Since some CPUs also come with integrated graphics for how small they are, i don't see why not GPU with an integrated CPU also mini-ITX motherboards can fit too
@@tobiid1585 yep and then there is the thing about when we are going to plug a cord straight from our gpus to power outlets :D Also when are we transferring power to the mbo from gpu, hmm :D Future idea, make a computer that is only one component. Shouldn't be hard i guess.
Your laughing but this is it my dude, gpu with built in CPU as approximate values are enough now. Same goes for society and that’s what really worries me. You don’t have to know stuff any more just approximates and where to look up the real values
My mini-ITX would be happy. It has 2 full size M.2 slots. But only one of them has a cooler, and the other is on the bottom of the board with no head space for a heat sink or any airflow if you found a low profile one.
@@Nordlicht05GPUs will never match the full throughput of pcie technology. Ultimately that's a unnecessary and worthless endeavor unless pcie stops scaling
But wouldn't the heat created by the graphics card far exceed normal conditions that a hard drive would be sit at? Regardless of fans ... Water cooling I could concede being a more at reason for such close integration
@@hystaric_3013 Problem: the NVMe slot is going to sit next to the chip. Solution: Move it to the other side of the board. Can't be THAT hard.... right?
actually no cuz the gpu itself is not close to the ssd + the ssd is inserted through the plastic shroud so at worst it’ll get heated up by a few degrees by the hot air blown from the gpu heat sink
@@ichisenzy if your GPU and SSD share the same heat sink then your SSD would actually heat up quite significantly. Heat sinks are designed to be conductors of heat, and your thermal transfer elements (like thermal paste) aren't directional (you can't put thermal paste on upside down or round the wrong way). Therefore if your GPU heats up, the heat sink heats up, the SSD heats up.
@@redslatepretty sure SSDs have a lower maximum operating threshold. So in theory you could actually destroy your SSD because your GPU will thermal throttle at a temperature higher than is safe for your SSD
You still have. Except for the x16 and probably one m.2, they aren't directly connected to the cpu but handled by the chipset. It reduces the bandwidth (that SSDs especially use to it's maximum) thus using all direct PCIe lanes isn't too stupid. But nvidia not using the 16x and artificially throtteling their gpus is fucking stupid.
back in my day there was no pcie. you had isa. and maybe one pci slot. if you wanted to watch a multimedia clip you needed a gpu AND a mpeg decoder card. if you wanted a dvd player a few years later, you needed a dvd decoder card. I miss those days.
We understand how it works. It's just that like...the kinds of cards that aren't using 16 lanes, are all low end builds anyways...so it just kind of gimmicky. Usually hardware innovation comes to high end risk taking users first, and trickles down. This goes straight to the low end and the mediocre@@ohlala9546
Because future is unpredictable. Every new technology implementation requires try and error. Some stopped in r&d phase, some stopped because users and market didn't receive it quite well. It'll become standard once market received it well and less than 10% negative feedback from the users
In 10 years, 16 core - 32 thread, 128Gb of Ram, 400 TFLOPS GPU with 48Gb of VRAM just to run Unreal 7, or 8, at 1080p with no ray tracing at 47 FPS. LOL. I'm probably not to far off.
Love this, it's a great way to add value to your product and opens the door for ways to Differentiate Your product from the competition in a market Overly controlled by the Chip suppliers.
@@clintonweir7609 There is no time because the logical step will be using all 16 lanes for newer GPU generations. The idea is flawed untill they start using SSD for direct access for a GPU like consoles do.
This is the opposite. Its having less bandwith instead of more with less latency. The SSD is still connectet to the CPU but physically located on the GPU Board that cheaps out on PCIeLanes; limiting the GPU Bandwith to half.
@@thoreberlin that is for now though, whos to say that cant be developed? Its got to at least be something they are thinking about. What if you could have a drive that is optimised for direct storage that will get gpu access directly with cpu second. If that would be possible you could probably do some crazy stuff too if you need a lot of vram but not as fast for example. ssds are so fast already that it can probably be fast enough for a lot of uses if it was accessible by the gpu directly
@@thoreberlin Unless it's used to load directly from the NVME, it might be possible that the GPU could use it for direct storage. Like he said, this GPU doesn't use the full bandwidth anyway and the nvme wouldn't either. So it might be profitable for optimized games. From NVIDIA: GPUDirect® Storage creates a direct data path between local or remote storage, such as NVMe or NVMe over Fabrics (NVMe-oF), and GPU memory.
@@ryanpongracz8051 That would require a whole other type of solid state storage. If you just loaded, unloaded and flashed a standard M.2 disk like it was RAM, you'd have a fried M.2 disk very, very quickly
It does not require any m.2 storage at all , we're talking about vrams, they are totally different things, all they're saying is that the heat issues from m.2 drives can be cooled by your graphics card. Excuse me for bad English
“that if you confess with your mouth the Lord Jesus and believe in your heart that God has raised Him from the dead, you will be saved.” Romans 10:9 NKJV
@@astral_hazethat's on you for not doing research. The chip on each laptop card is also the same as one below it's equivalent on desktop. So a 4090 mobile would be 4080 on desktop, 4080 mobile would be 4070 desktop, etc.
@@arashkmahshidfar7780I don't think it's fair to blame the consumer for "not doing research" when it is *vidia who is doing the fraudulent advertisement, like you basically explained. A rational person would expect a "3080 (laptop)" to be almost exactly the same as a "3080 (desktop)" if they're both called "3080"...
@@irnehhenri I see the implications and I know that the naming scheme sucks, but also everyone knows laptops are slower than PCs. On top of that, every website and computer store I went to says the amount of VRAM what they are trying to sell you has. So in practice, if you say they scammed you, that's your fault for not knowing what is basically common info for the PC enthusiast.
@@-FAFO- Mainly the head you read of your GPU comes from the Vram and GPU-processor... And in the far behind on many cards there are just fins and fans, the actual card is pretty small. And in this video its almost as if there is added some extra space on the card to make it bigger, with no Vram anywhere near, which means the SSD is gonna do fine, usually they aren't cooled by fans at all, and here they are actually getting aircooled. My 4080 looks like a beast, but once you take the shell off and bring the real card out, you'll see its not that big of a card :)
@@FingalPersson Don't know if you've ever touched a 30 or 40-series card while running but it's hot af. Yes the insides get some air running through, but the SSD will only get some residual cooling from the casing the air is passing through. That's still not as ideal as having it on the motherboard where you can have fans directly blowing onto it. You're going to be running your SSD at constantly hotter temps than you would otherwise, but have at it.
@@-FAFO- I started building computers back in the days of 2002, when i was 17. I do know a thing or two about computers i think, and tbh i can touch my 4080 while gaming star citizen at 2160p any day on the rare end, mostly plastic anyways. Now i cant touch the fins while the card is running, but i touched the rare end of the backplate and it was not as hot as the gpu and vram that showed 65 and 73 degrees celius under heavy gaming. 65 being the gpu and 73 being the vram. Also got tons of sensors on my motherboard and in the case it self, so i dont need to touch stuff to know the temps, i can read em of software, so should you, stop touching inside your computer while running it, it can actually end up in a shortcut and break components in your system. Ill also let you know i own a Asus 3070 for my streamer rig and a ROG Strix RTX 4080oc so i guess that sorta answers your questions. In fact, i have had a suprim x 3090ti, 3060, 1070, 1060, 2080ti. i did also go for a 980 and i had a 480ti for quite a good while. A card i never forget was my voodoo 2 from 1999 when gta 2 launched. My first 3D card ever was a radeon saphire. I've build computers for 21 years, my own, and my friends, and their friends computers ive build and upgraded over the years, and i've assembled ( i dont like calling it building anymore, was not that easy to build computers before as it is today ) SO, when i say that, this can actually work, it is because i dont know you know the temps an nvme runs at, and i dunno if you have ever tried to run an nvme without block on top either from motherboard or one that comes with the nvme, but ill tell you this, ive done it.. ive run an nvme 1. gen without any block on top of it, no fan giving it air except for the air that is in the case. It never had a problem, in fact it works today in use in the very same rig it was build it, even tho that built is now longer mine as i gave it away to a friend. usually the readings of my ssd back then was 45-50 degrees, without a block... you wanna know the temps of my 2 nvme's that im currently using, which has speeds upto 7500/7200... well they have the blocks on that comes with the Rog Strix z590-E gaming mobo, and run about 35-45 degrees. Not much dfference. WD says that the maximum operating temperature for NVMe is 72°C. Now with that in mind that is pretty hot right? and even at that temp, you are still fine, but you don't want to push the temps any further. So with that said, imagine having it on my RTX 4080 which btw has output for fans on the card in the rare actually 2x4-pins are on the far end on that card, which means i could easily costumize additional fans to blow directly on the nvme, which would be in the back of the gpu, near fins, near the last gpu fan, near the intake fans and with additional fans if you choose tho even tho i think that would be overkill and unneeded. So yeah, i think it would actually not melt at all, and if you put a block on it that also help with temps, and in the very back of my card as i said, you could touch the backplate feeling it has temperature but NOTHING compared to if you touch the plate in the front which i would not recommend you doing, especially when the system is running. So now i have elaborated this alot, now please elaborate your thoughts that makes you so certain this will melt nvme's...
I thought this video was going in that direction, THAT would be cool. This will just make the SSD slower because the data has to pass through two interfaces instead of one. Plus the main M.2 slot on a mobo is usually very close to the CPU and there's a good reason for that. So this is adding an interface and moving the SSD further away. No thanks
@@EvilTim1911 well while the cpu is incharge of processing a majority of the file transfers, you could inturn use the verry many processors present in the gpu instead to process the files, this is the case for how workstation gpus work
DirectStorage support certainly will benefit from that. Better game loading and video encoding times in result. I believe that video encoding with saving on GPU SSD will run faster too.
@@myk1_sp simply not powerful enough, they can either fully use 8 or half use 16 so they just use 8 Note this applies to stuff like a 4060, a 4090 definitely uses all 16.
This is because we don't fulfill that capacity at this moment, your system is x64 bits, why you don't use 16 exabytes (17179869184 gigabytes) of ram instead of just 16 gigabytes? It's because we don't need such power by this moment, 32 bits used to accept 4GB of ram and we jumped to 64 bits, be patient, maybe in the future we can start needing those 8 extra.
@@erictsui8802 gpu w/o m.2 slot = x temp, gpu with m.2 slot = x temp + 'nothing was heated'. It's as if you don't understand simple logic or the word 'warmer' means.
@@Levinas-od5yx If you can understand Chinese, you could just check the test session of the original video. But you don't understand it, and you're still here writing a very simple equation intended to make your point, how sad is that?
@@Levinas-od5yx Just for pity's sake, I'll outline the test results for you. The SSD in this design is mounted opposite of what they normally are, with all the heat sources from the SSD coming into direct contact with the GPU heat sink, the temp of SSD could stay 42 C degree after 10 min Furmark GPU test.
Nah, what's really cool about this is the potential for the gpu to directly access the NVME storage and essentially use it as a T2 memory cache. That's only if the storage controller for the NVME is on the GPU rather than Mobo/CPU though.
no, what would really be cool is if games could be designed knowing that this is an option, and be able to install textures directly to a gpu-attached ssd, meaning that data could be loaded into vram much faster, with much less overhead, and using zero pcie bandwidth
@@daniillasso2916it was also ATI and not nvidia, which lets be honest, really is the dominating market so if they do something it has a better chance of making it further than if ATI tried it.
People complain about companies not being innovative enough. Then when the company show the consumer their innovations the consumers complains about the cost, about how it's not necessary, about it's too big/it's too small, about the color, etc. etc.
@@Demopans5990 still waiting for AMD SOC on par with consoles.. If they make a new socket with RAM/CPU/GPU all-in-one that'll be great. They are not catching Nvidia so they should start just doing something different instead.
@@VanSanProductions bro amd lterally did this. You can get a board with ram, vram etc and the Jaguar cores from the Xbox... And stuff it in an ATX case.
@@Demopans5990 Didn't Gamers Nexus make a News video about that? 1GB cache already exists on high end server/workstation CPUs, iirc. Edit: Yeah, search for it. It already exists.
Well, duh... you're still using measurements and terms like Terabytes from over sixty years ago. It's like when people from the 2020's still measured everything in bits... Wait, they didn't do that then either?!
That is definitely good, but I'm more excited for the Fiore System Airjet coolers to be available to the public. The Airjet can possibly be a great solution to over heating m.2
How would it cook from this? You do realise that the cooling solution of current GPU's are oversized? There is more then enough cooling reserve in those cooling designs that they can take care of several nvme ssds.
That came to my mind too. If the enormous amount of gigabytes of an SSD could be used somehow as a continuation of VRAM. Of course it's a lot slower than normal VRAM, but it would be at least close to the GPU unlike a normal SSD.
Heat transfer has some specifics as to why mounting SSDs onto GPUs is a bad idea. Sure, it can cool your SSD. But only if your GPU is not producing any heat or produces less heat than the SSD itself. And GPUs generate a lot of heat, in most setups they are the main part you are trying to cool down. And that heat goes into the heatsink and PCB. And from that PCB or heatsink it would transfer to the SSD. Upper limit for SSD temps is between 70-80 degrees Celsius. GPUs often operate at those temps during games and if the heatsink is at 60 degrees during gaming session, an SSD that would be otherwise at 45-50 is going to soak up the heat and get warmer. So while in theory it seems like a good concept, you'd need to thermally isolate the cooling, otherwise you'd just warm up the SSD for no reason.
@@SaraMorgan-ym6ue The GPU Chip is bigger that's why they are thicker, and the GPU generates more heat then the ssd, so it isn't adding heat. GPU's triple in size and complexity, so do cards, it's 1:1. In a closed system, adding heat will increase the temperature, but it will not rise infinitely. The temperature will eventually reach an equilibrium where the heat added is balanced by the heat lost to the surroundings. The temperature will not exceed the heat source because the system will reach a point where the heat input is balanced by the system's ability to dissipate heat. This is known as thermal equilibrium.
It'd be really useful if the gpu can access the ssd and copy data to vram, otherwise it still gonna have to allocate space in main memory and then be allocated in gpu, which doesn't change anything
thats not going to work SSD are far too slow, imagine running DDR2 with increased latency instead of GDDR6/X You going to have a stuttery mess, not only that the traces to the memory modules are the same length, now this SSD is completely off the timing. It can never work as a VRAM upgrade because VRAM is so much faster and integrated into the design of the board
This is exactly what Apple does with its Apple Silicone. Whatever ram you get also goes towards the GPU. So if you order 256 gb of RAM, youll also have 256 gb of video ram.
@@hajjdawoodthat's also what every single other laptop or desktop pc with integrated graphics does, but apple markets it like they came up with the idea
@@obj_obj There is several key differences, first its actually good integrated graphics (scoring better then the 3090 and worse then the 4090), second because its a SOC and the ram is also on the same chip the memory bandwidth is 400 GB/s verses a normal integrated cpu/gpu like then Intel i9 runs at only 89 GB/s I think its a bit unfair to act like Apple didn't make a very strong leap in CPU technology. Non-apple laptops are struggling to keep up with the nearly 4 year old M1 chips especially when it comes to performance per watt. Apple almost never invents anything new, theyve admitted as such many times over, but when they do something they typically do so really really well. iPads, Apple Watch, the M series chips, the iPod, even smartphones (iPhone) apple was definitely not the first but they did a really really great job that leap frogged them past competition for years and years.
@@hajjdawood This feature is extremely old and has been used by windows machines forever. However, if you actually think that you can buy 256GB of ram and allocate it all to vram, you're not just uneducated on the subject, you don't even know how Macs actually work, because what you're claiming isn't even possible on an M3 Max.
This could also be great for storing compiled shaders so they can be accessed faster than through the motherboard SSD. Perfect for when RAM isn't big enough and runtime compiled shaders require too much GPU power.
With trust replacing secure practices, stega usage getting really creative, & the price of meta analysis at an all time low, everything is now "interesting enough" to be monitored; it's just that very few things require overt intervention
This idea could be nice to let SSDs access to more memory to load game objects and/or shaders without the need to involve many CPU instructions at the process. Edit: in terms of global computation, drivers that communicate GPUs with CPUs follow interface rules that may not be as optimum as the GPU hardware itself loading information from own disk memory. For those who say this is a limitation, it really isn't because during the "long" process of initialization of the game/OS, the reality is that the GPU obtains one more read/write channel that is independent from the computer's main disk space, leading to faster effective load times just because the GPU is avoiding two things: 1. Driver control instructions overhead (may be different between applications) 2. OS main disk read/write operations that may be used for other system internal purposes, or not parallel instructions that require high performance CPU clock times. And this also introduces a new potential paradigm where GPU programming becomes essential to write programs that let the GPU store it's own internal files/resources into a totally different memory channel without generating overhead with main disk memory requests. So yeah, it is just better, whenever disk speeds are similar. And people complaining or throwing arguments each against other should shut up and think they're not spreading really accurate information, but can offer a different view where disadvantages instead of spreading a useless fight. Thanks.
@@The_Keeper GPU memory is GDDR6... An SSD would never come close to being as fast. I can't believe you guys thought you could just strap a 1tb SSD on a GPU and call it a day 😂 Edit: GDDR6 NOT GDDR6X
@@OldRedOwlthey already do this all 4xxx series cards use practically the same PCB. They just install less for each model. Con is less performance and lesser image quality. Cons are cheaper price.
You know, BMW thought about that. They try to sell subscription for heating seats. 😅😅😅 We can definitely see something like that on pc soon. Hopefully, people won't fall for it. Otherwise we are F***d😢
These graphics cards require you to split the x16 lanes into x8 for the GPU and x4 for the NVMe drive using what’s called “PCIe bifurcation”. Even attempting to use such an SSD on a GPU like this requires your motherboard to support PCIe bifurcation. I’ve ran server disks using PCIe bifurcation before. It is a nuisance to set up, a nuisance to manage, and the payoff is negligible (assuming you have space M.2 slots on a standard motherboard, use those instead). So no, no I am not excited for this GPU or this approach at all lol.
I don't think you inderstand what's going on here. bifurcation is a pain because it turns 1 lane into 2 lanes. this GPU only uses 8 lanes and so 8 lanes are still available for something else. 2 NVMe drives, for example, which only use 4 lanes each.
1. Most of the AMD motherboards support bifurcation by default, but motherboard manufacturers do not include BIOS interface to configure it. 2. This does not require lane splitting.
Great timing for this video. I was recently wondering if video card slots needed an update in order to improve GPU tech. I didn't realize that they had too many slots, but with how heavy graphic cards are they need all the mobo support they can get. Note: I got a video card stand recently.
Actually, they used to in the past. I still remember an old card from my early days that had 4 slots to plug in memory modules on the gpu for expandability. The only reason we don't see it anymore is because the timing of the memory modules running through the circuit to the gpu die itself would be too slow for modern components. Faster than you can think, but still not fast enough to keep up. The memory modules would have no way to compete with memory directly on the gpu die or on the card immediately next to it. Sad because it was such a novel idea and it was brilliant at the time. 500MB GPu, quadrupled to 2GB. That was a ton back then. Having 1GB even was like, baller.
it'll be on the coolest part of the card, add a layer of thermal transfer tape to it, touching the rad that can effectively cool the 2nd hottest component with no effort? ;) i would genuinely love to see thermals with/without an ssd on the card. I think it's a really good concept especially since gpus typically vent their hot air directly out the case . well alot of the air
Honestly if it works with the EGPU market this is a big W for that. Not only can you have an external gpu to run your more demanding games but you can have storage as well Without too much hassle. W
Also more storage options in mini-ITX builds. Cooling would be a pain, but a big enough heat sink to handle both the GPU and NVMe isn't a far dream. A singular 4090 chip probably runs hotter than both combined.
Kind of how back in the day we would install games on a USB stick to play on school PCs as we couldn't install any software on them, just now you bring your own GPU with it as well so you can play games on any pc with a decent processor.
I'm more interested about the DirectStorage implications that this may have, since the SSD sits closer to the GPU and could significantly improve framerates during asset loads.
Interesting, but if your SSDs sit closer to your GPU then just run GPU Direct Storage instead of Direct Storage. Both GPU and CPU have specific condition in which their workload can be handle efficiently.
@@Lagger625on this concept build yes, but the point is theres no reason why a manufacturer couldn't build in an SSD like this and run it as GPUDirect storage, passing between the gpu and the ssd without needing to run through the cpu for everything.
Not that unusual in the past. A lot of Amiga boards often had multiple items on them such as RAM on a SCSI HD Controller. I think it had something to do with how Autoconfig worked with Amiga Zorro and CPU boards. So this is a retro idea that is still a good idea.
In this case it's more akin to a pass-through port. The GPU doesn't talk to the unused PCI lanes, it just routes them to somewhere potentially useful, like an M.2 socket.
This is the symtom of moving from SATA based SSDs to PCI-e SSDs. While they are fast, they ultimately take up lanes. Its one of the few reasons to run SATA. But if you want real speed, you need SAS. Its niche to the enterprise market, but think a simple upgrade to SAS-2 or SAS-3 even would be a game changer for those that want to run SSD RAID setups. I personally run SAS and I love it! There's nothing better than SAS for both external and internal capability. Its a truly flexible system. It's a truly the wonderful successor to parallel SCSI.
Imagine if GPU manufactures took that concept, and used an SSD on your GPU as a way to handle shader caching in games, freeing up your PC HDD from bulky shader files that build up, with some games.
"Freeing up your blah blah blah to do more..." You know full fucking well they ain't freeing up shit. Performance will get even worse with more hardware resources available as lazier and lazier unskilled devs are able to use modern specs to skate by with even worse optimized games. The 360 lasted over 10 years on market running the same games under the hood as modern pc's and consoles with only graphical hits taken, all on MODEST specs for when it came out back in 05. Now with 8gb and now 16gb of ram standard, internet browsers take up 10gb for like 5 tabs and youtube, and standard run of the mill fps games can max out ram usage qhen they used to get by on a gig or two max for the same amount or characters, physics and underlying game logic being performed. Even new graphics technology isn't being utilized by the majority of games, just the additional vram for people who don't understand how to optimize their texture cache usage and just demand you pay more for a car with more GB, or demand that the vram be gddr6 instead of gddr 3 or 4 like it used to only a few years ago. The new technologies ARE NOT ground breaking or genuine improvements, just a slightly larger bandwidth which is nowhere even close to being needed for consumer software applications. These nvme's on gpu's will only be another burden placed on the consumer's shoulders by shitty third world devs that cannot actually develope software at a professional level
@@kayukshi4271nah ram and dram are quite different. Pretty sure like the 1030 or something had a varient with ddr4 instead of gddr5. And the difference it made was a lot
Well there is also a chance that the SSD will corrupt due to the heat of the gpu, which we all know, it heats up a lot especially on gaming. If the SSD can withstand the heat of the gpu, then I'll support this concept.
If it is designed with that in mind by competent engineers, it would not be an issue. The real problem is that you’d need an additional, expensive chip to do the PCIe bifurcation.
That device was a poorly conceived product. It was a x16 GPU meant to utilize that slot. The two SSDs added 2 each X4 devices to that x16 slot. All of these lanes connect to the CPU meaning that the slot is 150% oversubscribed. You would only see contention between the lanes when you utilize them. In the end the device would flounder and drop frames when you put it to task. As for the GPU speaking directly to the GPU, this could conceivably work but only the GPU would be able to utilize the SSDs . It’s better to keep devices on discrete slots/connectors for reliability and serviceability.
@@3DMarkRGBA true but some mods and even new games going crazy with ram requirements this might be helpful although the heat being pumped out by some of these cards might just cook SSD's
Speaking out of your ass about it floundering and dropping frames, first of all it was a professional card not for gaming, the use case was very specific in video production and CAD.
GPUs are becoming brotherboards
Underrated comment 😭😭
My thoughts exactly. How only 4 likes!?@@tanvir.ahamed0
So the brotherboard plugs into the motherboard? That's a fucked up computer
@@perochialjoenah frl🤣
Fatherboards, since they still need mother to be effective
"my gpu ran out of space"
Edit 7 months later: i got this video reccomended again and got shocked.
💀💀💀
4090: hm? What you say?
@@TotallyNoAimlmao
@@TotallyNoAimthis argument fits all the new nvidia gpus😂
cries in stable diffusion
"GPU: look at me, I'm the motherboard now" meme is starting to get too realistic
Since some CPUs also come with integrated graphics for how small they are, i don't see why not GPU with an integrated CPU also mini-ITX motherboards can fit too
Oh s#it... Oh no. :D Here we go again! :D
@@mmrox4320Oh god what have you done
the funny thing is, gpu is basically can be called a computer itself with all the component inside that, except for the lack of psu.
@@tobiid1585 yep and then there is the thing about when we are going to plug a cord straight from our gpus to power outlets :D
Also when are we transferring power to the mbo from gpu, hmm :D
Future idea, make a computer that is only one component. Shouldn't be hard i guess.
“How much vram do you have?”
“Uhhh 2 tb”
it's not for vram sadly
@@valentds 😑
In the future, we may expand VRAM like that. Until then, you ask an electronics guy to upgrade your GPU (delicate and expensive task).
Being able to add VRAM would be clutch.. but let's be honest, they'd never do that.
That’s what they should’ve done. But VRAM is way too fast at ddr6, the drive wouldn’t be able to keep up.
Previously: built-in graphics on the motherboard
The future: built-in motherboard on the graphics card
Your laughing but this is it my dude, gpu with built in CPU as approximate values are enough now. Same goes for society and that’s what really worries me. You don’t have to know stuff any more just approximates and where to look up the real values
Technically, that's an APU.
How about a full on prebuilt pc inside the gpu 😂
@@FocusedCatlike that one that runs a fighter jet?
@@ichdissdich98 the fuck are you going on about lol
- 4060: "What is my purpose?"
- "You provide cooling for SSDs"
- 4060: "Oh my god"
Lmao
Rick and Morty
😂😂😂 nice reference
😂
That'll be me when I have a 4060 and all I play is Factorio games smaller than 20GB with pixel graphics.
"What computer do you have?"
"An RTX 4090."
That's already how it is, pretty much. Lead with the most expensive part
This reminds me of diesel truck owners. I don’t drive a 3500, I drive a Cummins!
@@karlhendriksefr
That one has no free pci express lanes though, so it's more likely a 4060 :P
Ohh I got the rtx 6090
My mini-ITX would be happy. It has 2 full size M.2 slots. But only one of them has a cooler, and the other is on the bottom of the board with no head space for a heat sink or any airflow if you found a low profile one.
Your GPU is becoming the System within a System
Always was.
Meanwhile, SSD has its own little OS and 4 core ARM CPU that manages the data.
Cells within Cells
Not really. The traces are practically going straight from the ssd to the motherboard.
All pc components are systems within a system
GPU: look at me, look at me. I am the computer now
From discrete GPU to discrete Computer 😂
Hahahahahaha, YES 👍🏻 exactly that
Kek
Somebody ported Linux to OpenCL, so yeah basically
DPU: look at me, look at me. I am the computer now
Bro got that RTX 4060 512GB version 😂
Lmao
Yeah, things like that will break minds!
RTX 4060 2048GB edition
PC: master I never felt this much power
Bro I got the 4060 ti 10tb upgrade
Would super love this idea especially for small form factors where you're limited in space and mobo features
2077: and now you can also put some ram on your gpu
Finally can actually download more ram
more like "now you can also put some ram in your head"
Am I getting wooshed here, or are you not aware that GPUs already have their own dedicated RAM?
@@KaityKat117 i know, i mean we can upgrade it, its a joke anyway
@@zyuradesu89 oh like give the GPU DIMM slots.
actually....... that doesn't sound like too bad an idea
How can I cool my SSD?
Nvidia: How about you strap it to the second hottest component in your computer?
I wonder how that should be the future? That means there will never be GPUs wich use all lanes on any Mainboard ever!?
@@Nordlicht05if its not necessary, they won't do it. gold and copper are expensive the less the better for them
Hottest component requires the coolest fans
@@Nordlicht05GPUs will never match the full throughput of pcie technology. Ultimately that's a unnecessary and worthless endeavor unless pcie stops scaling
@@infernaldaedra than they could make less lanes on Mainboards ☝️
4060: "What is my purpose?
Me: "You are a house for my SSD."
4060: "You've gotta be fckin kiddin me!"
Honestly being a place for an ssd is worth more than the worthless garbage 4060
4060 should apologize for existing... That thing is waste of my semen.
😂
@@junkpowyou fucked your GPU? 😂
But wouldn't the heat created by the graphics card far exceed normal conditions that a hard drive would be sit at? Regardless of fans ... Water cooling I could concede being a more at reason for such close integration
Now that’s something I am looking forward too. Great concept. There may be few degrees hotter temps overall for gpu but the gains are worth it
this brings the „I am the motherboard now“ joke to a whole different level
@@ghst-vq8sv😂😂😂
@@ghst-vq8svasus 4090 matrix has entered the chat lol
This will no longer be a joke in 10 years
I still use a intel celeron now and run kali linux on it
In the future, your graphics card will just be your computer.
C*nsole be like
@@lucythefrogge😨
like smartphones?
This is what I’ve been saying from long time
No genius, gpus will be obsolete and cpus will do both in the future.
this is a perfect opportunity for nvidia to make a 4060 that has an SLI bridge and like have a super gpu
or better yet, a compact 4060
They won't its NVIDIA were taking about
A double 4060 would not make a super gpu
Sli barely did any good. It was a waste of time and money.
problem with SLI is driver support and getting game devs to not only for new games but old games as well.
Manufactures should be selling the cpu, gpu, ram, and ssd already installed and discounted. Just like a phone with different tare levels and size.
I believe that's called a pre-built pc..
I was gonna say that@@tekirandlime
Have u not heard of pre-built pcs? 🤦🏻♂️
you would love laptops
Or consoles
I almost thought those cards would need 2TB VRAM
Ah yes, I can now load the textures of the entire GTA V map.
@@Roach_Dogg_JR So that should mean no load time for the game, everything is just there and ready to go.
Dude same
@@AnzaiDior Even unarchived GTA V files wouldn't be taking 2 TB of space
@@smsff7 you can load your entire games library onto the SSD just waiting to be launched
"So what we can do to make 60°C SSDs cooler?"
"Stick it to the board that can get up to 100°C, duh..."
Exactly, it's not by the actual chip. It's on the fucking backplate lmao That shit gets hot...
Cool concept, gonna take an arm and a leg to engineer.
@@Not_interestEd- What can i get for a rib? Guess a grilled SSD.
Attach the heatsink to the board of the SSD itself. It will be even more efficient.
@@hystaric_3013
Problem: the NVMe slot is going to sit next to the chip.
Solution: Move it to the other side of the board.
Can't be THAT hard.... right?
Imagine your gpu overheating and cooking your 2TB ssd
actually no cuz the gpu itself is not close to the ssd + the ssd is inserted through the plastic shroud so at worst it’ll get heated up by a few degrees by the hot air blown from the gpu heat sink
If it got that hot, your GPU would be toasted first.
Imagine your 2TB ssd overheating and cooking your gpu 🙃
@@ichisenzy if your GPU and SSD share the same heat sink then your SSD would actually heat up quite significantly. Heat sinks are designed to be conductors of heat, and your thermal transfer elements (like thermal paste) aren't directional (you can't put thermal paste on upside down or round the wrong way). Therefore if your GPU heats up, the heat sink heats up, the SSD heats up.
@@redslatepretty sure SSDs have a lower maximum operating threshold. So in theory you could actually destroy your SSD because your GPU will thermal throttle at a temperature higher than is safe for your SSD
Imagine releasing a custom firmware that uses that ssd slot as an extra vram slot. I might request that.
Nice multi tb of vram
@@leonhardmollney8638 😂😂
There is a reason why we don't use SSDs for RAM or VRAM...
"Yo dawg I heard you like SSDs"
An older meme, but it checks out
THAT'S MEAN MORE VRAM ??????
No@@Bin0mar
😂😂😂
“So we putting an SSD, ON YOU MOTHA FKIN GPU!”
Soon, your SSD needs a SSD
😂
SSD²
@@khyenssd gonna be using micro SD
Just as I need her 😢
Makarena
Back in my day, we just used to have more than one PCIe slot.
You still have. Except for the x16 and probably one m.2, they aren't directly connected to the cpu but handled by the chipset. It reduces the bandwidth (that SSDs especially use to it's maximum) thus using all direct PCIe lanes isn't too stupid.
But nvidia not using the 16x and artificially throtteling their gpus is fucking stupid.
That shares lanes.
back in my day there was no pcie. you had isa. and maybe one pci slot. if you wanted to watch a multimedia clip you needed a gpu AND a mpeg decoder card. if you wanted a dvd player a few years later, you needed a dvd decoder card. I miss those days.
We understand how it works. It's just that like...the kinds of cards that aren't using 16 lanes, are all low end builds anyways...so it just kind of gimmicky. Usually hardware innovation comes to high end risk taking users first, and trickles down. This goes straight to the low end and the mediocre@@ohlala9546
@@krzysztofb9279 Not a card just a decoder installed like the k-lite codec pack.
this is actually so good for matx builds
"I am all about this if this is the future."
- Every Tech Company
narrator: it was not the future
Hahahaha, 😂
"I am rooting for whichever team is winning"
Because future is unpredictable. Every new technology implementation requires try and error. Some stopped in r&d phase, some stopped because users and market didn't receive it quite well. It'll become standard once market received it well and less than 10% negative feedback from the users
@@JhanOjan This isn't some new technology, just strapping an SSD to the hottest PC component.
Can’t wait to see what computers look like in 10 years. Gonna be one RGB block
Mac studio lol
"What's you're specs?"
34893020696969 rgb strips
In 10 years, 16 core - 32 thread, 128Gb of Ram, 400 TFLOPS GPU with 48Gb of VRAM just to run Unreal 7, or 8, at 1080p with no ray tracing at 47 FPS. LOL. I'm probably not to far off.
@@northwastakenbasically lmao
Rgb ssd
I wish vram could be expanded like that
for real...
Radeon Pro SSG. Sort of. Kind of. Eh.
@@bck187lucky you
how much vram do you need?
This is exactly what crossed my mind when I clicked on this video.
Love this, it's a great way to add value to your product and opens the door for ways to Differentiate Your product from the competition in a market Overly controlled by the Chip suppliers.
If only they used this tech to store prerenders and shader caches to take load off the GPU.
I expected it would be used for something like this. Super disappointed it wasn't, especially with how this guy presented it.
Give it time.
DirectStorage would be absolutely goated with this because the gpu can just utilise the ssd straight from the card instead of through the motherboard
@@clintonweir7609 There is no time because the logical step will be using all 16 lanes for newer GPU generations.
The idea is flawed untill they start using SSD for direct access for a GPU like consoles do.
@@flsendzz I would love to be able to store Stable Diffusion directly on my GPU so its already there for use
Taking "direct storage" to a different level 😭
This is the opposite. Its having less bandwith instead of more with less latency. The SSD is still connectet to the CPU but physically located on the GPU Board that cheaps out on PCIeLanes; limiting the GPU Bandwith to half.
@@thoreberlin it was a joke mate
@nandishmunjal1998 Well, there were professional graphics cards that had direct storage from AMD: Radeon Pro SSG. Not a joke ;)
@@thoreberlin that is for now though, whos to say that cant be developed? Its got to at least be something they are thinking about. What if you could have a drive that is optimised for direct storage that will get gpu access directly with cpu second. If that would be possible you could probably do some crazy stuff too if you need a lot of vram but not as fast for example. ssds are so fast already that it can probably be fast enough for a lot of uses if it was accessible by the gpu directly
@@thoreberlin Unless it's used to load directly from the NVME, it might be possible that the GPU could use it for direct storage. Like he said, this GPU doesn't use the full bandwidth anyway and the nvme wouldn't either. So it might be profitable for optimized games. From NVIDIA:
GPUDirect® Storage creates a direct data path between local or remote storage, such as NVMe or NVMe over Fabrics (NVMe-oF), and GPU memory.
Can't wait to upgrade my 12GB VRAM to 1TB.
It’s not for vram
@@ShiggyCompPt2it should be
@@ryanpongracz8051 its for storage, not vram. The gpu uses 8 lanes, They put a ssd on the gpu to make the most out of the x16 slot.
@@ryanpongracz8051 That would require a whole other type of solid state storage. If you just loaded, unloaded and flashed a standard M.2 disk like it was RAM, you'd have a fried M.2 disk very, very quickly
It does not require any m.2 storage at all , we're talking about vrams, they are totally different things, all they're saying is that the heat issues from m.2 drives can be cooled by your graphics card.
Excuse me for bad English
Dude that’s actually awesome. I can’t wait to hear more abt these future projects with pcs. Thanks Zach!
Imagine putting the SSD on the Hottest place in the whole PC
Genius
And the gpu in the video has the ssd mounted on the back, so basically the part of that gpu that has the absolute least cooling on the entire gpu
CPU is still hotter but yeah.
@@jankrnac3535 not necessarily, my cpu runs at 69 to 70 degrees under full load, my gpu goes up to 85 under full load
@@legominimovieproductionsNot always.
SSD: “AHHHHH GOD IT BURNSSSS AHHHHHH PLEASE END MEEEEEE GOD IT BURRRRR-“
"MICHAEEEL
DONT LEAVE ME HEEERE
MIICHAAEEL"
ruclips.net/user/shortsr7UG5bbjM60?si=hpy9siEeMNvbPC5M
@@fennex7575hey Vsauce Micheal here
@@fennex7575i don’t even want to tell you how bad i want to punch you for being a cringe furry
“that if you confess with your mouth the Lord Jesus and believe in your heart that God has raised Him from the dead, you will be saved.”
Romans 10:9 NKJV
At first I thought this was implying that we are now using NVMe drives as VRAM lmao
needed since nvidia scammed me out of 4 gb of vram (they make worse laptop gpus under the same name)
@@astral_hazethat's on you for not doing research. The chip on each laptop card is also the same as one below it's equivalent on desktop. So a 4090 mobile would be 4080 on desktop, 4080 mobile would be 4070 desktop, etc.
@@arashkmahshidfar7780I don't think it's fair to blame the consumer for "not doing research" when it is *vidia who is doing the fraudulent advertisement, like you basically explained. A rational person would expect a "3080 (laptop)" to be almost exactly the same as a "3080 (desktop)" if they're both called "3080"...
@@irnehhenri I see the implications and I know that the naming scheme sucks, but also everyone knows laptops are slower than PCs. On top of that, every website and computer store I went to says the amount of VRAM what they are trying to sell you has. So in practice, if you say they scammed you, that's your fault for not knowing what is basically common info for the PC enthusiast.
It got me too. Who woulda thought it was something even stupider 😂
Let's just rember that this is all 20 year old military technology.
This is probably the most useful feature for RTX 4060 😆😆
Only*
dumb idea for dumb gpu
@@0AThijs I think that might be the joke.
This is how you can cook an NVME in no time!
Exactly what i was thinking, that thing is gonna get toasted.
LoL! Uh... LOL im just gonna state that again without saying any further and see if you can understand how dumb you just sound by saying so LOL!!! :-D
@@-FAFO- Mainly the head you read of your GPU comes from the Vram and GPU-processor... And in the far behind on many cards there are just fins and fans, the actual card is pretty small. And in this video its almost as if there is added some extra space on the card to make it bigger, with no Vram anywhere near, which means the SSD is gonna do fine, usually they aren't cooled by fans at all, and here they are actually getting aircooled. My 4080 looks like a beast, but once you take the shell off and bring the real card out, you'll see its not that big of a card :)
@@FingalPersson Don't know if you've ever touched a 30 or 40-series card while running but it's hot af. Yes the insides get some air running through, but the SSD will only get some residual cooling from the casing the air is passing through. That's still not as ideal as having it on the motherboard where you can have fans directly blowing onto it. You're going to be running your SSD at constantly hotter temps than you would otherwise, but have at it.
@@-FAFO- I started building computers back in the days of 2002, when i was 17. I do know a thing or two about computers i think, and tbh i can touch my 4080 while gaming star citizen at 2160p any day on the rare end, mostly plastic anyways. Now i cant touch the fins while the card is running, but i touched the rare end of the backplate and it was not as hot as the gpu and vram that showed 65 and 73 degrees celius under heavy gaming. 65 being the gpu and 73 being the vram.
Also got tons of sensors on my motherboard and in the case it self, so i dont need to touch stuff to know the temps, i can read em of software, so should you, stop touching inside your computer while running it, it can actually end up in a shortcut and break components in your system.
Ill also let you know i own a Asus 3070 for my streamer rig and a ROG Strix RTX 4080oc so i guess that sorta answers your questions. In fact, i have had a suprim x 3090ti, 3060, 1070, 1060, 2080ti. i did also go for a 980 and i had a 480ti for quite a good while. A card i never forget was my voodoo 2 from 1999 when gta 2 launched.
My first 3D card ever was a radeon saphire.
I've build computers for 21 years, my own, and my friends, and their friends computers ive build and upgraded over the years, and i've assembled ( i dont like calling it building anymore, was not that easy to build computers before as it is today )
SO, when i say that, this can actually work, it is because i dont know you know the temps an nvme runs at, and i dunno if you have ever tried to run an nvme without block on top either from motherboard or one that comes with the nvme, but ill tell you this, ive done it.. ive run an nvme 1. gen without any block on top of it, no fan giving it air except for the air that is in the case.
It never had a problem, in fact it works today in use in the very same rig it was build it, even tho that built is now longer mine as i gave it away to a friend.
usually the readings of my ssd back then was 45-50 degrees, without a block... you wanna know the temps of my 2 nvme's that im currently using, which has speeds upto 7500/7200... well they have the blocks on that comes with the Rog Strix z590-E gaming mobo, and run about 35-45 degrees. Not much dfference.
WD says that the maximum operating temperature for NVMe is 72°C.
Now with that in mind that is pretty hot right? and even at that temp, you are still fine, but you don't want to push the temps any further.
So with that said, imagine having it on my RTX 4080 which btw has output for fans on the card in the rare actually 2x4-pins are on the far end on that card, which means i could easily costumize additional fans to blow directly on the nvme, which would be in the back of the gpu, near fins, near the last gpu fan, near the intake fans and with additional fans if you choose tho even tho i think that would be overkill and unneeded.
So yeah, i think it would actually not melt at all, and if you put a block on it that also help with temps, and in the very back of my card as i said, you could touch the backplate feeling it has temperature but NOTHING compared to if you touch the plate in the front which i would not recommend you doing, especially when the system is running.
So now i have elaborated this alot, now please elaborate your thoughts that makes you so certain this will melt nvme's...
What we really need then is the GPU being able to directly decode textures and things from the builtin SSD
I thought this video was going in that direction, THAT would be cool. This will just make the SSD slower because the data has to pass through two interfaces instead of one. Plus the main M.2 slot on a mobo is usually very close to the CPU and there's a good reason for that. So this is adding an interface and moving the SSD further away. No thanks
@@EvilTim1911 there is direct storage access or whatever it's called for GPU directly reading from the hard drive. So it might be useful
@@EvilTim1911 well while the cpu is incharge of processing a majority of the file transfers, you could inturn use the verry many processors present in the gpu instead to process the files, this is the case for how workstation gpus work
DirectStorage support certainly will benefit from that. Better game loading and video encoding times in result. I believe that video encoding with saving on GPU SSD will run faster too.
You say the SSD is gonna get cooler from the GPU but my GPU is just gonna overheat my SSD lmao
I’ll stick to my 1080ti that runs at 25C
@@jax3rdude is your ambient temp in the negatives or something?
@@JFat5158bro lives in Antarctica
-Wee need direct storage on PC!
-But we have direct storage at home
Direct storage at home:
What do you mean? PCs already have DirectStorage. And the drive doesn't need to be near the GPU, it makes no difference.
I’d be more excited if the gpus just used the full 16 lanes by default
Same.
I'm not sure why some GPUs don't utilize the 16 PCIe lanes that they get installed to.
@@myk1_sp it's cheaper to not do that. That's... The entire reason.
What? You're disappointed? Sucks to be you, just buy a new card! Casual.
@@myk1_sp simply not powerful enough, they can either fully use 8 or half use 16 so they just use 8
Note this applies to stuff like a 4060, a 4090 definitely uses all 16.
This is because we don't fulfill that capacity at this moment, your system is x64 bits, why you don't use 16 exabytes (17179869184 gigabytes) of ram instead of just 16 gigabytes? It's because we don't need such power by this moment, 32 bits used to accept 4GB of ram and we jumped to 64 bits, be patient, maybe in the future we can start needing those 8 extra.
Actually an insane concept that would sell amazingly
There already was a product that is this exact concept, The Radeon Pro SSG
Ahh yes i definitely need something that will warm my GPU up even more
They have tested the temp, nothing was heated.
@@erictsui8802 gpu w/o m.2 slot = x temp, gpu with m.2 slot = x temp + 'nothing was heated'.
It's as if you don't understand simple logic or the word 'warmer' means.
@@Levinas-od5yx If you can understand Chinese, you could just check the test session of the original video. But you don't understand it, and you're still here writing a very simple equation intended to make your point, how sad is that?
@@Levinas-od5yx Just for pity's sake, I'll outline the test results for you. The SSD in this design is mounted opposite of what they normally are, with all the heat sources from the SSD coming into direct contact with the GPU heat sink, the temp of SSD could stay 42 C degree after 10 min Furmark GPU test.
@@Levinas-od5yx Oh the GPU was 62 degrees C after 10 minutes of Furmark testing, lest you say I missed the point
Nah, what's really cool about this is the potential for the gpu to directly access the NVME storage and essentially use it as a T2 memory cache. That's only if the storage controller for the NVME is on the GPU rather than Mobo/CPU though.
That WAS a thing back then.
no, what would really be cool is if games could be designed knowing that this is an option, and be able to install textures directly to a gpu-attached ssd, meaning that data could be loaded into vram much faster, with much less overhead, and using zero pcie bandwidth
ATI already did this,
It didn’t pan out
@@daniillasso2916 that doesn't mean it was a bad idea, just came too early and didn't give it long enough to be adopted
@@daniillasso2916it was also ATI and not nvidia, which lets be honest, really is the dominating market so if they do something it has a better chance of making it further than if ATI tried it.
I had been paralyzed for 7 years before watching this video, but this video gave me enough energy to get on my feet to open my window and jump out.
Seems that jump max paralyzed your legs this time 😜
fairs
Then you remember you are at the ground floor
Glad I'm not the only one who was disappointed you can't dislike a video twice
Are you paralyzed again?
"bro i got the 8TB graphics memory 4060"
-bro's last words
Companies getting more and more clever on how get consumers to spend
I mean, yeah. That's the point of a business lmao.
@@X_irtzand they could lose that business by overcharged and nickel and dimming their consumers.
@@quest2782That's why competition is a good thing.
@@Nim... yeah until they all start offering the same price
People complain about companies not being innovative enough. Then when the company show the consumer their innovations the consumers complains about the cost, about how it's not necessary, about it's too big/it's too small, about the color, etc. etc.
Today on How to cook your SSDs and lose your data😂
right, imagine this came out on the 4090 FE THE ROFLS WOULD BE R34L
@@extendedclipricco3214please never say that again
@@extendedclipricco3214 what the fuck
I couldn't tell you how many SSDs that were mounted under the GPU that ive had to replace because of this.
@@xero6774my man speaks early 2000s internet
We're slowly evolving PCs into SoCs.
Well, AMD did get there first. I'd expect GB of cache from them soon
@@Demopans5990 Looks like we might get there in 5-10 years.
@@Demopans5990 still waiting for AMD SOC on par with consoles.. If they make a new socket with RAM/CPU/GPU all-in-one that'll be great. They are not catching Nvidia so they should start just doing something different instead.
@@VanSanProductions bro amd lterally did this. You can get a board with ram, vram etc and the Jaguar cores from the Xbox... And stuff it in an ATX case.
@@Demopans5990
Didn't Gamers Nexus make a News video about that?
1GB cache already exists on high end server/workstation CPUs, iirc.
Edit: Yeah, search for it. It already exists.
RUclipsr:-Are you excited?
Broke me:-Oh Hell No! Im Good! Thank You!
imagine being excited to overheat your GPU by attaching a SSD to it.
💀
Imagine thinking that your GPU will overheat because of an SSD...
@@DesroQc depending on the gpu the thermal situation would be far from ideal to say the least
"He's all about it, if it's the future"
@@DesroQcit’s entirely possible if you’re running higher settings than your hardware can handle…
Year 2090: ughhhh, my 10000 terabyte gpu is full again
Well, duh... you're still using measurements and terms like Terabytes from over sixty years ago. It's like when people from the 2020's still measured everything in bits... Wait, they didn't do that then either?!
@@NickNieves112 well, sorry! Misspelling. I meant 10000 gixabyte
@@revenant_z it's just 10 petabytes
@@AbsoluteHuman but its still a lot 😭 i am poor
They will be using googlebytes 😂😂😂
That is definitely good, but I'm more excited for the Fiore System Airjet coolers to be available to the public. The Airjet can possibly be a great solution to over heating m.2
this
I would be happier if they made pci-e x32. These gpus have 2 maybe even 4x more computing power than pci-e can deliver.
He sounds so happy
"How to cook your NVME 101"
How would it cook from this? You do realise that the cooling solution of current GPU's are oversized? There is more then enough cooling reserve in those cooling designs that they can take care of several nvme ssds.
@@germanmoscaSSDs get damaged if they get over 60 degrees and GPUS go way above that even on a 4060
@@ichisenzy Bruh my laptop SSDs are constantly getting cooked
@@gabor6200 yeah you should expect that on a gaming furnace
M.2 ssd heatsinks exist, even with their own fans.
Use that SSD as VRAM… now that would be some good stuff!
That came to my mind too. If the enormous amount of gigabytes of an SSD could be used somehow as a continuation of VRAM. Of course it's a lot slower than normal VRAM, but it would be at least close to the GPU unlike a normal SSD.
Wouldn't want to write much of anything to it considering how much data a Gpu goes through, it would kill the SSD fast. For reads that's ok.
my thoughts exactly! or even a dedicated slot for AMD SAM (smart access memory)
@@SlofSi
AMD smart access memory is kind of in line with that.
would be really cool to see.
This is what I was thinking he was gonna say, but nope, just more storage 😩
Not only that it cools off the SSD using the GPU fan, but also transfer the GPU's heat to the SSD itself
Toasty
ikr.. putting the SSD on my already 80°C GPU that is at its cooling limits does not really sound right 🤷♂️
micro builds would go crazy with those
Heat transfer has some specifics as to why mounting SSDs onto GPUs is a bad idea.
Sure, it can cool your SSD. But only if your GPU is not producing any heat or produces less heat than the SSD itself. And GPUs generate a lot of heat, in most setups they are the main part you are trying to cool down. And that heat goes into the heatsink and PCB. And from that PCB or heatsink it would transfer to the SSD.
Upper limit for SSD temps is between 70-80 degrees Celsius. GPUs often operate at those temps during games and if the heatsink is at 60 degrees during gaming session, an SSD that would be otherwise at 45-50 is going to soak up the heat and get warmer.
So while in theory it seems like a good concept, you'd need to thermally isolate the cooling, otherwise you'd just warm up the SSD for no reason.
the i9-13900K says hi
@@MünWeeew oh you mean my sandwich warmer it works as well as my microwave LOL
finally someone who says right, this it's just like the sd card burning issue from rog ally , same spot near transistors
Water cooled GPU and SSD lmfao
Back in the history, graphics cards had user installable RAM slots. I was excited we were going back to the future for a second.
the only problem is that it ads heat to the gpu which makes a lot of heat by itself that's why they are getting so damned fat🤣
@@SaraMorgan-ym6ue The GPU Chip is bigger that's why they are thicker, and the GPU generates more heat then the ssd, so it isn't adding heat. GPU's triple in size and complexity, so do cards, it's 1:1.
In a closed system, adding heat will increase the temperature, but it will not rise infinitely. The temperature will eventually reach an equilibrium where the heat added is balanced by the heat lost to the surroundings. The temperature will not exceed the heat source because the system will reach a point where the heat input is balanced by the system's ability to dissipate heat. This is known as thermal equilibrium.
Number still goes up
@@ParadoxISPower
This is actually pretty cool.
Kinda how like sound cards would pull double-duty as a joystick adapter back in the day.
Ahh man, i forgot about those times.. pretty cool
It'd be really useful if the gpu can access the ssd and copy data to vram, otherwise it still gonna have to allocate space in main memory and then be allocated in gpu, which doesn't change anything
Did you even watch the video? Its not about performance, but putting the PCIe lanes to use.
@@banaana1234 Yeah, and they're saying that it'd be useful if it did that.
@@banaana1234 Which is about performance anyway.
"Look at all this empty space. It could be used for something"
"Maybe it could be used for ?"
None of this is about memory allocation at all, just direct PCIE addressing to the extra device, which is supported in hardware already.
It's called resizeable bar. Look it up!
what would be more cool is to see something similar to that but as "VRAM upgrade" like you could buy VRAM and upgrade VRAM capacity of your GPU
thats not going to work SSD are far too slow, imagine running DDR2 with increased latency instead of GDDR6/X
You going to have a stuttery mess, not only that the traces to the memory modules are the same length, now this SSD is completely off the timing.
It can never work as a VRAM upgrade because VRAM is so much faster and integrated into the design of the board
@@imo098765 I mean who knows, we live in a world where technology is exponentially developing itself 🤷♂
vram is dirt cheap so they won't allow it , they want to charg us fortune for more vram on their cards.
It was a thing before, where there was ram slots on gpus
From what ive heard, you could do this on very old graphics cards.
Imagine a GPU... that comes with ability to add extra VRam to the card. This needs to happen.
This is what I want
This is exactly what Apple does with its Apple Silicone. Whatever ram you get also goes towards the GPU. So if you order 256 gb of RAM, youll also have 256 gb of video ram.
@@hajjdawoodthat's also what every single other laptop or desktop pc with integrated graphics does, but apple markets it like they came up with the idea
@@obj_obj There is several key differences, first its actually good integrated graphics (scoring better then the 3090 and worse then the 4090), second because its a SOC and the ram is also on the same chip the memory bandwidth is 400 GB/s verses a normal integrated cpu/gpu like then Intel i9 runs at only 89 GB/s
I think its a bit unfair to act like Apple didn't make a very strong leap in CPU technology. Non-apple laptops are struggling to keep up with the nearly 4 year old M1 chips especially when it comes to performance per watt.
Apple almost never invents anything new, theyve admitted as such many times over, but when they do something they typically do so really really well. iPads, Apple Watch, the M series chips, the iPod, even smartphones (iPhone) apple was definitely not the first but they did a really really great job that leap frogged them past competition for years and years.
@@hajjdawood This feature is extremely old and has been used by windows machines forever. However, if you actually think that you can buy 256GB of ram and allocate it all to vram, you're not just uneducated on the subject, you don't even know how Macs actually work, because what you're claiming isn't even possible on an M3 Max.
This could also be great for storing compiled shaders so they can be accessed faster than through the motherboard SSD. Perfect for when RAM isn't big enough and runtime compiled shaders require too much GPU power.
They should do this with VRAM. All my problems will be solved
Shit yeah.. it saved my marriage!
In another timeline:
Internal Email: NSA can now use the GPU to scan your files...
Don't worry mate, you're most likely not interesting enough to be monitored
@maxjes5 this. Hahaha, unless you got a few files thatll land you in prison i wouldnt worry. If the NSA wanted into your network theyd be there.
With trust replacing secure practices, stega usage getting really creative, & the price of meta analysis at an all time low, everything is now "interesting enough" to be monitored; it's just that very few things require overt intervention
@@Papitotihaha that's what you think...
Why would I care don't live in the US.@@maxheadman59
As an ITX builder this sounds amazing. No need for storage options on the back of the motherboard with no airflow.
then you put an ssd on a piece of hardware that gets near 80 degrees normally?
I kinda wish that Low End, Entry Level or Mid Range can handle any games at 4K max settings.
This idea could be nice to let SSDs access to more memory to load game objects and/or shaders without the need to involve many CPU instructions at the process.
Edit: in terms of global computation, drivers that communicate GPUs with CPUs follow interface rules that may not be as optimum as the GPU hardware itself loading information from own disk memory. For those who say this is a limitation, it really isn't because during the "long" process of initialization of the game/OS, the reality is that the GPU obtains one more read/write channel that is independent from the computer's main disk space, leading to faster effective load times just because the GPU is avoiding two things:
1. Driver control instructions overhead (may be different between applications)
2. OS main disk read/write operations that may be used for other system internal purposes, or not parallel instructions that require high performance CPU clock times.
And this also introduces a new potential paradigm where GPU programming becomes essential to write programs that let the GPU store it's own internal files/resources into a totally different memory channel without generating overhead with main disk memory requests.
So yeah, it is just better, whenever disk speeds are similar. And people complaining or throwing arguments each against other should shut up and think they're not spreading really accurate information, but can offer a different view where disadvantages instead of spreading a useless fight.
Thanks.
Right. that was my thoughtprocess too.
Letting it act as an on-card cache.
True
@@The_Keeper the memory on a GPU isnt close to being the same as an SSD LOLL
@@The_Keeper GPU memory is GDDR6... An SSD would never come close to being as fast. I can't believe you guys thought you could just strap a 1tb SSD on a GPU and call it a day 😂
Edit: GDDR6 NOT GDDR6X
@@micaheiber1419 what does closer proximity have to do with anything?
EA's graphic card: We've locked out a portion of the RAM. Pay 9.99 a month for full access, or 59.99 a year for full access.
Barf. They would do this too
@@OldRedOwlthey already do this all 4xxx series cards use practically the same PCB. They just install less for each model. Con is less performance and lesser image quality. Cons are cheaper price.
You know, BMW thought about that. They try to sell subscription for heating seats. 😅😅😅
We can definitely see something like that on pc soon. Hopefully, people won't fall for it. Otherwise we are F***d😢
Intel actually did this on CPUs at one point. People were not happy.
@@piotrkowalski3869this.. isn't even remotely close to the truth. You don't understand how GPUs are manufactured
These graphics cards require you to split the x16 lanes into x8 for the GPU and x4 for the NVMe drive using what’s called “PCIe bifurcation”. Even attempting to use such an SSD on a GPU like this requires your motherboard to support PCIe bifurcation.
I’ve ran server disks using PCIe bifurcation before. It is a nuisance to set up, a nuisance to manage, and the payoff is negligible (assuming you have space M.2 slots on a standard motherboard, use those instead).
So no, no I am not excited for this GPU or this approach at all lol.
But I feel the marketing team will like it...
I could imagine this adding additional input lag tbh
What a load of nonsense lol. What you say is simply not true.
I don't think you inderstand what's going on here. bifurcation is a pain because it turns 1 lane into 2 lanes.
this GPU only uses 8 lanes and so 8 lanes are still available for something else. 2 NVMe drives, for example, which only use 4 lanes each.
1. Most of the AMD motherboards support bifurcation by default, but motherboard manufacturers do not include BIOS interface to configure it.
2. This does not require lane splitting.
Great timing for this video.
I was recently wondering if video card slots needed an update in order to improve GPU tech. I didn't realize that they had too many slots, but with how heavy graphic cards are they need all the mobo support they can get.
Note: I got a video card stand recently.
Honestly this could be super cool if GPUs in the future had like expandable VRAM
Actually, they used to in the past. I still remember an old card from my early days that had 4 slots to plug in memory modules on the gpu for expandability. The only reason we don't see it anymore is because the timing of the memory modules running through the circuit to the gpu die itself would be too slow for modern components. Faster than you can think, but still not fast enough to keep up. The memory modules would have no way to compete with memory directly on the gpu die or on the card immediately next to it. Sad because it was such a novel idea and it was brilliant at the time. 500MB GPu, quadrupled to 2GB. That was a ton back then. Having 1GB even was like, baller.
@@lexecomplexe4083 Yes I’m certainly much newer to PC building than you the first ever GPU I bought for myself was a 1080
But how will nvida be able to make money? ;(
The 4070 would easily become a commonly used card😊
I have an old computer with an IGPU on board, with a slot for a VRAM card
Me who uses an 13 year old laptop:
Get external one like me
Don’t we all friend
I'm still using acer 8920
Hey - just one R in the recycling triangle: reduce, it helps.
I also have a new one but it's more laggy
Putting an ssd on an oven is the most idiotic thing I’ve heard.
it'll be on the coolest part of the card, add a layer of thermal transfer tape to it, touching the rad that can effectively cool the 2nd hottest component with no effort? ;) i would genuinely love to see thermals with/without an ssd on the card. I think it's a really good concept especially since gpus typically vent their hot air directly out the case . well alot of the air
dude that end of the gpu is not very hot at all. there's a fan blowing through the heatsink, that's it
Then don't put it on the hot part moron? 😂
Do you even solder, bruh?
The backplate runs hot as hell, you can cook an egg on it. I put a fan on my gpus backplate because it is too damn hot for my comfort.
It sounds like theyre selling us incomplete graphics cards
Honestly if it works with the EGPU market this is a big W for that. Not only can you have an external gpu to run your more demanding games but you can have storage as well Without too much hassle. W
Also more storage options in mini-ITX builds. Cooling would be a pain, but a big enough heat sink to handle both the GPU and NVMe isn't a far dream.
A singular 4090 chip probably runs hotter than both combined.
Kind of how back in the day we would install games on a USB stick to play on school PCs as we couldn't install any software on them, just now you bring your own GPU with it as well so you can play games on any pc with a decent processor.
I'm more interested about the DirectStorage implications that this may have, since the SSD sits closer to the GPU and could significantly improve framerates during asset loads.
Interesting, but if your SSDs sit closer to your GPU then just run GPU Direct Storage instead of Direct Storage. Both GPU and CPU have specific condition in which their workload can be handle efficiently.
The short distance is irrelevant. The gpu is only giving the SSD its 8 unused lanes, their communication is still through the cpu
@@Lagger625on this concept build yes, but the point is theres no reason why a manufacturer couldn't build in an SSD like this and run it as GPUDirect storage, passing between the gpu and the ssd without needing to run through the cpu for everything.
@@branbrokenthey would need to basically add storage controller lite to GPU for that to work without involving CPU PCIe
I like how you think.
Not that unusual in the past. A lot of Amiga boards often had multiple items on them such as RAM on a SCSI HD Controller. I think it had something to do with how Autoconfig worked with Amiga Zorro and CPU boards.
So this is a retro idea that is still a good idea.
Alienware x51 kinda did this
In this case it's more akin to a pass-through port. The GPU doesn't talk to the unused PCI lanes, it just routes them to somewhere potentially useful, like an M.2 socket.
I had to cut a hole in my A1200 for a 80GB drive that cost £315 back in 1999
@@Space-O-2001 Did you stick a full-sized drive in there? lol
There's a reason I really liked the big-box Amigas. :D
@@MarkReed-smokindeist Yeah 3.5", 2.5" weren't really that common. A2000/A4000s weren't really in my budget back then lol
This is the symtom of moving from SATA based SSDs to PCI-e SSDs. While they are fast, they ultimately take up lanes. Its one of the few reasons to run SATA. But if you want real speed, you need SAS. Its niche to the enterprise market, but think a simple upgrade to SAS-2 or SAS-3 even would be a game changer for those that want to run SSD RAID setups. I personally run SAS and I love it! There's nothing better than SAS for both external and internal capability. Its a truly flexible system. It's a truly the wonderful successor to parallel SCSI.
It’s a pretty interesting concept. But what if you could add more VRAM via a port on the back? Any takers?
Too slow for gpu
Too slow, terrible bandwidth too.
Some video cards in the 90s had a slot for extra VRAM
@@Pasi123yeah and it wasn’t good
Imagine trying to configure VRAM timings just as well as RAM timings... and then there's the additional variable of extra VRAM port. UGH.
Sounds preety good in theory and i am excited for the result😅
dead nvme back plate of gpu gets hot and isn't cooled by the heatsink and fans from the front. lol
Imagine if GPU manufactures took that concept, and used an SSD on your GPU as a way to handle shader caching in games, freeing up your PC HDD from bulky shader files that build up, with some games.
Thats smart ngl but why not just use sn ssd than? Moreover a lot of bypassing required to do so
HDD? What year was that written in
"Freeing up your blah blah blah to do more..." You know full fucking well they ain't freeing up shit. Performance will get even worse with more hardware resources available as lazier and lazier unskilled devs are able to use modern specs to skate by with even worse optimized games. The 360 lasted over 10 years on market running the same games under the hood as modern pc's and consoles with only graphical hits taken, all on MODEST specs for when it came out back in 05. Now with 8gb and now 16gb of ram standard, internet browsers take up 10gb for like 5 tabs and youtube, and standard run of the mill fps games can max out ram usage qhen they used to get by on a gig or two max for the same amount or characters, physics and underlying game logic being performed. Even new graphics technology isn't being utilized by the majority of games, just the additional vram for people who don't understand how to optimize their texture cache usage and just demand you pay more for a car with more GB, or demand that the vram be gddr6 instead of gddr 3 or 4 like it used to only a few years ago. The new technologies ARE NOT ground breaking or genuine improvements, just a slightly larger bandwidth which is nowhere even close to being needed for consumer software applications. These nvme's on gpu's will only be another burden placed on the consumer's shoulders by shitty third world devs that cannot actually develope software at a professional level
and why exactly would they do that when they could just solder additional memory if they wanted to without need for ports and shit?
Good thing you mentioned the SSD Temperature issue. When an SSD overheats it freezes its read and write to zero.
Waiting for gpu that has expandable ram 😅
holy that would be so awesome
i also thought of that like we can just add a sodimm ram stick to the gpu and get better fps
@@kayukshi4271nah ram and dram are quite different. Pretty sure like the 1030 or something had a varient with ddr4 instead of gddr5. And the difference it made was a lot
@@wolvreigns maybe a sodimm ram stick with vram for gpus might work
@kayukshi4271 no, it wont.
Well there is also a chance that the SSD will corrupt due to the heat of the gpu, which we all know, it heats up a lot especially on gaming. If the SSD can withstand the heat of the gpu, then I'll support this concept.
The position they installed is pretty far away from the hotspot of the GPU
Non-issue. The air coming out of the heatsink is lower temperature than the SSD heatsink and will cool the SSD, even under load.
Probably best for watercooling. Some sort of waterblock that cools both
If it is designed with that in mind by competent engineers, it would not be an issue. The real problem is that you’d need an additional, expensive chip to do the PCIe bifurcation.
@@Demopans5990Pretty sure that if you have the money for a water block, you might as well get a higher end gpu that uses all 16 pcie lanes.
Could work, though it shouldn't be standard since most gpus do infact use all 16x lanes.
i am as nerdcited as u are. This is awesome.
AMD did this 5+ years ago with multiple SDD for TBs of extra storage. Imagine if they had their current RDNA and card architecture.
That device was a poorly conceived product. It was a x16 GPU meant to utilize that slot. The two SSDs added 2 each X4 devices to that x16 slot. All of these lanes connect to the CPU meaning that the slot is 150% oversubscribed. You would only see contention between the lanes when you utilize them. In the end the device would flounder and drop frames when you put it to task. As for the GPU speaking directly to the GPU, this could conceivably work but only the GPU would be able to utilize the SSDs . It’s better to keep devices on discrete slots/connectors for reliability and serviceability.
@@3DMarkRGBA true but some mods and even new games going crazy with ram requirements this might be helpful although the heat being pumped out by some of these cards might just cook SSD's
Speaking out of your ass about it floundering and dropping frames, first of all it was a professional card not for gaming, the use case was very specific in video production and CAD.
This is a smoother loop than those smooth loop videos
This is so simple yet it took someone with an unorthodox vision to think of it! I hope whoever was behind this gets appropriately credited.
Soon we'll have threadrippers built into our 5090s lol
Use an SSD as VRAM, now u got 8TB of VRAM
great for if you want your gpu to be both low-end AND high-end at the same time.
😎😎😎😎
@@namedauuidhell yeah time to minmaxx my 4090