system can´t load it. It don´t matter, with in a fraction of a second you will have a higher LOD, and with in a aditionall fraction you will have the full LOD.
So let me get this right.We can't compare Tflops from an architecture to another because they have different IPC's and T-flops are only based on clock speeds and shaders right?
You can't even compare from one instance to another. Put the same chip on a card with a smaller memory bus or slower memory, and you'll get drastically reduced performance. TFLOPs is a computational maximum that would be reached if every shader did a calculation every single clock cycle. Anything that forces a shader to remain idle reduces the actual TFLOPs figure. Stalled front end, slower memory, slow system bus, etc. Just look at how close the 5700 XT is to the Radeon VII in a lot of games (beating it in some). That despite the former having a peak of about 9.7 TFLOPs, which the latter has a peak of about 13.4 TFLOPs. And that's with less than half the memory speed as well. The front end of RDNA is just better at keeping shaders active in games.
3rd party developers wont even use it as they will design their games the same way they are the now so they run on current PCs and XSX. Only First party PS5 developers will use it and they most likely wont even utilise it a launch either. Sony focused on the SSD and spouted their rubbish because they know they have the inferior console if it had been the other way around like the PS4 and Xbox one Sony would be driving this power gap hard but PS5 isnt so their playing the SSD card as it's all they have and it is litterly a gimmick I can see 95% of developers never even using nor caring for it. XSX has DXR ray tracing, mesh shaders, more powerful console in almost everyway etc etc.
Xbox CPU clocked higher at a consistent frequency, Xbox GPU more grunt power and at a constant frequency , (In Sony’s marketing speech , than a RTX 2060 clocked at 2.100mhz is faster than a RTX 2080 clocked at 1.800mhz.. huh? ) Xbox GDDR6 is faster than the PS5s.. Yet Sony touts and brags about how fast is their SSD.. smh. NVME and SSDs are very fast in todays standards and nothing to brag about.
This is perfect for my work-from-home confinement, as I provide support for an important retail company and over 2/3 of the stores are closed due to COVID-19, so I don't have a lot to do. The expected efficiency of RDNA2 makes me extremely hopeful for the next generation of Radeon cards. It will probably be an interesting year.
It's not just diverse setups in PC land that stop you relying on the SSD outside of level loading The thin-spine architecture of PCs mean that the secondary storage is only loosely coupled to the GPU. It's miles away on the other side of a few busses. Whereas with XSX and PS5 being fixed and custom, you can have the SSD butted right up against the APU. And then on top of that you have custom DMA units for streaming data straight into RAM. And even more then you can build systems to give developers an incredible API toolkit to leverage these devices, custom compression blocks, etc. Eventually this and the SSD will be on die or on module and the PC will not be able to do this without itself starting to look more and more like an SoC.
The biggest thing I took from the playstation reveal was how long the silhouettes sat still to the point we thought they were doing a home alone moment.
Edit: My original post was incorrect, however, the more probably configuration is in the replies. Something potentially interesting is the memory bandwidth. I was trying to figure out how the Xbox Series X was able to achieve its 320-bit, 10/6 GB configuration, with their quoted bandwidths of 560/336. Even more curious was the fact that there only appeared to be 10x GDDR6 chips in the images of the main board. It turns out that to have a 10/6 configuration for a total of 16 GB, and to also obtain the bandwidth stated, there would need to be 16 GDDR6 chips total (since they are 1GB each). That means that there are 6 additional chips on the underside of the board. Furthermore, something that DF was incorrect about was the bus width / sharing (they speculated that the 320-bit bus was shared between the 10/6 with 64-bits for the CPU). To get the 560 MB/s claimed with 10 chips at 14 GT/s, all 320-bits would have to be used. And then to get 6 GB of GDDR6 at the claimed 336 GB/s, there would be an additional 192-bits used by the "slower" memory (it's not actually slower, it's just narrower - though it could be "slower" due to a more complicated memory interconnect, especially since most of the hardware units will probably use that controller). So it would actually be 320-bits + 192-bits. Note that if for some reason the slower memory was not GDDR6, it wouldn't be possible to achieve that sort of memory bandwidth for the 6GB memory in the footprint that they would need. I would guess that the 192-bit bus are the two memory controllers closest to the CCXs, and the 320-bit controllers are the ones further away. I find it a bit strange given that configuration though, that Microsoft would specifically split them up, which would lead me to suspect that the 320-bit controllers are exclusive to the GPU CUs, or at least are indirectly accessed by the CPUs via some sort of memory bridge. Also, the actual configuration of the GDDR6 is interesting, since the chips are dual channel x16. So that means that while each chip is technically 32-bits, it has 2x address/command buses, allowing for two independent 16-bit accesses simultaneously. In theory, this could allow for the 320-bit bus to access 20 independent 16-bit "words" (so a 20-channel memory controller), while the 192-bit bus could access 12 different 16-bit "words" (so a 12-channel memory controller). Keep in mind that GDDR6 supports bursts up to 16 lines too, so while the bus width is 16-bits, a single transaction can be 256-bits per channel (i.e. 32 Bytes). It will be interesting to see how the controllers are actually configured (since you could run both channels in lock-step effectively making a 2 channel x16 chip into a 1 channel x32 chip), and what the configuration means for memory access performance in the CCXs - especially when compared to 8-core PCs which are typically only 2 channels. There isn't enough information on the PS5 to make any similar comments there, though I suspect the bandwidth will be significantly less due to the fact that memory transfers appear to go through the IO die.
The xbox only has 10 gddr6 memory chips. 6 are 2GB and 4 are 1 GB. This is like plugging in 3 8GB Dimms into a 2 channel memory controler on a CPU. The first 16 GB of memory used would run at full speed, the last 8 GB would run at half speed. In effect the memory bandwidth numbers are shared. It's not like you can add both the 560 and the 336 numbers to get the total bandwidth. The 560 GB/s is the maximum bandwidth the console has at any point.
@@draco10111b Interesting. It looks like Samsung does have 16 Gb chips, I didn't bother checking since both Micron and SK-Hynix only had 4 Gb and 8 Gb ones. And looking back at the DF video, there appear to be 6x K4ZAF325BM-HC14 and 4x K4Z80325BC-HC14, which would be what you said. It also looks like the Samsung ones are single-channel x32, which gives the 320-bit memory. But 320/8*14 = 560 GB/s. If I understand correctly, you are suggesting that the controller is in a fly-by-topology, where the closer chips have a 336 GB/s bandwidth. However, as far as I know, that's not possible with GDDR6. The Samsung datasheets aren't accessible from what I can tell, but the Micron and SK-Hynix ones specifically mention point-to-point or T-topology (in special cases) ONLY. And it's not using the T-topology method. Do you have any thoughts as to specifically why a subset would run at a slower speed in that case? The only thing that I could think of, is it's the memory controller itself for the 6 GB, however, that would make the 560 GB/s impossible.
@@hjups It's using 1GB on all 10 chips to reach 560 GB/s. The extra 1 GB on those 6 2 GB chips comes out to 336 GB/s. So total bandwidth comes out somewhere between the 336 and 560. It's not the easiest thing to explain, but it's similar to the memory layout of the GTX 970, but in that case only one of the memory chips had an effective double capacity. Not sure about the fly-by-topology stuff. This is more a feature of the memory controller.
@@draco10111b I think I understand what you are describing, but I don't really believe that it's the case - i.e. I would have to see the datasheet to be convinced. Although it does seem plausible, I don't understand why the GDDR6 would be implemented that way. Or are the 16 Gb chips actually dual channel? Micron calls their chips x16 even though they have 32 data lines (16-bits per channel - they do this by having multiple dies in the package). So if that were the case, and the 8Gb chips were single channel instead, or dual channel 16-bit, then I could see exactly what you mean. But if the 16 Gb Samsung chips only have 32 data bits on their pins, with a single CA bus, then I don't see how you could get different speeds for the different GB banks, nor could you connect the different GB banks to different memory controllers, since they would only have a single CA bus. Hopefully what I said makes sense.
The superiority of Microsoft when it comes to software is unquestionable. They are indeed a software company, and they have also stepped up on hardware. This is exciting. Sony not being sure about backwards compatibility is a joke right now.
@@evalangley3985 Majority of "Ps4 games" will be backwards comparable. Microsoft has been buying studios left and right they could catch up to Sony in terms of gaming content this upcoming generation.
@@billykotsos4642 The ability to play older games is a major selling point of PSNow. I wonder if Sony doesn't wanna go as far into backwards compatibillty because of that.
21:50 Khronos has standardised the rt extension (finally) and AMD is in on it. Expect out of the box vk_rt support on RDNA2. And I bet PS5 gnmx is going to be heavily "influenced" by Vulkan in that sense, plus their own instructions.
Ha @AdoredTV, remember a while ago on twitter I speculated that APUs would be the first chips to migrate on a higher density node this time around? Judgeing from that XBox die, I might be right.
Mug of tea in hand. watching an Adored video, Getting both information and, more importantly, insight on a Tuesday morning. You learn a lot and are also helped in coping with the lock down. Thank you Jim
@AdoredTV great video, only thing I am missing is that the PS5 clocks for the GPU and CPU seems to be the maximum BoostClocks. So I think in reality the PS5 will run the CPU at 3,2 GHz and the GPU at around 1,8-2 GHz... They just didn't want to use the "boost"-word because than everyone would know that these numbers will rarely be reached and that the PS5 in reality seems to be much slower than the new Xbox
@@Sulleei he just says that it is possible to reach this speeds, but I think it will most likely only be possible, if the game for example doesn't need much CPU power, than the GPU can clock higher or the opposite... So I think for the most AAA games, since they need both in a good amount, both will not run at full speed.
I think the majority of the time the superior ps5 ssd will be used for faster load times and im not convinced 4gb of 0.5 seconds worth of textures would be vastly better then the xsx 2gb per 0.5 seconds of textures. This brings up the question of texture quality, textures are already pretty good on current gen, I think its an odd thing to focus on for next gen, bigger graphical improvements will come from increased polycount, lighting and things like hair and cloth physics.
I don't hear a lot of talk about die size and cost. If we're talking a 10-15% difference in launch price, that could be a real dealbreaker for buying the xbox -- especially since most people won't be seeing a huge increase in fidelity of one vs the other at 10+ feet away.
Sony shouldn't be worried by DXR, vulkan has an specification for raytracing done by Kronos in colaboration with AMD and NVDIA. Also Cerny make a mistake talking about Back compat. The PS5 has 3 modes, one for PS4, other for the pro and one with PS5 clocks. The last one is the mode they tested with the top 100 of games. The other two work for all games. In the part of the chip potencial, they have coprocessors to offload some work with the io (in the case of the cpu) and the Audio part (in case of the GPU), thats why some developers talk about getting equal performance with the two consoles. The only part that could bite them in the long run is memory bandwith. I think that in the inmidiate future Microsoft wins, they have the numbers, but for the long run it's difficult to Say. Microsoft has the todays technology and Sony It's More experimental...
What I like about these consoles is that the GPU's in them both are not equivalent to the usual $150-$250 PC discrete GPU, but are looking more like a $699-$799 GPU. This could mean one of two possibilities, either the consoles are going to be very expensive, or more likely we are going to see 2080 super equivalent performance on the PC discrete GPU price drop to around $300 or below. PS4 and Xbox one were based on a custom version of a $150 GPU (bonaire), PS4 pro and xbox onex were based on an equivalent $229 GPU (Polaris 10/20). Another trend I remember is that the PC equivalent of a $499 console usually begins on console launch at roughly $800 starting build price and quickly drops (over the next 12 months) to about a $100-$150 premium over the console. So sometime in 2021 I expect that we will get a 6700/6800 series card for $200-$350, assuming the 5700 gets refreshed to around the $200-$250 mark as a 6700 and the RDNA 2.0 with RT comes in at the $300-$350 mark as the 6800. I do believe something similar happened with the Xbox 360 (X1800XL), that its GPU performance was well beaten for under $300 within 6 months (X1900GT ($350 MSRP street low price $280) www.hardwaresecrets.com/connect3d-radeon-x1900-gt-review/10/ and much much less a bit over a year later (x1950GT ($150)) hothardware.com/reviews/sapphire-radeon-x1950-gt-256-mb?page=8. One can only hope that those historical trends continue.
The new Xbox, at least, is almost certainly going to beat all sub-$1K graphics cards on the market. I have no real guesses on the console prices, but I'm sure they'll still be lower than any PC you can build with the same performance, by quite a margin. It'd be awesome if that meant lower prices at the high end when big Navi comes out (can't count on nVidia alone lowering their prices), but I'm a bit skeptical. AMD seems to be enjoying the high-GPU-price market that nVidia has created a bit too much.
@@TrueThanny The mainstream market isn't the high end and as the XSX particularly is a high end spec that you'd find on a $699-$799 GPU presently, any GPU for the PC of equivalent performance is going to have to drop to the console price within 6 months to remain viable and well below within 12 months or people will drop away from PC gaming at an increasing rate as the mid range shifts to consoles. The last couple of console generations were built on mid range technology, the Xbox 360 and PS3 were built on higher end technology for the time and we saw massive drops on the high end cards of that same generation within 6 months with PC performance just blowing consoles away within two years. We should see similar this time with RDNA 3 looking to come online by end of 2021 early 2022, RDNA 2 discrete cards should have better performance CU for CU as I'd expect better cache arrangements that would be more limited on the console SOC due to die size considerations. Big Navi is still going to be expensive, but the smaller Big Navi GPU's may just have the card we're looking for to match consoles and find 5700 cards (slower than consoles) repurposed for around $200 not long after the console launch.
If I had to choose one of them, I would choose the Xbox, because of the broader backcompat. Since several month I've tried to find and replay some of the OG Xbox games I had when I was a teen, and finding some hardware for that in working order is quite difficult. I just hope the system will get rid of region locking fot those older games.
@@rickbhattacharya2334 The console's OS is specifically tuned to work with/around the properties of GDDR6 memory, while PC OSes are designed to work with traditional DDR3/4. Since GDDR isn't used as system memory on a PC, no one is going to spend dev time to work with it.
yeah that's the wrong angle he approached it. What most devs can't do is to properly use so many threads. most games don't use more than 4 or 6 threads still. for those, the extra 200mhz will obviously be a better choice. the ones that can use all the available 14 threads can opt for that instead.
@@300maze But devs have used multi GPU cores to companstate for how shite the Jag core are in this gens consoles(PS4 & X1), SMT is common practice now. Sony have developed tools they used for PS3(ported too PS4) that allows for several Jag cores too be split up having different stuff running together.
Great anaysis Jim. Great way to enlighten my day as we all sit in isolation. As a PC gamer its been a very long time since I've been excited about consoles. Better consoles will result in better quality PC games and competitive parts, thus reducing prices. It's a win-win as I see it.
Thanks again for another top notch tech video. Currently I have neither a PS4 or a Xbox. I play all of my games on a PC. I was thinking about getting a PS5 but after watching your video, I might take a closer look at the Xbox.
Adrian Maybe because exclusives don’t actually sell consoles? Rather it’s the console itself. Why buy a ps5 when 99% of the games are available on XSX and pc? The exclusives argument never goes both ways.
ValkOptics I agree, Sony fans value the Sony exclusives far more than most. Not that they aren’t very good games but there is just as good if not better in most cases in third party and in reality Microsoft has some very good games also. The other point to buying a console and in this case the Xbox even if you own a pc is that your pc may not be graphically up to par with the Xbox and then there’s always the convenience of console gaming. Another nice thing with Microsoft’s ecosystem is in its game pass. Buy it on either and you can play it on either for no extra charge.
the 2.4GB/s of guaranteed throughput is impressive, but it's the software APIs and custom hardware built into the SoC that deliver what Microsoft believes to be a revolution - a new way of using storage to augment memory >(an area where no platform holder will be able to deliver a more traditional generational leap). The idea, in basic terms at least, is pretty straightforward - the game package that sits on storage essentially becomes extended memory, allowing 100GB of game assets stored on the SSD to be instantly accessible by the developer. It's a system that Microsoft calls the Velocity Architecture and the SSD itself is just one part of the system. "Our second component is a high-speed hardware decompression block that can deliver over 6GB/s," reveals Andrew Goossen. "This is a dedicated silicon block that offloads decompression work from the CPU and is matched to the SSD so that decompression is never a bottleneck. The decompression hardware supports Zlib for >general data and a new compression [system] called BCPack that is tailored to the GPU textures that typically >comprise the vast majority of a game's package size." The final component in the triumvirate is an extension to >DirectX - DirectStorage - a necessary upgrade bearing in mind that existing file I/O protocols are knocking on for >30 years old, and in their current form would require two Zen CPU cores simply to cover the overhead, which DirectStorage reduces to just one tenth of single core. "Plus it has other benefits," enthuses Andrew Goossen. "It's less latent and it saves a ton of CPU. With the best competitive solution, we found doing decompression software to match the SSD rate would have consumed three Zen 2 CPU cores. When you add in the IO CPU overhead, that's another two cores. So the resulting workload would have completely consumed five Zen 2 CPU cores when now it only takes a tenth of a CPU core. So in other >words, to equal the performance of a Series X at its full IO rate, you would need to build a PC with 13 Zen 2 >cores. That's seven cores dedicated for the game: one for Windows and shell and five for the IO and >decompression overhead." Asset streaming is taken to the next level, but Microsoft wasn't finished there. Last-gen, we enjoyed a 16x increase in system memory, but this time it's a mere 2x - or just 50 per cent extra if we consider Xbox One X as >the baseline. In addition to drawing more heavily upon storage to make up the shortfall, Microsoft began a >process of optimising how memory is actually used, with some startling improvements. "We observed that typically, only a small percentage of memory loaded by games was ever accessed," reveals Goossen. "This wastage comes principally from the textures. Textures are universally the biggest consumers of memory for games. However, only a fraction of the memory for each texture is typically accessed by the GPU during the scene. For example, the largest mip of a 4K texture is eight megabytes and often more, but typically only a small portion of that mip is visible in the scene and so only that small portion really needs to be read by the GPU." As textures have ballooned in size to match 4K displays, efficiency in memory utilisation has got progressively worse - something Microsoft was able to confirm by building in special monitoring hardware into Xbox One X's Scorpio Engine SoC. "From this, we found a game typically accessed at best only one-half to one->third of their allocated pages over long windows of time," says Goossen. "So if a game never had to load pages >that are ultimately never actually used, that means a 2-3x multiplier on the effective amount of physical memory, >and a 2-3x multiplier on our effective IO performance." A technique called Sampler Feedback Streaming - SFS - was built to more closely marry the memory demands of the GPU, intelligently loading in the texture mip data that's actually required with the guarantee of a lower quality mip available if the higher quality version isn't readily available, stopping GPU stalls and frame-time spikes. Bespoke hardware within the GPU is available to smooth the transition between mips, on the off-chance that the higher quality texture arrives a frame or two later. Microsoft considers these aspects of the Velocity Architecture to be a genuine game-changer, adding a multiplier to how physical memory is utilised.
This is almost word by word what Mark Cerny also talked about. Except he said it would take 8 Zen Cores without dedicated hardware probably because of the higher performance SSD of PS5. Also, Sony is using a newer compression format called Kraken that is 10% more efficient than Zlib.
@21st Hmm that might explain why on the revealed specs Microsoft got from 2.4 GB/s RAW IO Throughput to 4.8 GB/s after compression while Sony got from 5.5 GB/s raw to "only" 8-9 GB/s. However, if I remember correctly Mark Cerny said that the reason they went with Kraken was that it was already popular with developers so that might have been a good decision based on that alone.
So basically AMD just starting using compression techniques that Nvida has used for years and console applies them. Short story: With faster SSDs and newer compression techniques less information needs to be stored in ram allowing for memory being used for other features. Freeing up more space then what the initial x2 increase in memory reveals.
@@88oscuro Consoles have always used compression Mark Cerny just mentioned they are moving to a more efficient method that is trending among developers. That is not the interesting part though. It is the ultra-fast SSD combined with custom processors that handle compression without even touching the CPU and also the 12 DMA channel SSD that will enable low latency high bandwidth access that Mark Cerny thinks is the key to next generation. This video explains what he was talking about well I suggest watching it: ruclips.net/video/PW-7Y7GbsiY/видео.html
@21st BCPack only handles textures and it should also be noted that if developers optimize the Kraken compression particularly well, that 8-9 GB/s figure can go as high as 22 GB/s.
It sounds like I'm going to have to wait for the PS5 pro. If my whole console gaming collection wasn't PS1/PS2/PS3/PS4, I'd be going Xbox this round. My gaming PC is now 10 years old and would have to be rebuilt from scratch, just a 2080 by itself is more expensive, the whole build would be much more costly than $500-$600 for a console. I didn't want to have to wait but oh well I don't have time to game much anymore anyway 😱😭🤣😂
Excellent video as usual, agree with all points. Cant wait for an upcoming video on what the consoles will bring for PC. - Future of games that are coded for more than 8 Threads and ultra-fast SSDs - RDNA2 arq improvements: 50% better perf per watt in the same node?? - Xbox very early demo already at RTX2080 perf. What does this mean for the current GPU price to perf and upcoming gpu launches. - RayTracing
Don't get me wrong, I'm super excited about these consoles and what they will bring to the table but a part of me does wish we waited a year longer and got 5nm, Zen 4, RDNA 3.0 based system with probably closer to 100 CUs if 7nm can give us 56CUs on just 360mm^2. I know people will always say "if you wait for that then you will want to wait for 3nm and Zen 5" but I just think 5nm is looking to be such an amazing (and probably really long lived) node that it would be worth the wait.
@@DJ_Dopamine That is unlikely to be next year and games will always be built for the lowest denominator, ie the base console specs. They are doing well to have RDNA 2.0 in it but it would have been nice if everything was cutting edge like that, so a brand new 5nm node, new CPU, new GPU. These consoles will likely be around for at least 4 or 5 years, Zen 4 and 100CUs of RDNA 3 on 5nm in 5 years times will still be pretty sweet but Zen 2 and 36/52 CUs of RDNA 2 on 7nm in 4 years will be a little old.
I wonder if the SONY SSD is actually 1TB+ in capacity but the rest is reserved for longetivity reasons and thats why they only advertise 825GB because thats whats actually usable.
All non-bargain-basement SSD's are overprovisioned. When you buy a "1TB" SSD, you're getting 10^12 bytes of capacity, but the chips themselves are 2^40 bytes. Even mechanical hard drives are overprovisioned a bit, which is why by the time you can detect a bad sector, the drive should be replaced, since it's already remapped all the spare capacity. Just how much Sony or MS is overprovisioning is anyone's guess at this point, but the one that really needs it is MS, given their feature of suspending a game to SSD, which will drastically increase write cycles the more it's used.
The PS5 is going to have Half-Life: Alyx on the PSVR2, while Microsoft has neither game nor hardware for that. If that happens, and more games like that come out, Microsoft might miss this essential new step in gaming.
If Microsoft get their act together, they could have Minecraft and Hololense available, as they've already demonstrated it before, but they just need to release it and start X-Box's 1st-gen VR. Sony is already miles ahead, already selling over 4 Million VR headsets to Playstation owners, making them the only console owners with VR. Microsoft isn't even competing there.
Tom Guy I’m not a pony but it was, in fact a GDC presentation. That’s common knowledge. So was MS’s reveal, they just had their marketing department gussy it up for them to marvelous effect.
i would have preferred that the consoles still had a 1TB HDD, with a dedicated 256GB SSD Internal that only the devs and system can access. that way the game loads into the SSD, and you still have a huge cheap amount of storage. Also the Xboxs SSD is plenty fast enough, sony has pretty much no advantage there.
I'm actually hoping that most of the Xbox titles will come to PC also, so I won't have to buy the Xbox. Unfortunately I am a major jrpg player, and historically, those tend to be exclusive to PlayStation, so if I do buy a console it will probably have to be Sony's based on my primary game style. But I am glad to see Microsoft making a great console, because it will mean that the games will be that much better even if I get them for PC instead. Really hoping Microsoft will make all their games cross platform will PC since they own both os anyway, then I'll never have to buy an Xbox console to take advantage of the games, as long as I have a good enough PC with Windows on it.
Finally somebody that knows what they are talking about! I do recall AMD saying something about density improvement in their latest chips. I’m still curious about how effective ray tracing will be on these consoles. I don’t foresee Sony’s use of it being as good as Microsoft’s. Sony’s SSD looks really good but costly. I don’t think the cost to benefit is worth it as it’ll likely push the console to similar costs as the Xbox. It’s good news for Sony first party developers though. My biggest worries about Sony are the overuse of the very small SSD and the noise of the system as Sony is well known for having consoles that run hot and loud.
07:55 TDP isn't calculated from the base clock. It applies to the boost clocks, whether maximum boost over one or two cores, or a lower boost across all cores. A 3700X will dissipate about 65W when running at between 4.0GHz and 4.1GHz on all cores. Input power to the socket at that TDP will be 88W. So this APU at 3.6GHz will be using substantially less power.
Adshir, an open source raytracing tech startup "Localray" had recently announced that they were in "one of the next gen consoles" - I can only assume that they would be in the playstation? The demo's they have are very impressive including being able to work with smartphones. On the software side of things this gen - Microsoft has far more interesting tech. I've been waiting on HDR everything since it released on tv's, a huge win for Microsoft in this department.
Has to be, Microsoft always use their own software that is derived from one of their API like DirectX. And since it's ray tracing, Microsoft would use Nvidia against Sony and form a Team Green, due to Nvidia having Ray-tracing capable GPUs and 'sales and history'.
I feel like since the debacle that was the 2013 xbox one launch, Microsoft has been trying a little too hard to always have the highest specs and top flops. Kinda feels like a high level dick measuring contest at this point. I mean yeah, i can totally nerd out over the details and numbers, but these two companies are losing their what fees like their ''personality''? Consoles before the 7th generation all had vastly different approaches and couldn't even be compared apples to apples, which made each of them kinda unique. Nintendo kept their ''personality'' for example. Now it's all just moving close to PC architecture and it's all about who has more CU's and more RAM etc. etc. Dunno thats just my 2 cents..
PlayStation's architecture is definitely further from a PC than Xbox's is. On top of that, the emphasis on 3D Audio for every kind of setup, DualShock improvements, and PSVR, give PlayStation a much more distinct identity than the Xbox. Outside of Kinect, which is barely a thing any more, I can't even remember the last time Xbox talked about any new way to interact, only improvements to their controllers existing functions.
Being an underdog means you must bet hard, or else your platform will loose interest and slowly diminish. They got cocky after the 360, and to be fair there's always been conflicting goals within MS as to what Xbox should be. Regarding architecture, all other have perished. There's nothing else to choose.
Uh oh Sony fanboys gonna be butthurt. But, still, their SSD tech is impressive and I must say I love their approach to 3D Spatial audio using HRTFs and whatnot. After watching both "reveals" I felt from the beginning that the Series X was the better machine. The problem with the PS5 approach is of course, the lowest common denominator argument. No one but first party devs are going to be running games that require 5-9 GBs per second of bandwidth from the SSD. As cool as the tech is, most multi platform games will, at best, be designed for more mainstream SATA and NVMe SSD speeds.
I can't wait until ray tracing becomes mainstream, especially in virtual reality games. Half Life: Alyx has so many cube map reflections that break immersion.
In flat screen games, cube maps aren't as distracting to me. I think it's because I'm able to move my head physically in the game space while watching the fake reflections..well..be fake. And then there's the lack of reflections for physics items. Byah! At least give me some screen space reflections here and there.
@@yulusleonard985 Well we don't know if they will enforce that, also if this quick resume is used frequently, that is a lot of writing if switching between games or apps frequently if they store large parts of in in the SSD's.
@@prich0382 Probably for Sony, since their ssd size is weird. it coud be 1tb deliberately lockedup to 850 just so it can last for at least 5 years. Tech savvy gamers will use less than that. For Microsoft they already said they have 1tb. If retards know nothing about ssd and use up all of it it will last only 2 years. Dont know if their actual size is 1.25 TB then their ssd can last up to 10 years.
Awesome. Thanks for the vid Jim. I have been eagerly awaiting your take on the next gen consoles and you absolutely delivered, as always. Thanks again👍And please, keep you and yours safe Jim.
Digital Foundry somehow calculated the TFLOPS of the XSX GPU slightly wrong somehow, they said 12.155 when I calculated it to be 12.1472, no idea how they came to that
Note that Sony clarified the confusion over their backward compatibility -
www.kotaku.com.au/2020/03/sony-clarifies-ps5-backwards-compatibility-says-we-believe-most-ps4-games-will-be-playable/
Watch this video Series X destroying a RTX 2080Ti at while PC running1440P and Series X running at 4K using Microsoft DirectX 12 API: What witch craft is this ?! | Developer Details and shows Insane Power: //ruclips.net/video/VatBO3CknGQ/видео.html
Original video here: ruclips.net/video/0sJ_g-aWriQ/видео.html
Another thing to consider is that in terms of expandable storage the Series X is using a proprietary storage format for playing Series X games within that expanded storage while the PS5 will let the user use an NVMe off the shelf (granted that it meets the requirements but still).
Sony went with a proprietary storage format with their PS Vita and it did not go well to say the least. Seems like Sony learned that lesson at least.
The co-processors in the I/O die is very important to accessing data off the SSD. If the co processors can manage the data on the SSD, then the cpu/gpu won't have to do unnecessary arithmetic/operations on bits of data. It woud just need to manage what's already available. Beware of the next gen Zen 3/4 with this technology built into I/O dies ( *there is a patent on this from AMD already* )
Archived link: archive.is/ql0JK
Microsoft has DX12, what does Sony have? Can anyone clarify whether the current Sony consoles 'have' any flavor of DX and if not, why doesn't it seem to be a problem with current gen? Something else of note: the current premium consoles from Sony and MS have the same 1.8TF difference that the next gen consoles are reported to have. I feel this likely means the difference in performance between Xbox series x and PS5 will be much less noticeable for next gen.
Looking through some of Digital Foundry's comparisons, the differences seem to largely boil down to a slightly higher detail setting on Xbox One X, with native 4k output vs 1800p checkerboarded to 4k on the PS4 Pro.
After all this time, I finally care about an upcoming console generation. All the nerd talk from Sony that most console gamers don't care about made me happy
That SSD is nuts, it'll definitely be interesting to see the types of game worlds that come from all that access and whether they'll use it as a form of expandable RAM.
All this tech and it's news made me very happy... except for the fact that apparently AMD's gpu source code got leaked/stolen. That's not good AT ALL!!!!
@@TestarossaF110 apparently it's useless code that can't be taken advantage of, idk tho
@@unoriginalyoutubename8761 let's hope so!! Would be fantastic news.
These announcements are exciting for both console fans and PCMR-heads.
Consoles: More info on new Xbox and PS.
PC: More info on AMD's RDNA 2 systems.
Everyone who says that tflops do no matter and XSX with 40% more CU's does matter needs to STFU. The XSX is way more powerful than the PS5.
@@ssaini5028 Hardware aside, I'm really excited to watch you guys fight over the consoles every day it's top entertainment
SSaini But even though it’s more powerful. Most developers would just code for the weaker console. So it doesn’t really matter. Unless you are only playing first-party games or sponsored games
@@chrisvicera6696 When you look at the life span of a console higher specs are better to have. Last gen is last gen, it has no bearing on the future.
SSaini I don’t get why you are saying last gen is last gen. I’m clearly talking about the next gem consoles. As for higher specs, nothing is stopping a refresh mid way through the console’s generation. We could see a PS5 pro, or Xbox series X extreme or smth. As for higher specs, it doesn’t mean much, again look at all the previous generations for examples, games run perfectly fine on the weakest ones, the higher specs will just allow “special” features. We are yet to see what they are, but an example from last generation was 4K 60 “native” on the Xbox one X
Console winner: AMD
sadly nvida and intel quit the console game in 2001 (original xbox)
Correction: nvidia is still in the game (Nintendo switch) - thx dogen
nvidia is the joint developer of the PS3 GPU - Thx Pooya Ansary
Only if the fangirls on both sides would not understand that
@@choitszho switch is tegra based
@@yazata1915 opps, i made a mistake, thx for correction.
Winner: TSMC
Holyshit, RDNA2 may actually live up to the hype. Wtf has AMD been cooking?
That Zen2 goodness. AMD's new leadership has definitely been on a hot streak, and I saw RDNA like Zen: Showed promising potential, took a new generation to make it shine. RDNA2 I feel will be the Zen2 of GPUs. I've been seeing some insanity going on with AMD's designs, and it's got me pretty excited. Their engineers are scary when given the budget
They got rid of Raja thats what happen. It also helped they have the funds to put into RDNA 2 thanks to Sony and Microsoft. Consoles saved AMD now console helping them with gpu.
@@ZeroFallout1 Microsoft not really.. They work witu amd before and disnt give them credit and amd are demanding more from them.. So amd showing their hand for notoriety
@@elitebeing21 yeah Microsoft choice of apu helped keep the lights on while AMD was working on Zen. If not for the 2 consoles we might not of had AMD anymore. They were that close to shutting the doors. I'm just glad they have some money to work on gpu side of business now.
@@ZeroFallout1 when you put it like that. I agree.
19:37 SSD's have limited write cycles, not limited read cycles. Constant read access on an SSD will have no real impact on its lifetime.
Not true, NAND flash suffers from Read Disturb errors. Constant reads of memory rows can affect cells in different rows of the same block. This can shift the threshold voltage of nearby cells corrupting data and potentially damaging endurance. Due to the way multiple bit per cell flash works, higher bit counts are going to be n^2 more susceptible to Read Disturb related damage. MLC is susceptible enough for it to be a noticable phenomenon under certain workloads, QLC based SSD's as used in the PS5 and XSX are exponentially more vulnerable.
www.delkin.com/blog/a-closer-look-at-read-disturb-errors/
arxiv.org/abs/1805.03283
@@williamforbes6919 So-called "read disturb" is a transient problem. You have to read the same row over and over tens of thousands of times. All flash controllers mitigate this problem away automatically, by rewriting a block before it becomes susceptible to "read disturb".
Streaming in a lot of assets stored at different locations of the SSD isn't going to trigger a "read disturb". Pretty much no normal operation will do it.
You're not going to wear out an SSD by streaming data from it constantly.
@@TrueThanny The two modes of potential damage I'm aware of include invalid (out of spec) voltage states in erased rows of memory occuring when they are finally written to, and automatic block refreshes due to file system rules or bit error correction eating into endurance.
As far as I'm aware, these drives are Samsung QLC based, and that there is going to be an extremely high volume of reads across the drive. QLC has a low number of programming cycles and requires many more block refreshes (35000 vs 1000 P/E cycles, and ~100,000 vs ~25000 between block refresh).
In this specific scenario, QLC would (using magical numbers I made an educated guess on) have around 140 times the read related endurance penalties as experienced by Samsung TLC, this is combined with a hell of a lot more random reads than I personally imagine you would get in almost any other workload. Classical workloads which will run on a platter drive probably wouldn't ever experience measurable drive endurance loss due to read accesses.
I'm not claiming it's a showstopper or anything that can't be engineered around, I'm extremely confident this is something they have already solved with their filesystem and drive controller. I do however think it is something that should have a tangible impact on drive endurance in this very specific scenario, and requires at least minimal efforts for mitigation.
@@TrueThanny This is assuming that they actually use it as an extension of vram and aren't just BS'ing us like they usually do. If they are embellishing their use of the drive, then it probably won't even be a measurable effect.
@@williamforbes6919 Dude I've been running an MX 500 for YEARS. That's for games, work, Hyper-V VMs etc. If anyone would have worn out an ssd in home use it's me. It runs as fast as the day I got it. Its 512 gigs and cost me $180.00 new. That's how long I've had it. Please stop spreading your FUD. SSDs have come a long way in reliability and longevity. A console won't change that.
There's no way you'll be able to build a pc with similar performance for the same price at launch this time around lul
Not even close.
While i agree with the assesment it really comes down to how much the PC prices will drop after prices have been announced on the consoles, as well as the actual price on consoles. This is for sure a much more interesting console generation than the last one and it will more than likely be by far the best bang for buck for gaming.
Prices of consoles and next gen pc parts TBD still.
Could the new consoles force a drop in overpriced pc hardware ..... let's hope so.
@@gismo3564 The Xbox Series X is going to be faster than a 2080 Super. Depending on a few factors, it could even be as fast as a 2080 Ti in some cases, given the latter's memory bottleneck.
You will not be able to build a PC as fast for anywhere near the same amount of money. That will probably remain true after big Navi is released on desktop, unless AMD and nVidia are finally forced to cut prices somehow.
this happens every console launch
I keep telling myself I don t care about consoles, but my curiosity for hardware disagrees. thanks Jim for the vid!
Make some room on that boat. Bought an Xbox one a little after release just to play Destiny with a friend... it's been gathering dust since about 3 months after that. After realizing how much of a cut down game Destiny was, I finally talked him into a gaming pc with some spare parts I had to help him. Good to see that consoles are finally stepping up in hardware to current pc levels instead of using bargain bin pc tech. Looking forward to see if developers actually having current tech will finally push better optimization in pc games as well. Won't hold my breath, but maybe cross my fingers.
In terms of hardware consoles and PCs are the same field of interest right now. Mark Cerny himself said they build the GPU of PS5 together with AMD and that a lot of the stuff they developed for PS5 will also end up on PC GPUs at the end of this year.
I don't care about consoles but I do care about hardware and how it progresses and in this case, consoles matter for PC gamers as the hardware stack is almost the same now.
AMD dabbling toes in high CU count with Microsoft XboX, and dabbling toes in high clock speeds with Sony PS5.
Next up: AMD using what they've learned for a high-CU, high-clock desktop Navi2... and win!
Fabian Tschopp This is my interest / next pc
With a 300 watt TDP you should easily get both, long live RDNA 2
These designs are owned by Microsoft and Sony
@@BudgetGamerz So what? It's AMDs IP and I'm 100% sure they have a two-way licencing so that any design improvement proposed by Sony or Microsoft is owned back by AMD again.
@@naibaf710 It's Microsoft's and Sony's IP. What tech from PS4 and Xbox one have you seen in PC? Direct X? Lol
16:50 Glad someone has picked up on this. Seen a ton of PC Gaming talking-heads talk about fast SSDs as this completely ubiquitous feature in PC gaming, when really we only have one part of the puzzle: shorter load screens. Genuinely interested to see how this develops on the PS5.
I was also worried if the jump to 16GB RAM was too little for next-gen but that alleviates a lot of the worries. It is also interesting how much further Mark Cerny wanted to push the speed of PS5 SSD because even with a regular SSD you still get a massive boost over the hard drives.
Yeah I'm very curious how its gonna be handled on the Windows side when devs start building games that require SSD speeds equal to at least xbox series x. I wouldn't be surprised if we see SSD storage and maybe even I/O speeds being listed in system requirements. Yea SSD's are nothing new on PC but like you said it's dif on the console side where the dev can count on every single user on the new consoles having SSD storage and count on exact I/O specs.
Just what I want to hear on a Tuesday morning - Scotsman talking about Computer Tech
Any morning, if it gives me an excuse not to work!
Tuesday night here. This is the stuff I fall asleep too!
Ai
@@nathans5259 What do you mean?
@@jimmyjango5213 Ai is yes in Scotland
Just to clear something up: Cerny in his presentation talked about the top 100 PS4 games being able to run in boost mode on the PS5 at launch. The boost mode enhances the graphics so that the games look better or run with more FPS. Every PS4 game will be playable at launch of the PS5, it‘s just some games aren‘t able to handle the boost mode. Sony already cleared this up after the presentation.
21st he was talking about boost mode. The top 100 PS4 games will be playable in boost mode but every PS4 game will be playable in normal mode at launch. Check the statement they made after the press conference.
Enhanced graphics as in full resolution,no dynamic resolution,to get improved graphics a la Xbox one x they'd have to bake that into the silicon,they haven't ,so therefore.....they fooled you
Where? Not even that rectifying blog post is this clear. It makes sense, but I wont take anything for granted in this confusing situation. But if so, that's great! Why do they have so much communication difficulties?
Stop spreading that all PS4 games will be playable at launch. This was not specified during Cerny's keynote or in the official PS blog.
@@pmpmpmpmpmpmpm Sony cleared it up, Google it it's been out for a few days already. Almost all of the 4000+ games will be available. Spin that into a negative.
Sony's higher clocks would've been fine if the difference was between 52 vs 46 CU's or something like that. 36 is just such a massive difference no ammount of clock speed would be able to level the performance.
Mateusz Wojtkiewicz I don’t agree with you at all the gpus are actually a lot closer to each other than last gen you cant just look at CUs and just say series x is the more capable machine it’s more than just tflops you guys still don’t get it the clock speed of the CUs matter and the ps5 gpu is significantly clocked higher than series x Sony went for 36CUs at 2.23ghz Microsoft went for 52CUs at 1.8ghz Sony’s gpu can outperform the series x gpu and that’s what people don’t understand
@@chasethechosenone8478
I'm sorry for my lack of tech fortitude, but can you elaborate any further on your comment?
@@mutalix He can't cause he's full of crap!
@@chasethechosenone8478 sorry bro but you don't make sense.
SUPER USER or maybe you just lack understanding
"Bang per flop" has overtaken "butterdoughnut" as my favorite Jim-ism
This the most honest breakdown of both consoles I’ve seen. You didn’t sugarcoat anything or tried to make things seem bigger than what they actually are and you explained everything perfectly. Great job
@@evalangley3985 and what's wrong with you?
Hi Jim! Nice video as always, but there are just a couple things I'd like to discuss about.
The short one is about the SSD: NAND doesn't wear out with reads only, at least not in a tangible way. So the idea of loading assets on the go is quite clever if they manage to translate it in a lower load for CPU and GPU.
The longer one is about the node: according to TMSC there's no other node other than N7 and N7P on DUV, which are the same density and IP combatible with each other. According to Dr. Ian Cutress they have 90% of the same steps and machinery in common.
Also according to Dr. Cutress, Zen2 mobile CPUs (the closest thing to next-gen console CPUs) are still on N7, which means that Series X/PS5 CPUs are about 10% more efficient at the same clocks, on average, even compared to mobile chips. Infinity Fabric on 7nm is also smaller and much more efficient as stated by AMD and all of this translates to a quite large power budget available for the GPU section only.
RDNA1 already is on N7P, this means the efficiency gains all come from the architecture itself: I'd call it the "Maxwell step" much needed by AMD and I suppose it's been enabled by further distancing from the GCN structure, leaving the bare minimum for backwards compatibility. I'm pretty sure they decided to make the SoCs on N7P because of the already mentioned IP compatibility with both Zen2 and RDNA1/2 and the decreasing demand of non-EUV products from now to the next couple years, bringing the cost down by a lot at least on the AMD side. Moreover, TSMC announced "N6", an EUV node which is IP compatible with N7 and N7P which I'm sure will power the future mid gen consoles or "slim" ones.
I've also done some calculations, based on the info given by Microsoft and Sony, about power consumption and die space occupied by the single sections of the SoC. RDNA2 CUs look like ~10% larger than RDNA1 ones and the 50% efficiency gain can also be confirmed quite easily with Series X power supply data. I've used the number to write a video for a friend of mine if you're interested: it's in Italian but most of the numbers are shown on screen.
I understood very little of this, but it was an ineresting read just the same.😂
@@jonathanhensley4883 I omitted all of the calculations for obvious reasons, but they're pretty straightforward once you account for every information given from Microsoft, Sony and AMD
What did you find harder to understand?
@@_Agosto_ Just the tech speak in general. Went way over my head, but it was still interesting. Much better than the fanboy pissing contests just arguing about who's is better.
@@jonathanhensley4883 I just used console data to get a glimpse into next gen AMD cards, since I don't really care about either xbox or ps5.
The explanation is in another video, even though it's also not 100% complete for time and convenience reasons
The assumption that NAND flash doesn't suffer from Read only related wear is incorrect. Read Disturb errors are caused by flash memory reads causing a gate bias charge degrading the threshold voltage of other unread rows of flash cells in the same block. This affect is exponentially greater in QLC memory than in MLC, where is already considered to be an issue.
Because of this, read wear levelling is a commonly used technology in flash controllers, but it does require the equivalent of defragmentation to be run on a drive, making sure related data is stored in the same blocks and that no specific memory block sees a much greater number of reads than any other.
arxiv.org/abs/1805.03283
Platform exclusives are what sell consoles, but in all honesty Microsoft has done a better job I think.
@21st Games sell consoles simple as that. For the end-user first or third party is irrelevant, of course, GTA V sell more consoles than any exclusive from any of the console companies but at the same time, Marvel's Spider-man on PS4 was the best selling superhero game of all time surpassing even all the multiplat superhero games so we can see that people just go where the games are regardless of the platform.
@@weaverquest spiderman brah I'm not a child anymore their single play survival games you play one you've played all of them they keep milking the same concept over and over again
Not with that tarded ram config. Devs are not going to be happy later on.
It is all about the exclusives which is why I personally am always on PC and Nintendo Switch these days because the Switch has such a huge variety of exclusives that when paired with a good gaming PC you really have access to many games. Plus I've always been a handheld enthusiast and have a bunch of different handheld consoles going all the way back to the GBC. I've had a PS4 since it came out but haven't used it much the last couple of years. Graphics are nice but the gaming experience as a whole is more important, to me anyway. As long as you're having fun and enjoying your purchases is what matters at the end of the day.
@@TerraWare You should check out more stuff on your PS4 library. There a ton of amazing games to play for very cheap at this point and it is certainly no coincidence you see Sony games at GOTY awards each year. My favourites are God of War, Horizon Zero Dawn, Bloodborne, Gravity Rush 2, Shadow of the Colossus, Persona 5 Royal, Marvel's Spider-Man, Death Stranding, Dreams, and also the upcoming Ghost of Tsushima looks super promising.
PC users can already see the benefits of an SSD in asset streaming firsthand: Get a GPU with limited vRAM (Say, 1GB or so) but a powerful enough core to play with loads of texture mods and run it on a Hard Drive, you'll likely wind up in a situation where you'll be at a decent FPS...until something spawns or you turn around too fast or the like, at which point you'll hear the ticking of a HDD head seeking and have a pause until it finishes. Move that install to even a slow SSD (I have an 840Evo) and it'll be much more fluid, you'll still get small hiccups here and there but it'll still be playable. Sims 3 also runs far faster on an SSD than a HDD, it's actually the only game I know of that generally is storage bottlenecked to the point where setting up a synced RAMDisk for certain bits of data (eg. Objects, Saves) is actually something I've considered at times when playing it...
I have a fast NVME, also an EVO 900 series SSD and a bunch of others. I put all my fave games on NVME as its massively faster but honestly I can't notice any difference. They load about the same speed and big ass games with streaming textures seem to work seamlessly regardless of what kind of solid state storage you're using. I believe the new Doom really has texture streaming down to a fine art and looks amazing coming straight from any old SSD eliminating load times completely. I'm fascinated but also dubious of Sony's claims that it can do more with faster storage than is already being done. Its gonna be fun watching it all unfold.
PREY (Bethesda game not older one) stuttered like crazy. A constant stutter/judder as I moved. One or more per second. Then I move the game to an SSD and it all went away. Turns out it was asset streaming constantly which is odd considering the game was fairly modern and I had 8GB of VRAM.
@@ClayMann You really have to game the setup to show the benefits of an SSD because by default the games usually are made to run off a slow HDD: That's why I noticed it when I was modding Skyrim on a GTX 275. The only one I've noticed that comes from the factory as HDD bottlenecked is Sims 3 but you quickly run into a single-thread CPU speed bottleneck there too.
NVMe drives, the good ones are beast, I upgraded from a 840 Evo to a Firecuda 510 and games are so smooth that I enjoy playing them. My 840 was so little in capacity, I had to use games on my 2TB HDD and games where near enough unpleasant and i didn't enjoy them as I had micro/largestuttering and hitching.
@@ClayMann Well PC got way more overhead to the hard-drives. What Sony has done is using on-die custom controllers. Fascinating if it works and if it can work, a custom eco-system is the best spot.
"...though it most likely went over the heads of most console gamers."
Yeah, shortly after this presentation I was reading in various RUclips comments sections about how Cerny is a liar and how he promised this and that with the PS4, how he spent all that time talking about nothing, how SSDs don't really matter, blah blah blah. But for people who pay attention to what he actually says, it's clear that he chooses his words carefully and puts a lot of thought into the hardware design criteria.
It's really interesting to see how this next generation is starting out. We've gotten unprecedented access not only to hardware specs and design criteria, but Microsoft even let DF disassemble the damn thing. This kind of transparency is great, and I think it's interesting to see how both companies are attempting to innovate their hardware in very different ways. I personally think that Sony is going to have an advantage with that I/O throughput, and I love the fact that even though the included SSD is way too small, the PS5 will support third-party NVMe drives, which will make upgrading cheaper. On the other hand, I kind of like how Microsoft's SSD design allows for easy swapping, and if they are affordable enough, I could envision just having several on hand and swapping them whenever necessary rather than having to attach slow, bulky external storage.
Great analysis, Jim. I agree with most of it. I'm more excited for this generation of console than I've been in a LONG time. The technology is becoming a lot more interesting, and these machines very well could make a strong argument for going console versus a budget gaming PC.
"UPDATE: A quick update on backward compatibility - With all of the amazing games in PS4’s catalog, we’ve devoted significant efforts to enable our fans to play their favorites on PS5. We believe that the overwhelming majority of the 4,000+ PS4 titles will be playable on PS5."
Dilly Vesper wait, what about PS2 and PS3!?
It baffles me how many Japanese devs are still tying game logic / physics / audio to framerate, and have their games broken with faster hardware.
In any war, the victor is the one who supplies the weapons
I've listened to a few comparisons between both consoles but the answer was always up in the air,
Happy to see you tackle this particular topic Jim.
Power wars aside, I really want to know more about the knock on effects of the fast SSD they seem to be alluding to. Does it save CPU cycles, and perhaps avoid the usual memory allocation steps in some cases etc etc.
Mark Cerny covered a lot of this in his presentation.
@@jc_dogen Beyond that.
@@StraightcheD This video seems to have some info from developers ruclips.net/video/vqDHY7geZ00/видео.html.
It will function as Virtual Ram.
@@sydius4248 Yes it seems to me as well that it's blurring the line between ram and virtual ram. It's true that there's a large bulk of data on DDR ram that isn't often used, but needs to be memory allocated because of the way a PC works. Offloading all that onto an ultra fast SSD and only load on demand would certainly free up a big chunk of that 16GB of ram which will have a noticeable effect.
For instance the texture limitations might not be as big of an issue if the SSD is so fast. I think I'm beginning to piece together how this change in PC paradigm comes together on PS5. I need more detail though. Far more.
Love your videos Jim!
Cheers bud.
As a programmer that spent lot's of time with codecs using AVX2, I can tell you for a fact that running AVX code on, at least intel CPU, causes the CPU to limit it's clock. AVX2 doesn't run has fast has some other stuff. And that just the clock speed limit, using AVX to get things done quick and consistant requires lot of hard work, and pretty much require hand writing assembly. It's a lot of pain.
Probably XBox Series X ditched that hardware, to get better clock and smaller packages. Opting to use co-processors/GPU for the gaming use case of AVX2/AVX512.
I wouldn't be surprised if both Sony and microsoft ditch most of the ZEN2 fixed hardware functions. I remember a cpp talk on youtube about how much of the size of a CPU is fixed function hardware. I mean each and every core needs AVX, crypto and tons of other instruction. But most of a program/most program only plays with interger or basic floating point, so in a fast majority of cases, most of the CPU is actually idle...
In console, where hardware backward compatibility can be limited, you don't need SHA-1 fixed function, opting for SHA-2 only. 3Dnow might be skipped (or partly skipped)...
So microsoft probably removed some of the complex instructions that forced the CPU to be more complex and running slower, probably opting for a mix of custom instruction, co-processors and "they won't need it". Sony probably got more conservative and kept more instruction in the CPU, thus limiting the clock becomes required, based on what extension is used, and due to a more complex scheduler. But doing so probably cost a lot less, not having to overly customize the hardware has much; nor compiler stack.
I know AVX2 lowers clock speeds on Intel but Mark Papermaster said a long time ago that it wouldn't affect clocks on Zen 2.
My real point though was that I just don't see why it should be an issue as AVX2 is use how often? With that said, I did consider your point on Microsoft disabling it (my thoughts being, if it were an issue then that would be the smart option).
I think disabling stuff like AVX2 would actually cost more as they are working with IP blocks which already has the tech. I found it interesting that Sony didn't mention 4-bit and 8-bit support though as well.
That's not entirely accurate. For example, SHA and AES block you'd utilize for fast hashes needed in probabilistic and associative structures (hashmaps). And some cpus, especially those customized for the Super 8, maintain clocks even in AVX modes (first and most famous is the E3-2676v3). Could be the case here as well. And although you might write AVX seldom, you might use it frequently through libraries. They benefit tremendously from fast memory, hence on consumer Zen (for example) you'd often run in memory bottlenecks; which is what might contribute to ditching it or go with a mixed algorithm (64b prescan, 128b+ compute on matches). The consoles wouldn't have that, hence I expect the opposite: devs actually using AVX more! - Neither does removing fixed functions contribute to a faster cpu (pipeline deals with any such latency mismatch) nor to a cooler one (units are put asleep if not used). Indeed removing is more work ($$) as you'd need to revisit power domains.
@@EditioCastigata Ok well put it this way - I think AVX2 is a convenient excuse for Sony lacking (at least) a hundred MHz, when the far more likely reason is simply that MS spent more on cooling.
@@adoredtv I see another gamble at play here, essentially answering the question “when do enough devs learn using the new computing paradigm:” Sony might've done what's been done in the past while betting on devs are not ready yet, analyzed what's been used currently, and hence gone for increased clocks. Microsoft is betting on devs picking up the skills now (and/or they provide a rich library of examples and best-practices as offset, like Intel does!) and hence have gone wide yet remaining in the performance-power sweet spot.
3Dnow is phased out since Bulldozer.
Fantastic video as usual. Been having a rough few days and I really needed this. Thanks Jim!
bro, your videos are always top notch and high quality. Appreciate it man.
Well about the load on the fly from SSD... you have to take in to consideration that there is LOD as well. So if you have say 5 LOD levels. You can dump it one or two level for the stuff that is behind you . Then read in the full LOD JIT.
If the textures is really smartly configured, the same area of memory can contain all LOD, so its just a matter of how much you read of the texters to get the full load. If you read the full texture you get the full load. If you read a quarter, you get the second LOD. If you read 1/16, you get the third.. and so on.
Even if you manage to turn so fast that the system can´t load it. It don´t matter, with in a fraction of a second you will have a higher LOD, and with in a aditionall fraction you will have the full LOD.
Can hide the progressive loading behind motion blur as well.
So let me get this right.We can't compare Tflops from an architecture to another because they have different IPC's and T-flops are only based on clock speeds and shaders right?
Yes, basically.
You can't even compare from one instance to another. Put the same chip on a card with a smaller memory bus or slower memory, and you'll get drastically reduced performance. TFLOPs is a computational maximum that would be reached if every shader did a calculation every single clock cycle. Anything that forces a shader to remain idle reduces the actual TFLOPs figure. Stalled front end, slower memory, slow system bus, etc.
Just look at how close the 5700 XT is to the Radeon VII in a lot of games (beating it in some). That despite the former having a peak of about 9.7 TFLOPs, which the latter has a peak of about 13.4 TFLOPs. And that's with less than half the memory speed as well. The front end of RDNA is just better at keeping shaders active in games.
Sony SSD promises sounds like the old ESRAM thing Microsoft had in the original Xbox one. Ended up a total fizzer.
3rd party developers wont even use it as they will design their games the same way they are the now so they run on current PCs and XSX. Only First party PS5 developers will use it and they most likely wont even utilise it a launch either. Sony focused on the SSD and spouted their rubbish because they know they have the inferior console if it had been the other way around like the PS4 and Xbox one Sony would be driving this power gap hard but PS5 isnt so their playing the SSD card as it's all they have and it is litterly a gimmick I can see 95% of developers never even using nor caring for it. XSX has DXR ray tracing, mesh shaders, more powerful console in almost everyway etc etc.
Xbox CPU clocked higher at a consistent frequency, Xbox GPU more grunt power and at a constant frequency , (In Sony’s marketing speech , than a RTX 2060 clocked at 2.100mhz is faster than a RTX 2080 clocked at 1.800mhz.. huh? ) Xbox GDDR6 is faster than the PS5s.. Yet Sony touts and brags about how fast is their SSD.. smh. NVME and SSDs are very fast in todays standards and nothing to brag about.
So, on the Xbox, does that mean if you used the two memory block in parallel you'd get 896 Gbps?
This is perfect for my work-from-home confinement, as I provide support for an important retail company and over 2/3 of the stores are closed due to COVID-19, so I don't have a lot to do.
The expected efficiency of RDNA2 makes me extremely hopeful for the next generation of Radeon cards. It will probably be an interesting year.
I am on the same boat with you, work for a opera production house, everything cancelled till June.
It's not just diverse setups in PC land that stop you relying on the SSD outside of level loading
The thin-spine architecture of PCs mean that the secondary storage is only loosely coupled to the GPU.
It's miles away on the other side of a few busses.
Whereas with XSX and PS5 being fixed and custom, you can have the SSD butted right up against the APU.
And then on top of that you have custom DMA units for streaming data straight into RAM.
And even more then you can build systems to give developers an incredible API toolkit to leverage these devices, custom compression blocks, etc. Eventually this and the SSD will be on die or on module and the PC will not be able to do this without itself starting to look more and more like an SoC.
I'd imagine they'd be using VulkanRT seeing as it's an open standard and has also just been updated IIRC.
Somehow one of the shortest 30+ minute videos I've ever watched. Loved every minute though.
Best breakdown of both console specs I've seen so far
The biggest thing I took from the playstation reveal was how long the silhouettes sat still to the point we thought they were doing a home alone moment.
Edit: My original post was incorrect, however, the more probably configuration is in the replies.
Something potentially interesting is the memory bandwidth. I was trying to figure out how the Xbox Series X was able to achieve its 320-bit, 10/6 GB configuration, with their quoted bandwidths of 560/336. Even more curious was the fact that there only appeared to be 10x GDDR6 chips in the images of the main board. It turns out that to have a 10/6 configuration for a total of 16 GB, and to also obtain the bandwidth stated, there would need to be 16 GDDR6 chips total (since they are 1GB each). That means that there are 6 additional chips on the underside of the board. Furthermore, something that DF was incorrect about was the bus width / sharing (they speculated that the 320-bit bus was shared between the 10/6 with 64-bits for the CPU). To get the 560 MB/s claimed with 10 chips at 14 GT/s, all 320-bits would have to be used. And then to get 6 GB of GDDR6 at the claimed 336 GB/s, there would be an additional 192-bits used by the "slower" memory (it's not actually slower, it's just narrower - though it could be "slower" due to a more complicated memory interconnect, especially since most of the hardware units will probably use that controller). So it would actually be 320-bits + 192-bits. Note that if for some reason the slower memory was not GDDR6, it wouldn't be possible to achieve that sort of memory bandwidth for the 6GB memory in the footprint that they would need.
I would guess that the 192-bit bus are the two memory controllers closest to the CCXs, and the 320-bit controllers are the ones further away. I find it a bit strange given that configuration though, that Microsoft would specifically split them up, which would lead me to suspect that the 320-bit controllers are exclusive to the GPU CUs, or at least are indirectly accessed by the CPUs via some sort of memory bridge.
Also, the actual configuration of the GDDR6 is interesting, since the chips are dual channel x16. So that means that while each chip is technically 32-bits, it has 2x address/command buses, allowing for two independent 16-bit accesses simultaneously. In theory, this could allow for the 320-bit bus to access 20 independent 16-bit "words" (so a 20-channel memory controller), while the 192-bit bus could access 12 different 16-bit "words" (so a 12-channel memory controller). Keep in mind that GDDR6 supports bursts up to 16 lines too, so while the bus width is 16-bits, a single transaction can be 256-bits per channel (i.e. 32 Bytes). It will be interesting to see how the controllers are actually configured (since you could run both channels in lock-step effectively making a 2 channel x16 chip into a 1 channel x32 chip), and what the configuration means for memory access performance in the CCXs - especially when compared to 8-core PCs which are typically only 2 channels.
There isn't enough information on the PS5 to make any similar comments there, though I suspect the bandwidth will be significantly less due to the fact that memory transfers appear to go through the IO die.
The xbox only has 10 gddr6 memory chips. 6 are 2GB and 4 are 1 GB. This is like plugging in 3 8GB Dimms into a 2 channel memory controler on a CPU. The first 16 GB of memory used would run at full speed, the last 8 GB would run at half speed. In effect the memory bandwidth numbers are shared. It's not like you can add both the 560 and the 336 numbers to get the total bandwidth. The 560 GB/s is the maximum bandwidth the console has at any point.
@@draco10111b Interesting. It looks like Samsung does have 16 Gb chips, I didn't bother checking since both Micron and SK-Hynix only had 4 Gb and 8 Gb ones. And looking back at the DF video, there appear to be 6x K4ZAF325BM-HC14 and 4x K4Z80325BC-HC14, which would be what you said. It also looks like the Samsung ones are single-channel x32, which gives the 320-bit memory. But 320/8*14 = 560 GB/s. If I understand correctly, you are suggesting that the controller is in a fly-by-topology, where the closer chips have a 336 GB/s bandwidth. However, as far as I know, that's not possible with GDDR6. The Samsung datasheets aren't accessible from what I can tell, but the Micron and SK-Hynix ones specifically mention point-to-point or T-topology (in special cases) ONLY. And it's not using the T-topology method.
Do you have any thoughts as to specifically why a subset would run at a slower speed in that case? The only thing that I could think of, is it's the memory controller itself for the 6 GB, however, that would make the 560 GB/s impossible.
@@hjups It's using 1GB on all 10 chips to reach 560 GB/s. The extra 1 GB on those 6 2 GB chips comes out to 336 GB/s. So total bandwidth comes out somewhere between the 336 and 560. It's not the easiest thing to explain, but it's similar to the memory layout of the GTX 970, but in that case only one of the memory chips had an effective double capacity.
Not sure about the fly-by-topology stuff. This is more a feature of the memory controller.
@@draco10111b I think I understand what you are describing, but I don't really believe that it's the case - i.e. I would have to see the datasheet to be convinced. Although it does seem plausible, I don't understand why the GDDR6 would be implemented that way. Or are the 16 Gb chips actually dual channel? Micron calls their chips x16 even though they have 32 data lines (16-bits per channel - they do this by having multiple dies in the package). So if that were the case, and the 8Gb chips were single channel instead, or dual channel 16-bit, then I could see exactly what you mean. But if the 16 Gb Samsung chips only have 32 data bits on their pins, with a single CA bus, then I don't see how you could get different speeds for the different GB banks, nor could you connect the different GB banks to different memory controllers, since they would only have a single CA bus. Hopefully what I said makes sense.
The Xenon CPU in the Xbox 360 was a 3 core 6 thread part, so they have used SMT in console programming before.
The superiority of Microsoft when it comes to software is unquestionable. They are indeed a software company, and they have also stepped up on hardware.
This is exciting.
Sony not being sure about backwards compatibility is a joke right now.
@@evalangley3985 Majority of "Ps4 games" will be backwards comparable.
Microsoft has been buying studios left and right they could catch up to Sony in terms of gaming content this upcoming generation.
Software company that Sony beats in better games that does not make sense.
@@ramoncruz4369 "Software" is a lot more than games.
I'm specifically talking about the backwards compatibility capabilities of the consoles here.
"The superiority of Microsoft when it comes to software is unquestionable."
Apple: Am I a joke to you?!
@@billykotsos4642 The ability to play older games is a major selling point of PSNow. I wonder if Sony doesn't wanna go as far into backwards compatibillty because of that.
Series X has over a THOUSAND more shaders. Sony can crank the MHZ all they want,that is an unassailable lead.
Always enjoy starting my day with a new AdoredTV. I feel like it's good for the brain. Cheers Jim!
21:50 Khronos has standardised the rt extension (finally) and AMD is in on it. Expect out of the box vk_rt support on RDNA2.
And I bet PS5 gnmx is going to be heavily "influenced" by Vulkan in that sense, plus their own instructions.
Ha @AdoredTV, remember a while ago on twitter I speculated that APUs would be the first chips to migrate on a higher density node this time around?
Judgeing from that XBox die, I might be right.
Mug of tea in hand. watching an Adored video, Getting both information and, more importantly, insight on a Tuesday morning. You learn a lot and are also helped in coping with the lock down. Thank you Jim
The Sony SSD thing sound like what Intel has been promising /w Optane forever. Nothing's really come of it.
Holy shit Jim, I'm supposed to be working from home, now I know that I'm taking a 34 minute and 40 seconds break.
7:55 you made a mistake here, Zen TDP isn't calculated at base clock.
Yep, 3700x can do 75W consumption at full PBO boost.
@AdoredTV great video, only thing I am missing is that the PS5 clocks for the GPU and CPU seems to be the maximum BoostClocks. So I think in reality the PS5 will run the CPU at 3,2 GHz and the GPU at around 1,8-2 GHz... They just didn't want to use the "boost"-word because than everyone would know that these numbers will rarely be reached and that the PS5 in reality seems to be much slower than the new Xbox
And get cerny says the opposite. According to him he fully expects to console to run at those freqs. I know who I'd rather believe.
@@Sulleei he just says that it is possible to reach this speeds, but I think it will most likely only be possible, if the game for example doesn't need much CPU power, than the GPU can clock higher or the opposite... So I think for the most AAA games, since they need both in a good amount, both will not run at full speed.
Yes Jim! Saving our sanity during quarantine!
I think the majority of the time the superior ps5 ssd will be used for faster load times and im not convinced 4gb of 0.5 seconds worth of textures would be vastly better then the xsx 2gb per 0.5 seconds of textures. This brings up the question of texture quality, textures are already pretty good on current gen, I think its an odd thing to focus on for next gen, bigger graphical improvements will come from increased polycount, lighting and things like hair and cloth physics.
There goes my productivity this morning... Awesome!
I don't hear a lot of talk about die size and cost. If we're talking a 10-15% difference in launch price, that could be a real dealbreaker for buying the xbox -- especially since most people won't be seeing a huge increase in fidelity of one vs the other at 10+ feet away.
Sony shouldn't be worried by DXR, vulkan has an specification for raytracing done by Kronos in colaboration with AMD and NVDIA. Also Cerny make a mistake talking about Back compat. The PS5 has 3 modes, one for PS4, other for the pro and one with PS5 clocks. The last one is the mode they tested with the top 100 of games. The other two work for all games.
In the part of the chip potencial, they have coprocessors to offload some work with the io (in the case of the cpu) and the Audio part (in case of the GPU), thats why some developers talk about getting equal performance with the two consoles. The only part that could bite them in the long run is memory bandwith.
I think that in the inmidiate future Microsoft wins, they have the numbers, but for the long run it's difficult to Say. Microsoft has the todays technology and Sony It's More experimental...
What I like about these consoles is that the GPU's in them both are not equivalent to the usual $150-$250 PC discrete GPU, but are looking more like a $699-$799 GPU.
This could mean one of two possibilities, either the consoles are going to be very expensive, or more likely we are going to see 2080 super equivalent performance on the PC discrete GPU price drop to around $300 or below.
PS4 and Xbox one were based on a custom version of a $150 GPU (bonaire), PS4 pro and xbox onex were based on an equivalent $229 GPU (Polaris 10/20).
Another trend I remember is that the PC equivalent of a $499 console usually begins on console launch at roughly $800 starting build price and quickly drops (over the next 12 months) to about a $100-$150 premium over the console. So sometime in 2021 I expect that we will get a 6700/6800 series card for $200-$350, assuming the 5700 gets refreshed to around the $200-$250 mark as a 6700 and the RDNA 2.0 with RT comes in at the $300-$350 mark as the 6800.
I do believe something similar happened with the Xbox 360 (X1800XL), that its GPU performance was well beaten for under $300 within 6 months (X1900GT ($350 MSRP street low price $280) www.hardwaresecrets.com/connect3d-radeon-x1900-gt-review/10/ and much much less a bit over a year later (x1950GT ($150)) hothardware.com/reviews/sapphire-radeon-x1950-gt-256-mb?page=8.
One can only hope that those historical trends continue.
The new Xbox, at least, is almost certainly going to beat all sub-$1K graphics cards on the market. I have no real guesses on the console prices, but I'm sure they'll still be lower than any PC you can build with the same performance, by quite a margin.
It'd be awesome if that meant lower prices at the high end when big Navi comes out (can't count on nVidia alone lowering their prices), but I'm a bit skeptical. AMD seems to be enjoying the high-GPU-price market that nVidia has created a bit too much.
@@TrueThanny
The mainstream market isn't the high end and as the XSX particularly is a high end spec that you'd find on a $699-$799 GPU presently, any GPU for the PC of equivalent performance is going to have to drop to the console price within 6 months to remain viable and well below within 12 months or people will drop away from PC gaming at an increasing rate as the mid range shifts to consoles.
The last couple of console generations were built on mid range technology, the Xbox 360 and PS3 were built on higher end technology for the time and we saw massive drops on the high end cards of that same generation within 6 months with PC performance just blowing consoles away within two years. We should see similar this time with RDNA 3 looking to come online by end of 2021 early 2022, RDNA 2 discrete cards should have better performance CU for CU as I'd expect better cache arrangements that would be more limited on the console SOC due to die size considerations.
Big Navi is still going to be expensive, but the smaller Big Navi GPU's may just have the card we're looking for to match consoles and find 5700 cards (slower than consoles) repurposed for around $200 not long after the console launch.
Vulkan also has a ray tracing implementation as well. AMD has historically played very nice with Vulkan implementations.
Vulkan is basically a merger of OpenGL and AMD's own Mantle API, so they have a vested interest in supporting it.
If I had to choose one of them, I would choose the Xbox, because of the broader backcompat. Since several month I've tried to find and replay some of the OG Xbox games I had when I was a teen, and finding some hardware for that in working order is quite difficult. I just hope the system will get rid of region locking fot those older games.
AMD give us a 6C12T 5700xt APU .
If you guys do that I am happy to buy a new Mobo and 200$ on the APU
You can't get 5700xt apu performance using DDR4 like you do on desktop
The only way would be an APU with embedded HBM2
@@viniciuserrero
Yeah I know that , memory is always an issue with APUs.
But can't they sell RAMs with GDDR6 dies on it ??
@@minus3dbintheteens60 just look what they r doing with consoles , and they can make a new APU line-up with that .
@@rickbhattacharya2334 The console's OS is specifically tuned to work with/around the properties of GDDR6 memory, while PC OSes are designed to work with traditional DDR3/4. Since GDDR isn't used as system memory on a PC, no one is going to spend dev time to work with it.
@@NightKev i didn't knew about that
Thanks for your insights! Hope that you're feeling better and had a great time taking the weekend off. Stay safe and well my friend!
It was nice seeing som Mass effect footage on your channel again Jim. I liked them.
Interesting video keep i up!
Great video one of the best on the subject. Now that we know from Hot chips at 3:57 it was the coherency interconnect
"most game devs arent used to SMT"- xbox 360 had smt, still some truth to it I'm just saying.
the 360 had an IBM cpu, the SMT implementation and the ISA is completely different from Zen 2/ X86-64
Not really. I'd venture to say that Fecks CPUs had the _opposite_ of SMT, architecture-wise.
@@300maze i was under the impression it was a unique tri-core amd CPU modified compared to the pc counterpart
yeah that's the wrong angle he approached it. What most devs can't do is to properly use so many threads. most games don't use more than 4 or 6 threads still. for those, the extra 200mhz will obviously be a better choice. the ones that can use all the available 14 threads can opt for that instead.
@@300maze But devs have used multi GPU cores to companstate for how shite the Jag core are in this gens consoles(PS4 & X1), SMT is common practice now. Sony have developed tools they used for PS3(ported too PS4) that allows for several Jag cores too be split up having different stuff running together.
There is currently no ssd for pc that has 5.5gb i/o throughput!
1:26
Wichard*
Witcher Richard?
Great anaysis Jim. Great way to enlighten my day as we all sit in isolation. As a PC gamer its been a very long time since I've been excited about consoles. Better consoles will result in better quality PC games and competitive parts, thus reducing prices. It's a win-win as I see it.
Cheers bud.
A console delivering 2080S performance.
Nvidia wont be happy.
It’s hilarious in hindsight considering Jensen was declaring that a laptop 2080 Max-Q would be faster than the next-gen consoles a few months ago.
THIS WAS REALLY WELL MADE!!
Thanks again for another top notch tech video. Currently I have neither a PS4 or a Xbox. I play all of my games on a PC. I was thinking about getting a PS5 but after watching your video, I might take a closer look at the Xbox.
Why would you get an Xbox over a PlayStation when all of its games are on PC?
Adrian Maybe because exclusives don’t actually sell consoles?
Rather it’s the console itself.
Why buy a ps5 when 99% of the games are available on XSX and pc?
The exclusives argument never goes both ways.
ValkOptics I agree, Sony fans value the Sony exclusives far more than most. Not that they aren’t very good games but there is just as good if not better in most cases in third party and in reality Microsoft has some very good games also. The other point to buying a console and in this case the Xbox even if you own a pc is that your pc may not be graphically up to par with the Xbox and then there’s always the convenience of console gaming. Another nice thing with Microsoft’s ecosystem is in its game pass. Buy it on either and you can play it on either for no extra charge.
Rdna 2 is going to be huge
the 2.4GB/s of guaranteed throughput is impressive, but it's the software APIs and custom hardware built into the SoC that deliver what Microsoft believes to be a revolution - a new way of using storage to augment memory >(an area where no platform holder will be able to deliver a more traditional generational leap). The idea, in basic terms at least, is pretty straightforward - the game package that sits on storage essentially becomes extended memory, allowing 100GB of game assets stored on the SSD to be instantly accessible by the developer. It's a system that Microsoft calls the Velocity Architecture and the SSD itself is just one part of the system.
"Our second component is a high-speed hardware decompression block that can deliver over 6GB/s," reveals Andrew Goossen. "This is a dedicated silicon block that offloads decompression work from the CPU and is matched to the SSD so that decompression is never a bottleneck. The decompression hardware supports Zlib for >general data and a new compression [system] called BCPack that is tailored to the GPU textures that typically >comprise the vast majority of a game's package size." The final component in the triumvirate is an extension to >DirectX - DirectStorage - a necessary upgrade bearing in mind that existing file I/O protocols are knocking on for >30 years old, and in their current form would require two Zen CPU cores simply to cover the overhead, which DirectStorage reduces to just one tenth of single core.
"Plus it has other benefits," enthuses Andrew Goossen. "It's less latent and it saves a ton of CPU. With the best competitive solution, we found doing decompression software to match the SSD rate would have consumed three Zen 2 CPU cores. When you add in the IO CPU overhead, that's another two cores. So the resulting workload would have completely consumed five Zen 2 CPU cores when now it only takes a tenth of a CPU core. So in other >words, to equal the performance of a Series X at its full IO rate, you would need to build a PC with 13 Zen 2 >cores. That's seven cores dedicated for the game: one for Windows and shell and five for the IO and >decompression overhead."
Asset streaming is taken to the next level, but Microsoft wasn't finished there. Last-gen, we enjoyed a 16x increase in system memory, but this time it's a mere 2x - or just 50 per cent extra if we consider Xbox One X as >the baseline. In addition to drawing more heavily upon storage to make up the shortfall, Microsoft began a >process of optimising how memory is actually used, with some startling improvements.
"We observed that typically, only a small percentage of memory loaded by games was ever accessed," reveals Goossen. "This wastage comes principally from the textures. Textures are universally the biggest consumers of memory for games. However, only a fraction of the memory for each texture is typically accessed by the GPU during the scene. For example, the largest mip of a 4K texture is eight megabytes and often more, but typically only a small portion of that mip is visible in the scene and so only that small portion really needs to be read by the GPU." As textures have ballooned in size to match 4K displays, efficiency in memory utilisation has got progressively worse - something Microsoft was able to confirm by building in special monitoring hardware into Xbox One X's Scorpio Engine SoC. "From this, we found a game typically accessed at best only one-half to one->third of their allocated pages over long windows of time," says Goossen. "So if a game never had to load pages >that are ultimately never actually used, that means a 2-3x multiplier on the effective amount of physical memory, >and a 2-3x multiplier on our effective IO performance."
A technique called Sampler Feedback Streaming - SFS - was built to more closely marry the memory demands of the GPU, intelligently loading in the texture mip data that's actually required with the guarantee of a lower quality mip available if the higher quality version isn't readily available, stopping GPU stalls and frame-time spikes. Bespoke hardware within the GPU is available to smooth the transition between mips, on the off-chance that the higher quality texture arrives a frame or two later. Microsoft considers these aspects of the Velocity Architecture to be a genuine game-changer, adding a multiplier to how physical memory is utilised.
This is almost word by word what Mark Cerny also talked about. Except he said it would take 8 Zen Cores without dedicated hardware probably because of the higher performance SSD of PS5. Also, Sony is using a newer compression format called Kraken that is 10% more efficient than Zlib.
@21st Hmm that might explain why on the revealed specs Microsoft got from 2.4 GB/s RAW IO Throughput to 4.8 GB/s after compression while Sony got from 5.5 GB/s raw to "only" 8-9 GB/s.
However, if I remember correctly Mark Cerny said that the reason they went with Kraken was that it was already popular with developers so that might have been a good decision based on that alone.
So basically AMD just starting using compression techniques that Nvida has used for years and console applies them.
Short story: With faster SSDs and newer compression techniques less information needs to be stored in ram allowing for memory being used for other features. Freeing up more space then what the initial x2 increase in memory reveals.
@@88oscuro Consoles have always used compression Mark Cerny just mentioned they are moving to a more efficient method that is trending among developers. That is not the interesting part though. It is the ultra-fast SSD combined with custom processors that handle compression without even touching the CPU and also the 12 DMA channel SSD that will enable low latency high bandwidth access that Mark Cerny thinks is the key to next generation. This video explains what he was talking about well I suggest watching it: ruclips.net/video/PW-7Y7GbsiY/видео.html
@21st BCPack only handles textures and it should also be noted that if developers optimize the Kraken compression particularly well, that 8-9 GB/s figure can go as high as 22 GB/s.
It sounds like I'm going to have to wait for the PS5 pro. If my whole console gaming collection wasn't PS1/PS2/PS3/PS4, I'd be going Xbox this round. My gaming PC is now 10 years old and would have to be rebuilt from scratch, just a 2080 by itself is more expensive, the whole build would be much more costly than $500-$600 for a console. I didn't want to have to wait but oh well I don't have time to game much anymore anyway 😱😭🤣😂
you'd save that money because you don't have to pay for the privilege of online playing and also steam sales
Excellent video as usual, agree with all points.
Cant wait for an upcoming video on what the consoles will bring for PC.
- Future of games that are coded for more than 8 Threads and ultra-fast SSDs
- RDNA2 arq improvements: 50% better perf per watt in the same node??
- Xbox very early demo already at RTX2080 perf. What does this mean for the current GPU price to perf and upcoming gpu launches.
- RayTracing
Don't get me wrong, I'm super excited about these consoles and what they will bring to the table but a part of me does wish we waited a year longer and got 5nm, Zen 4, RDNA 3.0 based system with probably closer to 100 CUs if 7nm can give us 56CUs on just 360mm^2.
I know people will always say "if you wait for that then you will want to wait for 3nm and Zen 5" but I just think 5nm is looking to be such an amazing (and probably really long lived) node that it would be worth the wait.
Just wait for the PS5 Pro then... or the 'Xbox Series X X' (or whatever it will be called).
@@DJ_Dopamine Pffft... smalltime. I'll wait for PS6.
@@DJ_Dopamine That is unlikely to be next year and games will always be built for the lowest denominator, ie the base console specs.
They are doing well to have RDNA 2.0 in it but it would have been nice if everything was cutting edge like that, so a brand new 5nm node, new CPU, new GPU. These consoles will likely be around for at least 4 or 5 years, Zen 4 and 100CUs of RDNA 3 on 5nm in 5 years times will still be pretty sweet but Zen 2 and 36/52 CUs of RDNA 2 on 7nm in 4 years will be a little old.
I wonder if the SONY SSD is actually 1TB+ in capacity but the rest is reserved for longetivity reasons and thats why they only advertise 825GB because thats whats actually usable.
Yep I wondered about that too but his wording makes it seem unlikely.
All non-bargain-basement SSD's are overprovisioned. When you buy a "1TB" SSD, you're getting 10^12 bytes of capacity, but the chips themselves are 2^40 bytes. Even mechanical hard drives are overprovisioned a bit, which is why by the time you can detect a bad sector, the drive should be replaced, since it's already remapped all the spare capacity.
Just how much Sony or MS is overprovisioning is anyone's guess at this point, but the one that really needs it is MS, given their feature of suspending a game to SSD, which will drastically increase write cycles the more it's used.
The PS5 is going to have Half-Life: Alyx on the PSVR2, while Microsoft has neither game nor hardware for that. If that happens, and more games like that come out, Microsoft might miss this essential new step in gaming.
If Microsoft get their act together, they could have Minecraft and Hololense available, as they've already demonstrated it before, but they just need to release it and start X-Box's 1st-gen VR. Sony is already miles ahead, already selling over 4 Million VR headsets to Playstation owners, making them the only console owners with VR. Microsoft isn't even competing there.
Subscribed, i thought u did a better job explaining the specs of the consoles than digital foundry.
Did I make it before the sony fanboys got here to say it wasn't a reveal and it was a GDC presentation?
Tom Guy I’m not a pony but it was, in fact a GDC presentation. That’s common knowledge. So was MS’s reveal, they just had their marketing department gussy it up for them to marvelous effect.
it was literally the first thing they said in de presentation: this was meant for GDC. So i don't really understand what your point is.
@@evalangley3985 Oooh, now i understand 😁
i would have preferred that the consoles still had a 1TB HDD, with a dedicated 256GB SSD Internal that only the devs and system can access. that way the game loads into the SSD, and you still have a huge cheap amount of storage.
Also the Xboxs SSD is plenty fast enough, sony has pretty much no advantage there.
Making repeated huge writes to an SSD every time you load up a game is a surefire way to destroy its longevity.
@@ShawFujikawa that is true, but it should easily still last 7 years, the typical life of a console.
I'm actually hoping that most of the Xbox titles will come to PC also, so I won't have to buy the Xbox. Unfortunately I am a major jrpg player, and historically, those tend to be exclusive to PlayStation, so if I do buy a console it will probably have to be Sony's based on my primary game style. But I am glad to see Microsoft making a great console, because it will mean that the games will be that much better even if I get them for PC instead. Really hoping Microsoft will make all their games cross platform will PC since they own both os anyway, then I'll never have to buy an Xbox console to take advantage of the games, as long as I have a good enough PC with Windows on it.
Some people are also looking forward to next generation PSVR. What has Microsoft done on VR end?
They have stated that they are not currently interested (last summer IIRC).
HP and Valve are helping them with that.
Windows Mixed Reality, which is a flop, but excellent for VR on the cheap.
Finally somebody that knows what they are talking about! I do recall AMD saying something about density improvement in their latest chips. I’m still curious about how effective ray tracing will be on these consoles. I don’t foresee Sony’s use of it being as good as Microsoft’s. Sony’s SSD looks really good but costly. I don’t think the cost to benefit is worth it as it’ll likely push the console to similar costs as the Xbox. It’s good news for Sony first party developers though. My biggest worries about Sony are the overuse of the very small SSD and the noise of the system as Sony is well known for having consoles that run hot and loud.
07:55 TDP isn't calculated from the base clock. It applies to the boost clocks, whether maximum boost over one or two cores, or a lower boost across all cores. A 3700X will dissipate about 65W when running at between 4.0GHz and 4.1GHz on all cores. Input power to the socket at that TDP will be 88W. So this APU at 3.6GHz will be using substantially less power.
Adshir, an open source raytracing tech startup "Localray" had recently announced that they were in "one of the next gen consoles" - I can only assume that they would be in the playstation? The demo's they have are very impressive including being able to work with smartphones.
On the software side of things this gen - Microsoft has far more interesting tech. I've been waiting on HDR everything since it released on tv's, a huge win for Microsoft in this department.
Has to be, Microsoft always use their own software that is derived from one of their API like DirectX.
And since it's ray tracing, Microsoft would use Nvidia against Sony and form a Team Green, due to Nvidia having Ray-tracing capable GPUs and 'sales and history'.
Great video as always Jim.
I feel like since the debacle that was the 2013 xbox one launch, Microsoft has been trying a little too hard to always have the highest specs and top flops. Kinda feels like a high level dick measuring contest at this point. I mean yeah, i can totally nerd out over the details and numbers, but these two companies are losing their what fees like their ''personality''?
Consoles before the 7th generation all had vastly different approaches and couldn't even be compared apples to apples, which made each of them kinda unique. Nintendo kept their ''personality'' for example. Now it's all just moving close to PC architecture and it's all about who has more CU's and more RAM etc. etc.
Dunno thats just my 2 cents..
PlayStation's architecture is definitely further from a PC than Xbox's is. On top of that, the emphasis on 3D Audio for every kind of setup, DualShock improvements, and PSVR, give PlayStation a much more distinct identity than the Xbox.
Outside of Kinect, which is barely a thing any more, I can't even remember the last time Xbox talked about any new way to interact, only improvements to their controllers existing functions.
Being an underdog means you must bet hard, or else your platform will loose interest and slowly diminish. They got cocky after the 360, and to be fair there's always been conflicting goals within MS as to what Xbox should be. Regarding architecture, all other have perished. There's nothing else to choose.
Am I the only one who is more interested in the SSD difference then the GPU difference?
AMD's Raytracing "strategy" is both DXR and Vulkan RT. I'm going to assume the PS5 will use Vulkan
Probably their own proprietary API
They have their own API, But it's heavily based on Vulkan. Hopefully, they toss some code for improvements back into Vulkan so everyone wins.
Uh oh Sony fanboys gonna be butthurt. But, still, their SSD tech is impressive and I must say I love their approach to 3D Spatial audio using HRTFs and whatnot. After watching both "reveals" I felt from the beginning that the Series X was the better machine. The problem with the PS5 approach is of course, the lowest common denominator argument. No one but first party devs are going to be running games that require 5-9 GBs per second of bandwidth from the SSD. As cool as the tech is, most multi platform games will, at best, be designed for more mainstream SATA and NVMe SSD speeds.
I can't wait until ray tracing becomes mainstream, especially in virtual reality games. Half Life: Alyx has so many cube map reflections that break immersion.
It may be a while until ray tracing reaches VR, because VR requires high frame rates with no stutters. But yeah, I do hate those cube maps.
RT might not might be the future of gaming
@@GuyFromJupiter Might be sooner than you think with variable rate shading and eye tracking
I think Half Life Alyx's cube maps aren't that bad tbh. They certainly beat GTA V's awful cube maps that's for sure.
In flat screen games, cube maps aren't as distracting to me. I think it's because I'm able to move my head physically in the game space while watching the fake reflections..well..be fake. And then there's the lack of reflections for physics items. Byah! At least give me some screen space reflections here and there.
Thank you for your input.
It's 4am.... "Alright guys. How's it goin'."
Ok...
It's noon in europe. World isn't one timezone
@@hene193 yeah.... and?
All I got from this is that I might just upgrade my 5600 XT to an RDNA2 high end part. Now that I have a 240Hz display I really need the GPU power.
06:06 I thought the 5700XT has 40 CUs?
It does
Yep good catch, my mistake.
it's the 5700 that has 36
Honestly, big thanks to the people who make these technologies possible. It's like freaking magic how these little chips output such great graphics.
As we know, SSD's have finite write life, so I wonder how they are going to tackle this issue
deliberately leave large part of it empty. So it can last for atleast a decade.
@@yulusleonard985 Well we don't know if they will enforce that, also if this quick resume is used frequently, that is a lot of writing if switching between games or apps frequently if they store large parts of in in the SSD's.
@@prich0382 Probably for Sony, since their ssd size is weird. it coud be 1tb deliberately lockedup to 850 just so it can last for at least 5 years. Tech savvy gamers will use less than that. For Microsoft they already said they have 1tb. If retards know nothing about ssd and use up all of it it will last only 2 years. Dont know if their actual size is 1.25 TB then their ssd can last up to 10 years.
@@yulusleonard985 Actually the PS5 SSD is only 825 GB or whatever it is, they said it themselves, also covered in Digital Foundry video.
Or it could mean 825 gb useable space? Actual size could be 1tb?
Does Sony know exact xbox series x specs B4 they announce it?
Like knowing about it months Before announcement to decide what to focus on?
Yes
All PS4 games will work, but the top 100 are enhanced to PS5 standards.
Yep they clarified Cerny's comments later cheers.
they said overwhelming majority not all
@@thecooletompie Maybe Life of Black Tiger won't work... conveniently enough
Awesome. Thanks for the vid Jim. I have been eagerly awaiting your take on the next gen consoles and you absolutely delivered, as always. Thanks again👍And please, keep you and yours safe Jim.
Bang per flop. My favourite new unit of measure.
Digital Foundry somehow calculated the TFLOPS of the XSX GPU slightly wrong somehow, they said 12.155 when I calculated it to be 12.1472, no idea how they came to that
Same here. I calculated it in Excel and it comes to 12.147...
@@DJ_Dopamine Well you could of simply used a calculator doing 3328 x 2 x 1825 but okay aha