Nope... Because there are components that are extra costly and don't provide that much performance. And in fact, cheaper components can provide as much performance as pricier ones. So it's actually more about Finding the Best Value than just simply blindly go buy the pricier things.
@@Trip4man"uhm akshyually" lol. It was a good joke the dude made. Sure, you're right, it's not always true that spending more equals greater perf. But on average, for most people, increasing budget will allow for better performance.... And the joke was funny! So chill lmao
I just upgraded to 5800x3d from my old 2700x.I see about 30% performance gain in average FPS with my 6700XT, and the stuttering, frame time spikes are all gone, I can finally enjoy fluid gaming in most of the games.
it really is amazing how much of a difference going from a 3600 to a 5800x3d made, even while using a measly RX480 and technically "gpu bound" 99% of the time. Those moments when the CPU slogs you down really ruin gaming fun
30% sounds low, maybe lower some settings if playing anything competitive. I went 38xt to 58x3d, I was CPU limited in ACC and EPIC graphics settings everywhere didn't impact max / average FPS much. Dialled back some settings that are pretty useless when playing and almost doubled FPS at race starts, average 50% higher overall.
Isn't it great that V-Cache was just a skunk works for funsies thing one of their engineers cooked up? Wasn't even really in the plans until the prototype was super impressive. Info from someone's AMD tour... Maybe LTT? Gamers Nexus? I can't remember now
@@TheDarksideFNothing That was Gamers Nexus. Interestingly, Threadripper has a similar origin story. AMD seems to have a company culture that lets their engineers experiment a bit and it's paid off for them greatly.
@@Breakfast_of_Championsi7-5775C's technology isn't even close. That is an L4 eDRAM cache attached on the side. Not only is eDRAM much slower than SRAM, it's just a standard MCM (side-by-side) module attached through a standard bus. The incredible magic of 3D V-Cache is that it adds practically no additional latency, because it's literally right there, where the regular L3 cache logic is. Any company can make a giant chip with more L3 cache, but that will lead to additional latencies (bigger == further away), and cost more due to sinking yield.
When I saw the title I thought, "Haven't you already done this with Intel?" I wonder what would happen if you took the 64 core or 96 core Threadripper and disabled all but 8 cores. Would that give those 8 cores 384 MB of L3 cache?
Depends on the layout if that's beneficial or not. Accessing cache on a different CCD induces a hefty latency penalty that would reduce performance in most instances.
As someone who plays a lot simulation type games, I am continually grateful that you included Assetto Corsa Competizone in your testing suite. For people that primarily play racing sims, flight sims, and large scale military sims, like ARMA, testing and comparing CPU cache as well as core count is integral to find out what hardware is the best choice for these kinds of titles. The way these games operate is so greatly different than most other games, mostly being console ports with not a lot of instruction sets being sent to the CPU in comparison.
It's allways worth revisiting these types of subjects if only to help newbies learn more about the machines they are buying. Also updated / expanded testing data is allways good.
Funny we forgot the lessons learned during the Core 2 Duo and Quad era. The extra cache on Penryn vs Conroe (especially 2M Conroe) mattered more than the # of cores for gaming
Cache itself is important, but it depends on how accessible it is to all the CPU's cores. If you have 20mb of cache but a core can access only 1/8th of it, its far worse than if one core can access all of the cache. That's basically why zen 3 is so much faster than zen 2. A core can access twice as much cache on zen 3 compared to zen 2
This here is about L3-Cache that is shared between all cores. But yeah L1/L2-Cashes also matter but are not easily comparable, because they are usually the same for the same architecture. And when comparing between two different architectures, there are more factors than just L1/L2 that are responsible for the performance gain.
Generally speaking, only L1-I and L1-D caches are private in modern CPU architectures. L2 cache is "on the core" meaning it's physically allocated to each core, but other cores can "snoop" L2 caches of other cores. With Zen, this can only happen inside a CCD, so Core 0 (on CCD 0) cannot access the L2 cache of Core 8 (on CCD 1). This is partly why multi CCD Zen chips are not better in games. L3 cache is again CCD-public, meaning any core on the same CCD can access the L3 cache, but other CCDs cannot. As you mentioned, with Zen 2, a CCD was 4-cores and a CCX contained 2 CCDs. With Zen 3, Zen 4 and Zen 4c, a CCD is 8 cores.
A good example for Zen 2 is Ryzen 5 3600 vs Ryzen 7 3700X. The Ryzen 5 3600 has 32MB of shared L3 cache between cores and the other one does not. But the difference in performance is only 5%. But that can be attributed to more cores and a higher frequency rather than a difference in cache. People have learned that the cache makes a big difference in Zen 3, but then they try to apply it retroactively to older architectures as well. I don't think Zen 2 was designed to benefit significantly from more cache.
@@JackJohnson-br4qr the cache on zen 2 was shared between 4 cores, thus an 8 core 3700x did have 32mb of cache but each 4 core CCD could access 16mb. That changed with zen 3 where each core had access to the 32mb of cache as there is no CCD anymore
@@cpt.tombstone except that dual ccd chips are faster in games, not by much with varying results, but in general they are the same or faster than single CCD chips
Excel, browser and other productivity applications' impact on core frequency, ipc and cache will be much appreciated as most of the day to day tasks are still single threaded
Yes, I'd be very interested to see all the same CPU's tested against general apps/productivity. Especially as the much larger cache on the X3D chips usually results in lower frequency, which in theory matters more outside of games, but it would be nice to see all that confirmed.
I was running Intel for at least 4 of my last builds and was about to go for the 14700k a week ago. Then I stumbled over some information on the lifecycle and the fact that AM5 would be more future proof for another couple of years while being superior for gaming anyway due to the cache. And then I also noticed that the R7 7800X3D also was way more efficient and cooler. All that while costing less in total together with an Aorus Master mainboard. Had to reconfigure my cart eventually and go with AMD of course. Super glad right now.
@@johnfirst3986 I went with - Corsair 7000D Airflow (Big Tower) white - Aorus Master B650E (Mainboard) - 7800X3D (of course) cooled by Noctua NH-D15 - Corsair Vengeance RGB 32GB CL30 AMD EXPO (RAM) - Corsair RM850x (PSU) - MSI GeForce RTX 4080 Super 16G Gaming X Slim White (GPU) I had plenty of storage left from my older machine. And I game at 1440p for the most part. The build is overkill for what I need as I mostly play fighting games. Even the most recent games will not need the 4080 to consume more than 50 watts.
This was one of the most informative CPU videos I've ever seen. Good job making great content in a time where there isn't much happening as far as new parts.
When it can be utilized yeah, but a brute force single core combined with low latency is more consistent in its results. That's why I'm interested to see how arrow lake pans out considering Intel is ditching hyperthreading for the sake of single core performance.
Core count and mhz are linked. High cache is about removing latency issues. If your cores or high mhz cpu is waiting for a chunk of ram from system ram the it dosn't matter if you have 6 or 7ghz cpu. It will be idle. A high cache helps to mitigate this issue by guessing what you might want to load in the future and keep it closer. The P4 tried to do this but their long pipeline killed the gains from the cache. The P4 had to flush the long pipelines and it took a long time.
12600k alderlake vs 14400 "raptor lake refresh" Lock the cores and caches to the same frequency (and tdp's) and see if there's any change in architecture... because they're the same die.
I always follow the suggestion/raccomandation from Steve and Tim , NEVER let me down !! CPU , GPU , Monitors , I based my purchases on this channel , I did NEVER regret ANY decision. They are the best IMO.
It's been implied by testing, but L2 cache is incredibly important as well, arguably with more impact. Raptor Lake's performance improvements are almost entirely based on in a large increases in L2 cache (not all 13th gen did get L2 cache increases, basically it's 13600K+). The nVidia Lovelace architecture also saw massive gains by increasing L2 cache sizes. At a basic level, L2 cache is "easier" and less expensive to implement than stacked dies.
L2 just has less options for blowing up majorly in size. I think even the latency of V-Cache would be too great for L2 IIRC Would be interesting if they figured a way to use all the L3 space on the die for L2 and then use V-Cache only for L3. Best of both worlds.
@@TheDarksideFNothing "L2 just has less options for blowing up majorly in size..." The "problem" is that L3 size has significantly diminishing returns in performance gains. You could probably cut the L3 (per core) in half on an X3D chip and see very similar results.
@@awebuser5914 You are probably right about diminishing returns, but I wonder if the relative ceiling of L3 cache effectiveness on Ryzens has even been reached yet. Right now it's obvious that 96MB is much better than 32MB (for games), but what if they tried 128MB, 160MB or even 192MB of V-Cache? 😁
I will be forever thankful to you guys for the review of the 5800X3D almost 2 years ago. If it wasn't for your benchmark with ACC i probably wouldn't have jumped on the X3D train and wouldn't have experienced the monster that this chip is. 18 months plus and counting and still feel completely blown away by the performance every time i load a game.
It's a shame no one ever tested it with the original AC (+ CSP). I only recently decided to make the jump to X3D, upgrading from the 3700X to the 5800X3D. and my goodness was I not prepared for the difference. I legitimately get up to 105% higher framerate. Yes, over twice the framerate. And that's with a 4060 Ti, so the potential gains with a higher end GPU can probably be much higher still. X3D is such a blessing in many CPU-heavy titles.
To me bigger difference would be cache vs frequency I am using X99 based Xeons and some games benefit so much from cache that it doubles my framerate despite running
good call to be honest. we already know 12 threads is more than enough, but frequency VS cache is a whole different animal, as some games will heavily favor frequency, whereas others will heavily favor cache.
At this point, all desktop cpus run at outrageously high (inefficient) clock rates. Everything over 3ghz is mostly a waste of power. Look at gpu clock rates.
@@Eidolon2003 Agreed, this whole test is done on cpu's starved of ram info throughput. Zen2/3 have high latency to ram and ofcourse a cache buffer will diminish that problem somewhat. On Intel one can oc ring and cache for over 6Ghz scaling. This has been true for over 10years on Intel. Only HW likes to pretend this isnt doable on Intel, rofl. On Intel no oc guy leaves ring and cache on stock speed when oc'ing.
The other thing not mentioned is that cache can help vs clock speed. The 5800X and 5600X chips both have higher clocks and yet are showing some serious performance difference in gaming.
I use a AMD Ryzen 9 5900X with 64 MB of L3 cache along with 32 GB of DDR4 running at 3200 and i never feel it need more it runs every game and application i use with NO problems hopefully it will keep going for a few more years yet
To be honest, after the release & reviews of the X3D cpus we kinda already knew those results. Steve you are an absolute legend mate.... I was honestly just writing "What would be an interesting, but also almost impossible to accurate measure, is the number of cores that start to make a difference when playing games in a more real life example and not in a benchmarking environment. So, basically, what happens when you play a game on a windows pc that is not a clean installation to get accurate results for the hardware you test, but when you have youtube, discord, motherboard/peripheral background software, various tabs on a browser open, etc"... And I see you 've already covered this on 15:00
So 6 cores are fine with just gaming and more cache would be usually better . How does it hold up when u also streaming from the same pc? Would a cpu with more cores then be better, or a cpu with less cores but more cache? Dunno if this a yes or no question or 'it depends' :P
Multitasking requires the CPU cores to share the rest of the memory and storage subsystems and therefore increases latency and latency dependent games will suffer regardless of core count. That's why the E-cores on ADL and RPL taking care of all the background tasks is somewhat of a fallacy.
"That's why the E-cores on ADL and RPL taking care of all the background tasks is somewhat of a fallacy." Better scheduling fixes software CPU overhead, that's why my 12700K is perfect for a DAW and chips without Big-Little aren't that good. What Steve shows in bars doesn't reflect the real world when actually using these types of CPUs, same for a lot of results i watched on the internet when these techtubers are ignoring music producers. When i disable the e-cores on the i7, i lose singlethread speed and gaming perfomance even though I'm freeing cache for the big cores. Depending on the game, it could be a hit to the frametimes (a small one with some exceptions).
When the scheduling works perfectly, we get a massive 30% increase for gaming perfomance (watch APO on the 14900K as an example). Cache and memory aren't everything. Imagine how much better the Ryzen 7 1700 would have been at launch if the scheduling worked fine, nowadays it outperforms quad core i7s from that time due to the lack of cores. Simply search which CPUs have the highest IPC, singlethread speeds and thread counts, the rest is irrelevant (minus number of PCIe lanes).
Can we see a cache vs clock speed video? How much do the non 3D skus make up in performance with their clock speed increase over the 3D variants? Only a curiosity of mine because I moved from 7600x at 5.5ghz to a 7800x3d at around 4.6 ghz.
In these charts, the 5800X has faster clocks than the 5800X3D. The 5600X has faster clocks than the 5600X3D. In simple games, faster cores can be an advantage because those games don't fill the 'normal' cache sizes. (Anything in a modern engine like UE4 will be more complicated than this.) Productivity workloads are often spamming a small set of instructions on a large amount of unique data. Cores and clocks help those more than cache - if the task can be spread across cores, that is. Conversely, going from a 5000 CPU to a 7000 CPU is a change of architecture and memory goes from DDR4 to DDR5. The also doubled the L2 cache on each core. ~15% more instructions per cycle in each core and more memory bandwidth alone should give ~30% more performance for the same clock+core+L3 cache configuration. 7800X3D should boost up to 5050MHz per core. Have a read about PBO, curve optimiser and the +200MHz "Boost Clock" setting your motherboard BIOS should have. Task Manager might not report the full clock speed you are getting. HWinfo and similar tools will show it. If your motherboard has an external clock generator, ScatterBench has an article about overclocking X3D a bit. It may even have a built-in, pre-tuned overclocked profile.
I got my 5600 luckily by chance. I saw a 5600G for sale on Facebook Marketplace and I pm'ed the seller. We met the next day and he said to me that he has two CPUs, a 5600 and a 5600G. I paid for what I bought from him and when I got home I was surprised it was a 5600, instead of a 5600G. He said he forgot which box it is but I said to him it's fine. Best decision ever.
Hell even a modern game you could fit enough of it in there to easily mask any data swapping. We already see where some games see no benefit because they've already optimized to fit in normal amounts of cache so at some point you lose the benefits. But I do wonder if a dev KNEW they were getting 1GB if they'd be able to take advantage of it in really interesting ways.
Yeah the one thing I keep wondering about how one could optimize if they knew they were getting 1gb of cache. i personally wonder if it would help raytracing performance at all since that hits performance the hardest@@TheDarksideFNothing
@@OtherwiseUknownMonkey Yeah, I think a full GB of cache would be much more about seeing what new things you could do vs making existing things go faster. Right now all things are designed around small caches because that's the hardware that exists in mainstream. But some applications miss the mark, and that's where V-Cache shines.
@@TheDarksideFNothing i wish i could hear a tech artist talk and a programmer talk on it, with 1gb cache on cpu you could make worlds feel so much more lively eith more intricate ai routines i imagine, and if you had a gig of cache on the gpu you could keep whole lightmaps in there, like lets say the game let the gpu know where the lighting will be at in 10 seconds from now and you could smartly interpolate those lightmaps that will still be in cache the whole time making rendering sm faster
This was why i upgraded from a 3900X to a 5800x3D, despite the core deficit i mostly play games but also i wanted a single CCX 1x8 rather than 4x3 core because of the cross-talk between cores over the infinity fabric and accessing the cache. Still using an RTX2080 so while over all fps has not changed much (GPU limit) but my 1% lows has been reduced by over 50%. Also games don't like too many CPU cores as it messes up the scheduling which epic has came out about crashes on Intel CPU's because of core count.
The 3D V-cache absolutely makes a different in frametimes too, I went from the 5800X to the 5800X 3D and the caching stutters went away, I was getting insane shader cache stutters on the regular 5800X and they were non existent on the 3D version, this is after a full driver install and the cache cleared.
Thank you! I'd rather add as optional opponents 4600g and 4700g. They have only zen2 cores but twice less L3 when compare to 5xxxG parts, so L3 compare chain can be wider - from 8 to 96MB :)
I followed your recommendation (5600X) two years ago and never regretted it as it runs everything very nicely, in fact I built my whole system following your various videos on hardware components. I briefly thought of upgrading to the 5800X3D when it came out but then I realized that I don't really need it for the types of games I play (mostly strategy games @ 60 fps).
Agreed. I did the same , always follow the suggestion/raccomandation from Steve and Tim , NEVER let me down !! CPU , GPU , Monitors , I base my purchase on this channel , NEVER regret ANY decision. They are the best IMO.
An i7-12700K with DDR5 is faster than a 5800X3D, it's similar in gaming perfomance to the 7700X while clocking 800MHz lower. It's a streaming beast as well... If i waited a little bit more, maybe i would have a 7800X3D. The 5800X3D also has an annoying flaw, it's a jittery mess for the clocks, depending on the game i even watched a Ryzen 9 5900X being smoother for the frametime graphs than the 5800X3D for Rust. I don't like CPUs with bad clocks or power limits. I feel the jitter as well for the Windows UI on those CPUs and other things.
IMO the clockspeed masterace will be end soon (in main use) just like Intel did back in the days with the Pentium4. To much power/heat to dissipate comparate to performance. I belive ,the big and fast cache, will be the proxime future primary choice for the productor .@bra2867
With how a lot of CPU heavy games DON'T have multi-core support, it looks like the emphasis on cache is a better idea. While not optimized well in the first place, the X3D CPU's have been a blessing for Ryzen users who play Squad.
The "problem" with games is that each frame is a down-render of a specific point on a time line. This means that the synchronisation thread will always be the bottleneck. A tile based renderer doesn't give a rats arse about what happens if one thread finishes before another. It will just keep piling jobs on any available core until there are no more jobs. A game cares very much about exactly where each individual animation is at exactly the moment when it's time to freeze everything, down-render the geometry and dispatch it to the GPU. I'm not saying games are the epitome of hardcore optimisation, as we get plenty of examples where a new game has less going on than some other older title, yet still hogs CPU power much worse. But I am saying that at the fundamental level there is a hard limit to how well you can make a game utilize asynchronous batch processing without the synchronisation thread becoming overwhelmed.
Hey Hardware Unboxed. I was curious about Intel 12th gen. How does an i5-12600k with 20MB cache beat the Ryzen 5 5600X with 32MB cache in most games? There are still some other factors when talking specifically about gaming performance. I can see that cache improves performance quite a bit on AMD but how can intel cpu’s with lower cache keep up? 12:59
I remember many hears ago buying a used Opteron X2 170 for gaming. It had one more core and a boat load more cache then it's Athlon 64 equivalent, and even being 200Mhz slower it out gamed pretty much any Athlon 64 because of the added cache. I was also able to sink a 1Ghz overclock on that bad boy and REALLY crank out the performance.. good times.
Been saying this since Intel's mesh architecture (I had a 7820x), so it's great seeing more and more videos confirm cache is so important! Great video.
Cash is indeed king XD.. AMD really struck gold with 3d vcache.. would be interesting to interview a cpu engineer to understand more about how the increased cache helps
I LOVE the “for science” videos. Extremely educational for clearing up people’s oft-speculative, sometimes wishful musings. Since there’s a decent bit of time ahead before the next gen, perhaps we can revisit these varying cache configurations (especially the X3D variants) with cores disabled down to 4, 2, and even 3 cores! As performance and even stability starts to drop off, how much does cache influence the result for what number of cores? Does a triple- of even dual-core X3D config run some games where a stunted L3 quad-core can’t? Come on Steve! Do it FOR SCIENCE!!!
Something that has to be noted is that cache architecture is far more complex than just adding more even if it's possible, the corecount, the cores, the task and the L1 and L2 caches affect how much L3 is optimal. In general a larger cache is slower so there is an optimal value for the cache for each task. Slower cache means it takes longer to access it but it means you have to access the even slower RAM less often. If your task has very little memory requirements (takes very little ram space) then a cpu with larger L3 can actually become slower. Games get heavier and heavier so we might get L4 cache soon in the form of HBM on the chip and the L3 caches might get smaller.
Hey steve. Love the vid. Whould love see a vid abpit whether an ssd with dram cache makes a sense for a gaming pc. Does more expensive storage make a differnce?
Hey Steve, This was a great video, I was just looking into 5800X3D as a potential upgrade path as it does seem to regularly get discounts here in the EU. I am kind of curious how it would compare against one of the 12/16-core SKUs though. All of these have 64MB L3 cache, but boost higher and are generally better value than X3Ds for productivity. I'd imagine many of us also use their "gaming" PCs as a workstation, so I wonder if the gaming performance gain is significant enough to justify a 5800X3D vs. 5900x for example. In your original review the average 8-game 1% lows came up to 150 vs. 121 FPS respectively, but the X3D got blown away in all productivity tests. What I don't get is why the 64MB L3 cache on the 5900X/5950X did not seem to make any difference in these tests vs. the non-X3D 5800X (32MB cache). The way I understand it, the 64MB should be shared between both CCDs (as opposed to 32MB per CCD), so even if games don't use multiple cores, shouldn't they still benefit from the larger cache size?
This kind of video is the most valuable where we really learn something new!
7 месяцев назад
Problem with these infopacked videos is the fact that i get more questions to ask from them. My 5600X and 6750XT is looking a bit slow on my ultra-wide running games like Hogwarts and Cyberpunk in 3440x1440... so i am looking to upgrade and as i see it, my road forward is X3D and 7800XT. What i would like to see is a video showing what happens when you go X3D and switch-around on graphics. What would be my best way forward? Stay AMD or go nVidia? Too many questions :-)
It would be interesting to do a deep dive into this subject and trace the CPU operations to see how many times L3 cache is hit and how many times main memory during gaming. That would give you conclusive information.
Kinda shows how I/O bound CPUs are. You can go from dirt cheap sticks to premium ones and get 10% more performance, but add a few dozen mb of cache and you get huge boosts. Hopefully DDR ram is gonna die soon and we'll see something groundbreaking-ly fast. Modern CPUs need that. Also interested in Zen 4 version of these tests, those have doubled the L2 cache comapred to Zen 3.
Wild idea. Imagine getting RAM onboard the CPU and cutting out those slots entirety. I'm sure the skus would explode and that would lead to an overall increase in size and heat concentration, and of course less RGB performance, but it does make me wonder.
Question: is the gap between 3D variant and non-3D one the same on 4k/1440p Gaming? In terms of percentage. Asking these too see if is worth it getting the 3D variant for the high res gaming. In my country 7900X is cheaper than 7800X3D. So I'm considering the 12 core over the 8core if the gaming is not affected that much on high resolutions.
I would take the 7800X3D if gaming is your focus. Will scale way better in the long run with higher powered GPU's and will keep its resale value. Most new applications are GPU accelerated anyways.
tbh i always thought that increase in intel performance going up the stack was due to clocks: 10600K - 4.1 GHz (4.8 GHz Turbo) 10900K - 3.7 GHz (5.3 GHz Turbo)
I have the 5700G but I'm also running 4k resolution. I understand the 1080p testing reasons but it would be nice to see other resolutions even if it was for just 1 bar graph chart...I am fairly certain there's little to no fps difference between a 5700 and 5800x3d at 4k
These are CPU tests, that's why there's no 4k -- We're trying to see what the CPUs are capable of doing. At 4k, you're just benchmarking the GPU. Keep in mind these tests are for comparison purposes of the specific components, and not intended as a "This is the performance you can expect" sort of review content.
3:00 … 😂 … "I don't really have a good reason, I just wanted to. And I found this data interesting and so you're _going_ to find it interesting as well" (or else!) 😂
I agree with his findings. But games like Transport Fever where as you progress and build more assets, more cores will improve performance. I played the game on 4, 6, 8, 12, and 16 cores
In a span of like 4-5 years i been steadily uprading the basic used PC I bought back then. Had an FX-6100 and a 1060 3GB. Horrible but enough for me at that time to get away from Laptop gaming and over to PC gaming. Went from the FX-6100 (fried a board trying to OC it lmao ) to a FX-8300 ( black edition it was i think ), then R5 2600 with a nice all-core overclock and now a 5600x. You could really feel the jump from FX to Ryzen and then from the 2600 to 5600x. It's incredible how much more performance CPUs, even in the low and middle class, have compared to some years ago. Using the PBO + CO method for the 5600x which works flawlessly. Really nice and easy ( even though i love the normal overclocking via bios ). For my 1080p high refresh gaming it's a nice combination with a RTX 3070 as GPU. Not having watched the video, i am thinking that while it depends on the title,. more Cache might have a bigger impact than Core-Count. Next thing I could see me upgrade to, is a 5600/5800x3D. They offer an additional amazing performance jump and I dont need to change motherboard and stuff, cause legendary AM4. Too bad the 5600x3D is a microcenter only part again. It's a damn nice chip. Should get some new Ram as well. current one is not really impressive at all. xD
2:25 - Why is Steve taking credit for putting out the knowledge that cache improves performance? Admittedly I am an old person, but the performance gains were obvious even back in the 386/486 days when motherboards would offer sockets to add more cache to your system
I recently switched from a 12600k to 7800x3D and the difference in gaming is staggering - I was NOT used to seeing this much performance gain from just the CPU. Typically I expect around 20% increases when I switch, but in most games I get close to 40-50% which is insanity.
I'm curious to know if you locked the CPU Core clocks of the 5700G/5800X to be identical (and the same for the 5600G/X) just how much the cache itself makes a difference between those two. I mention this, as the cache clocks on Zen chips is tied to the CPU Core clocks. I.e. CPU Freq of 4GHz means the cache is also at 4GHz. Great content though!
I hope for future memory technologies, they work more on reducing latency rather than increasing bandwidth. That would likely have more effect on performance at this point. I really love this analysis. Great job!
The one thing that improves gaming performance for sure is more Cash.
Nope... Because there are components that are extra costly and don't provide that much performance. And in fact, cheaper components can provide as much performance as pricier ones. So it's actually more about Finding the Best Value than just simply blindly go buy the pricier things.
@@Trip4man someone didnt get the joke
@@Trip4man"uhm akshyually" lol.
It was a good joke the dude made. Sure, you're right, it's not always true that spending more equals greater perf. But on average, for most people, increasing budget will allow for better performance.... And the joke was funny! So chill lmao
You want more Cache, I want more Cash. We are not the same.
More cache: This does not spark joy.
More cash: This sparks joy.
I just upgraded to 5800x3d from my old 2700x.I see about 30% performance gain in average FPS with my 6700XT, and the stuttering, frame time spikes are all gone, I can finally enjoy fluid gaming in most of the games.
it really is amazing how much of a difference going from a 3600 to a 5800x3d made, even while using a measly RX480 and technically "gpu bound" 99% of the time. Those moments when the CPU slogs you down really ruin gaming fun
30% sounds low, maybe lower some settings if playing anything competitive. I went 38xt to 58x3d, I was CPU limited in ACC and EPIC graphics settings everywhere didn't impact max / average FPS much. Dialled back some settings that are pretty useless when playing and almost doubled FPS at race starts, average 50% higher overall.
I went from a 2700x to a 5600 on a Radeon 6700 10G and even losing two cores I still get massive performance gains.
@@fracturedlife1393 competetive games will see larger gains, Elden Ring maxed at 1440p just got smooth frame pacing now :)
@@fracturedlife1393 from cpu bound to gpu bound :)
AMD has hit a jackpot with it's 3D V-Cache technology.
Intel only did it as a one-off with the 5775C, never followed up on it.
Isn't it great that V-Cache was just a skunk works for funsies thing one of their engineers cooked up? Wasn't even really in the plans until the prototype was super impressive.
Info from someone's AMD tour... Maybe LTT? Gamers Nexus? I can't remember now
Yeah but they also hit a wall with it.....
Which is why they are putting it in EVERYTHING now.
Cuz they know that AI chip ain't shit, LMFAO!!!
@@TheDarksideFNothing That was Gamers Nexus. Interestingly, Threadripper has a similar origin story. AMD seems to have a company culture that lets their engineers experiment a bit and it's paid off for them greatly.
@@Breakfast_of_Championsi7-5775C's technology isn't even close. That is an L4 eDRAM cache attached on the side.
Not only is eDRAM much slower than SRAM, it's just a standard MCM (side-by-side) module attached through a standard bus.
The incredible magic of 3D V-Cache is that it adds practically no additional latency, because it's literally right there, where the regular L3 cache logic is.
Any company can make a giant chip with more L3 cache, but that will lead to additional latencies (bigger == further away), and cost more due to sinking yield.
I love these kinds of comparisons. Thank you for doing them.
When I saw the title I thought, "Haven't you already done this with Intel?" I wonder what would happen if you took the 64 core or 96 core Threadripper and disabled all but 8 cores. Would that give those 8 cores 384 MB of L3 cache?
peopIe should upvote this so hardwareunboxed sees it and tries it out
Depends on the layout if that's beneficial or not. Accessing cache on a different CCD induces a hefty latency penalty that would reduce performance in most instances.
@@MLWJ1993 is that latency penalty worth it over the latency penalty of having to access RAM?
@@scamdem1cGo check out the 7950x review, look at it's scores compared to the 7700x. The answer is no. It's at best equal, in some cases worse.
@@scamdem1cprobably, ram access penalty is hundreds of clocks.
As someone who plays a lot simulation type games, I am continually grateful that you included Assetto Corsa Competizone in your testing suite. For people that primarily play racing sims, flight sims, and large scale military sims, like ARMA, testing and comparing CPU cache as well as core count is integral to find out what hardware is the best choice for these kinds of titles. The way these games operate is so greatly different than most other games, mostly being console ports with not a lot of instruction sets being sent to the CPU in comparison.
It's allways worth revisiting these types of subjects if only to help newbies learn more about the machines they are buying. Also updated / expanded testing data is allways good.
Funny we forgot the lessons learned during the Core 2 Duo and Quad era. The extra cache on Penryn vs Conroe (especially 2M Conroe) mattered more than the # of cores for gaming
yep, this is why a eon with lots of chache is still relevant
Then Intel Ring Bus came on Sandy Bridge and made a huge improvement in perfomance.
cores = muscles
cache = oxygen
Spot on 👍
great analogy!
Ha classic.
Cores=allnattymuscles
Cache=steroids
VRAM = ???
More like cache=blood vessels
I actually upgraded from the 5700G to the 5800X3D last year and it'd one of the best PC components I've ever bought
@tilapiadave3234 nonsense
@@tilapiadave3234 That's why AMD is dominating the CPU market now...
@@tilapiadave3234given that in Germany you can get one for 277 Euros while the 5700X3D is 253..... nope!
@@tilapiadave3234said no one ever
yep @@LupusAries
Cache itself is important, but it depends on how accessible it is to all the CPU's cores. If you have 20mb of cache but a core can access only 1/8th of it, its far worse than if one core can access all of the cache. That's basically why zen 3 is so much faster than zen 2. A core can access twice as much cache on zen 3 compared to zen 2
This here is about L3-Cache that is shared between all cores. But yeah L1/L2-Cashes also matter but are not easily comparable, because they are usually the same for the same architecture. And when comparing between two different architectures, there are more factors than just L1/L2 that are responsible for the performance gain.
Generally speaking, only L1-I and L1-D caches are private in modern CPU architectures. L2 cache is "on the core" meaning it's physically allocated to each core, but other cores can "snoop" L2 caches of other cores. With Zen, this can only happen inside a CCD, so Core 0 (on CCD 0) cannot access the L2 cache of Core 8 (on CCD 1). This is partly why multi CCD Zen chips are not better in games. L3 cache is again CCD-public, meaning any core on the same CCD can access the L3 cache, but other CCDs cannot. As you mentioned, with Zen 2, a CCD was 4-cores and a CCX contained 2 CCDs. With Zen 3, Zen 4 and Zen 4c, a CCD is 8 cores.
A good example for Zen 2 is Ryzen 5 3600 vs Ryzen 7 3700X. The Ryzen 5 3600 has 32MB of shared L3 cache between cores and the other one does not. But the difference in performance is only 5%. But that can be attributed to more cores and a higher frequency rather than a difference in cache. People have learned that the cache makes a big difference in Zen 3, but then they try to apply it retroactively to older architectures as well. I don't think Zen 2 was designed to benefit significantly from more cache.
@@JackJohnson-br4qr the cache on zen 2 was shared between 4 cores, thus an 8 core 3700x did have 32mb of cache but each 4 core CCD could access 16mb. That changed with zen 3 where each core had access to the 32mb of cache as there is no CCD anymore
@@cpt.tombstone except that dual ccd chips are faster in games, not by much with varying results, but in general they are the same or faster than single CCD chips
Excel, browser and other productivity applications' impact on core frequency, ipc and cache will be much appreciated as most of the day to day tasks are still single threaded
Yes, I'd be very interested to see all the same CPU's tested against general apps/productivity. Especially as the much larger cache on the X3D chips usually results in lower frequency, which in theory matters more outside of games, but it would be nice to see all that confirmed.
3:10 wow, dictator Steve :D:D:D But hey, you are good dictator ! :D
3D cache also improves winRAR performance a lot, because the dictionary fits inside cache and the processor won´t go to main memory frequently.
7zip better
@@FateXO i'm excited about zstd and FSE-related compressors
@@FateXO
You’re pushing your luck little man.
@@Dankyjrthethird what you finna do about it old timer
Love the blowing up of the "More cores/multi-tasking!" argument points. Well done guys.
it's quite surprising to me, but good to know!
I was running Intel for at least 4 of my last builds and was about to go for the 14700k a week ago. Then I stumbled over some information on the lifecycle and the fact that AM5 would be more future proof for another couple of years while being superior for gaming anyway due to the cache. And then I also noticed that the R7 7800X3D also was way more efficient and cooler. All that while costing less in total together with an Aorus Master mainboard. Had to reconfigure my cart eventually and go with AMD of course. Super glad right now.
I was on intel for a decade. My last build was 5800x3d, very efficient never regret will last at least 5 more years.
hi planing also to buy 7800x3d AMD path, can you share me your build list ? thanks and have a nice day
@@johnfirst3986 I went with
- Corsair 7000D Airflow (Big Tower) white
- Aorus Master B650E (Mainboard)
- 7800X3D (of course) cooled by Noctua NH-D15
- Corsair Vengeance RGB 32GB CL30 AMD EXPO (RAM)
- Corsair RM850x (PSU)
- MSI GeForce RTX 4080 Super 16G Gaming X Slim White (GPU)
I had plenty of storage left from my older machine. And I game at 1440p for the most part. The build is overkill for what I need as I mostly play fighting games. Even the most recent games will not need the 4080 to consume more than 50 watts.
This was one of the most informative CPU videos I've ever seen. Good job making great content in a time where there isn't much happening as far as new parts.
very interesting, will be installing 5700X3D tomorrow from 3500X, my wife's PC ended up being a good upgrade path from 8700K PC.
upgraded my 3700x to 5700x3d and i'm loving it. the 3700x was great for its price back in the day, but the 5700x3d is just amazing
even from 5600 to 5700X3D i felt the game more smooth especially in CS2
I had a 5900x 12 core and changed to a 7800x3d 8 core and don't regret it for gaming now.
Did you get 5900x for gaming?
Does that mean cache are more important than cores?
I still have my 5900x and now waiting for the Zen5. Gaming in 4k and streaming at the same time
@@Mario211DE I don't think you need to upgrade
@@DungxxHen thing is tho im cpu bound even at 4k in different games already which is interesting
Cache is King.
Content is King, too!
Well I’d like both
Wu Tang said it first. Cache Rules Everything Around Me.
When it can be utilized yeah, but a brute force single core combined with low latency is more consistent in its results. That's why I'm interested to see how arrow lake pans out considering Intel is ditching hyperthreading for the sake of single core performance.
@ZaHandlesome did 😊
Just got my 5800x3d a few days ago. 🥳
Long live AM4
AM4ever!
AM4 is going to beat LGA1155 in terms of usable lifespan :)
Maybe we are Lucky And they will bring a other New cpu with x3d for am4
Core count and mhz are linked. High cache is about removing latency issues. If your cores or high mhz cpu is waiting for a chunk of ram from system ram the it dosn't matter if you have 6 or 7ghz cpu. It will be idle. A high cache helps to mitigate this issue by guessing what you might want to load in the future and keep it closer. The P4 tried to do this but their long pipeline killed the gains from the cache. The P4 had to flush the long pipelines and it took a long time.
12600k alderlake vs 14400 "raptor lake refresh"
Lock the cores and caches to the same frequency (and tdp's) and see if there's any change in architecture... because they're the same die.
I always follow the suggestion/raccomandation from Steve and Tim , NEVER let me down !!
CPU , GPU , Monitors , I based my purchases on this channel , I did NEVER regret ANY decision.
They are the best IMO.
It's been implied by testing, but L2 cache is incredibly important as well, arguably with more impact. Raptor Lake's performance improvements are almost entirely based on in a large increases in L2 cache (not all 13th gen did get L2 cache increases, basically it's 13600K+). The nVidia Lovelace architecture also saw massive gains by increasing L2 cache sizes. At a basic level, L2 cache is "easier" and less expensive to implement than stacked dies.
L2 just has less options for blowing up majorly in size. I think even the latency of V-Cache would be too great for L2 IIRC
Would be interesting if they figured a way to use all the L3 space on the die for L2 and then use V-Cache only for L3. Best of both worlds.
@@TheDarksideFNothing "L2 just has less options for blowing up majorly in size..." The "problem" is that L3 size has significantly diminishing returns in performance gains. You could probably cut the L3 (per core) in half on an X3D chip and see very similar results.
@@awebuser5914 You are probably right about diminishing returns, but I wonder if the relative ceiling of L3 cache effectiveness on Ryzens has even been reached yet. Right now it's obvious that 96MB is much better than 32MB (for games), but what if they tried 128MB, 160MB or even 192MB of V-Cache? 😁
They wouldn't just because of expense.
It may even be better to just tack another 32mb on instead of 64mb@kosmosyche
No, Steve did a comparison on that. Most of it is clock speed and RAM speed, the difference in L2 cache is like 1 frame.
I will be forever thankful to you guys for the review of the 5800X3D almost 2 years ago. If it wasn't for your benchmark with ACC i probably wouldn't have jumped on the X3D train and wouldn't have experienced the monster that this chip is. 18 months plus and counting and still feel completely blown away by the performance every time i load a game.
It's a shame no one ever tested it with the original AC (+ CSP). I only recently decided to make the jump to X3D, upgrading from the 3700X to the 5800X3D. and my goodness was I not prepared for the difference. I legitimately get up to 105% higher framerate. Yes, over twice the framerate. And that's with a 4060 Ti, so the potential gains with a higher end GPU can probably be much higher still.
X3D is such a blessing in many CPU-heavy titles.
To me bigger difference would be cache vs frequency
I am using X99 based Xeons and some games benefit so much from cache that it doubles my framerate despite running
Just look at 5700X3D vs 5800X3D reviews. They are almost the same perf while 5700X3D runs by around 400MHz-500MHz slower.
good call to be honest. we already know 12 threads is more than enough, but frequency VS cache is a whole different animal, as some games will heavily favor frequency, whereas others will heavily favor cache.
You need to test competitive games like a BF, COD, APEX, PUBG. It is for them that people upgrade their CPU\RAM in the first place.
Another dimension to check is Cache Size vs. ClockFrequency with similar corecount, as 3DVCache is usually lower clocked as standard Cache-CPUs
At this point, all desktop cpus run at outrageously high (inefficient) clock rates. Everything over 3ghz is mostly a waste of power.
Look at gpu clock rates.
@@PaulSpadeslooks over at my 6.3 ghz 14900k pc 👁️👄👁️
@@PaulSpadeswhy is that?
@@PaulSpades Performance scales with clock speed far past 3 GHz, what are you even talking about?
@@Eidolon2003 Agreed, this whole test is done on cpu's starved of ram info throughput. Zen2/3 have high latency to ram and ofcourse a cache buffer will diminish that problem somewhat. On Intel one can oc ring and cache for over 6Ghz scaling. This has been true for over 10years on Intel. Only HW likes to pretend this isnt doable on Intel, rofl. On Intel no oc guy leaves ring and cache on stock speed when oc'ing.
That's why I love you guys from down under. You're making videos to topics or questions the viewers would like to get answered. 👌👍
The other thing not mentioned is that cache can help vs clock speed. The 5800X and 5600X chips both have higher clocks and yet are showing some serious performance difference in gaming.
I use a AMD Ryzen 9 5900X with 64 MB of L3 cache along with 32 GB of DDR4 running at 3200 and i never feel it need more it runs every game and application i use with NO problems hopefully it will keep going for a few more years yet
Chiplet vs monolithic would also be interesting
Thanks for doing a video you found interesting to do and that's why ya did it ^^
Found it intereresting too, for sure =)
~a random canadian viewer
Super useful video, as always! Thanks Steve!
To be honest, after the release & reviews of the X3D cpus we kinda already knew those results.
Steve you are an absolute legend mate.... I was honestly just writing "What would be an interesting, but also almost impossible to accurate measure, is the number of cores that start to make a difference when playing games in a more real life example and not in a benchmarking environment. So, basically, what happens when you play a game on a windows pc that is not a clean installation to get accurate results for the hardware you test, but when you have youtube, discord, motherboard/peripheral background software, various tabs on a browser open, etc"...
And I see you 've already covered this on 15:00
So 6 cores are fine with just gaming and more cache would be usually better . How does it hold up when u also streaming from the same pc? Would a cpu with more cores then be better, or a cpu with less cores but more cache? Dunno if this a yes or no question or 'it depends' :P
Really impressive how big changes are in fps and how much L3 cache has changed over years!
Thanks for the benchmarks!
Multitasking requires the CPU cores to share the rest of the memory and storage subsystems and therefore increases latency and latency dependent games will suffer regardless of core count. That's why the E-cores on ADL and RPL taking care of all the background tasks is somewhat of a fallacy.
Yeah. For that argument to ever have any merit we will need quad channel memory on the consumer platform.
"That's why the E-cores on ADL and RPL taking care of all the background tasks is somewhat of a fallacy."
Better scheduling fixes software CPU overhead, that's why my 12700K is perfect for a DAW and chips without Big-Little aren't that good.
What Steve shows in bars doesn't reflect the real world when actually using these types of CPUs, same for a lot of results i watched on the internet when these techtubers are ignoring music producers.
When i disable the e-cores on the i7, i lose singlethread speed and gaming perfomance even though I'm freeing cache for the big cores. Depending on the game, it could be a hit to the frametimes (a small one with some exceptions).
When the scheduling works perfectly, we get a massive 30% increase for gaming perfomance (watch APO on the 14900K as an example).
Cache and memory aren't everything. Imagine how much better the Ryzen 7 1700 would have been at launch if the scheduling worked fine, nowadays it outperforms quad core i7s from that time due to the lack of cores.
Simply search which CPUs have the highest IPC, singlethread speeds and thread counts, the rest is irrelevant (minus number of PCIe lanes).
I appreciate your no nonsense video titles.
Can we see a cache vs clock speed video?
How much do the non 3D skus make up in performance with their clock speed increase over the 3D variants?
Only a curiosity of mine because I moved from 7600x at 5.5ghz to a 7800x3d at around 4.6 ghz.
In these charts, the 5800X has faster clocks than the 5800X3D. The 5600X has faster clocks than the 5600X3D.
In simple games, faster cores can be an advantage because those games don't fill the 'normal' cache sizes. (Anything in a modern engine like UE4 will be more complicated than this.)
Productivity workloads are often spamming a small set of instructions on a large amount of unique data. Cores and clocks help those more than cache - if the task can be spread across cores, that is.
Conversely, going from a 5000 CPU to a 7000 CPU is a change of architecture and memory goes from DDR4 to DDR5. The also doubled the L2 cache on each core. ~15% more instructions per cycle in each core and more memory bandwidth alone should give ~30% more performance for the same clock+core+L3 cache configuration.
7800X3D should boost up to 5050MHz per core. Have a read about PBO, curve optimiser and the +200MHz "Boost Clock" setting your motherboard BIOS should have. Task Manager might not report the full clock speed you are getting. HWinfo and similar tools will show it.
If your motherboard has an external clock generator, ScatterBench has an article about overclocking X3D a bit. It may even have a built-in, pre-tuned overclocked profile.
I got my 5600 luckily by chance. I saw a 5600G for sale on Facebook Marketplace and I pm'ed the seller. We met the next day and he said to me that he has two CPUs, a 5600 and a 5600G. I paid for what I bought from him and when I got home I was surprised it was a 5600, instead of a 5600G. He said he forgot which box it is but I said to him it's fine. Best decision ever.
i really want to see a 1gb cache one day lmao you could cram a whole old game in there
Hell even a modern game you could fit enough of it in there to easily mask any data swapping.
We already see where some games see no benefit because they've already optimized to fit in normal amounts of cache so at some point you lose the benefits.
But I do wonder if a dev KNEW they were getting 1GB if they'd be able to take advantage of it in really interesting ways.
Yeah the one thing I keep wondering about how one could optimize if they knew they were getting 1gb of cache. i personally wonder if it would help raytracing performance at all since that hits performance the hardest@@TheDarksideFNothing
@@OtherwiseUknownMonkey Yeah, I think a full GB of cache would be much more about seeing what new things you could do vs making existing things go faster.
Right now all things are designed around small caches because that's the hardware that exists in mainstream. But some applications miss the mark, and that's where V-Cache shines.
Intel is apparently working on a last level cache that goes up to 8GB.
@@TheDarksideFNothing i wish i could hear a tech artist talk and a programmer talk on it, with 1gb cache on cpu you could make worlds feel so much more lively eith more intricate ai routines i imagine, and if you had a gig of cache on the gpu you could keep whole lightmaps in there, like lets say the game let the gpu know where the lighting will be at in 10 seconds from now and you could smartly interpolate those lightmaps that will still be in cache the whole time making rendering sm faster
Are you going to do another video on CPU scaling with high end GPUs at all the resolutions ?
"Why am I doing this. Because I want to ". At this point, the video gets a like. Because I want to 😊
This was why i upgraded from a 3900X to a 5800x3D, despite the core deficit i mostly play games but also i wanted a single CCX 1x8 rather than 4x3 core because of the cross-talk between cores over the infinity fabric and accessing the cache. Still using an RTX2080 so while over all fps has not changed much (GPU limit) but my 1% lows has been reduced by over 50%. Also games don't like too many CPU cores as it messes up the scheduling which epic has came out about crashes on Intel CPU's because of core count.
What about L2 and L1 cache size?
The 3D V-cache absolutely makes a different in frametimes too, I went from the 5800X to the 5800X 3D and the caching stutters went away, I was getting insane shader cache stutters on the regular 5800X and they were non existent on the 3D version, this is after a full driver install and the cache cleared.
Just let 5600X3D sell worldwide already !!!
*ABSOLUTE GENIUS WORK*
As a geeky request, if you can compare the 7000 series x3d models. Please
A few certain RUclipsrs aren't going to like this one...
14:59 shots fired. Lookin at you byte size tech
Just went from a 5900x to 5800x3D to 7600x to 7800x3D in the last 3 months. I learned my lesson about higher core count for gaming. Great video.
Really fantastic video! Great information, great data presentation, and very educational for buyers. Thanks Steve!
You guys always seem to make videos answering questions I've wanted to know the answers to. Cheers!
Thank you! I'd rather add as optional opponents 4600g and 4700g. They have only zen2 cores but twice less L3 when compare to 5xxxG parts, so L3 compare chain can be wider - from 8 to 96MB :)
I followed your recommendation (5600X) two years ago and never regretted it as it runs everything very nicely, in fact I built my whole system following your various videos on hardware components. I briefly thought of upgrading to the 5800X3D when it came out but then I realized that I don't really need it for the types of games I play (mostly strategy games @ 60 fps).
Agreed. I did the same , always follow the suggestion/raccomandation from Steve and Tim , NEVER let me down !!
CPU , GPU , Monitors , I base my purchase on this channel , NEVER regret ANY decision.
They are the best IMO.
An i7-12700K with DDR5 is faster than a 5800X3D, it's similar in gaming perfomance to the 7700X while clocking 800MHz lower. It's a streaming beast as well...
If i waited a little bit more, maybe i would have a 7800X3D. The 5800X3D also has an annoying flaw, it's a jittery mess for the clocks, depending on the game i even watched a Ryzen 9 5900X being smoother for the frametime graphs than the 5800X3D for Rust.
I don't like CPUs with bad clocks or power limits. I feel the jitter as well for the Windows UI on those CPUs and other things.
IMO the clockspeed masterace will be end soon (in main use) just like Intel did back in the days with the Pentium4. To much power/heat to dissipate comparate to performance.
I belive ,the big and fast cache, will be the proxime future primary choice for the productor .@bra2867
oh wow this is a subject ive been wondering about for awhile thanks for the video
I was just wondering about this with my brother yesterday, thank you for the explanation!
Love this channel. Exponentially increases my enjoyment for pc tech and gaming.
This and the x3d comparison was super helpful!
With how a lot of CPU heavy games DON'T have multi-core support, it looks like the emphasis on cache is a better idea. While not optimized well in the first place, the X3D CPU's have been a blessing for Ryzen users who play Squad.
The "problem" with games is that each frame is a down-render of a specific point on a time line. This means that the synchronisation thread will always be the bottleneck. A tile based renderer doesn't give a rats arse about what happens if one thread finishes before another. It will just keep piling jobs on any available core until there are no more jobs. A game cares very much about exactly where each individual animation is at exactly the moment when it's time to freeze everything, down-render the geometry and dispatch it to the GPU.
I'm not saying games are the epitome of hardcore optimisation, as we get plenty of examples where a new game has less going on than some other older title, yet still hogs CPU power much worse. But I am saying that at the fundamental level there is a hard limit to how well you can make a game utilize asynchronous batch processing without the synchronisation thread becoming overwhelmed.
If anything this really highlights the impact of swapping data in and out of system ram has on overall performance.
Hey Hardware Unboxed.
I was curious about Intel 12th gen.
How does an i5-12600k with 20MB cache beat the Ryzen 5 5600X with 32MB cache in most games? There are still some other factors when talking specifically about gaming performance. I can see that cache improves performance quite a bit on AMD but how can intel cpu’s with lower cache keep up? 12:59
I remember many hears ago buying a used Opteron X2 170 for gaming. It had one more core and a boat load more cache then it's Athlon 64 equivalent, and even being 200Mhz slower it out gamed pretty much any Athlon 64 because of the added cache. I was also able to sink a 1Ghz overclock on that bad boy and REALLY crank out the performance.. good times.
Been saying this since Intel's mesh architecture (I had a 7820x), so it's great seeing more and more videos confirm cache is so important! Great video.
Very informative, thank you. There are some particularities, but but some things will never linear.
Cash is indeed king XD.. AMD really struck gold with 3d vcache.. would be interesting to interview a cpu engineer to understand more about how the increased cache helps
I LOVE the “for science” videos. Extremely educational for clearing up people’s oft-speculative, sometimes wishful musings.
Since there’s a decent bit of time ahead before the next gen, perhaps we can revisit these varying cache configurations (especially the X3D variants) with cores disabled down to 4, 2, and even 3 cores! As performance and even stability starts to drop off, how much does cache influence the result for what number of cores? Does a triple- of even dual-core X3D config run some games where a stunted L3 quad-core can’t?
Come on Steve!
Do it FOR SCIENCE!!!
Something that has to be noted is that cache architecture is far more complex than just adding more even if it's possible, the corecount, the cores, the task and the L1 and L2 caches affect how much L3 is optimal.
In general a larger cache is slower so there is an optimal value for the cache for each task. Slower cache means it takes longer to access it but it means you have to access the even slower RAM less often.
If your task has very little memory requirements (takes very little ram space) then a cpu with larger L3 can actually become slower.
Games get heavier and heavier so we might get L4 cache soon in the form of HBM on the chip and the L3 caches might get smaller.
Hey steve. Love the vid. Whould love see a vid abpit whether an ssd with dram cache makes a sense for a gaming pc. Does more expensive storage make a differnce?
Great video! But I feel like taking clock speed into consideration would be important
Hey Steve,
This was a great video, I was just looking into 5800X3D as a potential upgrade path as it does seem to regularly get discounts here in the EU.
I am kind of curious how it would compare against one of the 12/16-core SKUs though. All of these have 64MB L3 cache, but boost higher and are generally better value than X3Ds for productivity. I'd imagine many of us also use their "gaming" PCs as a workstation, so I wonder if the gaming performance gain is significant enough to justify a 5800X3D vs. 5900x for example. In your original review the average 8-game 1% lows came up to 150 vs. 121 FPS respectively, but the X3D got blown away in all productivity tests.
What I don't get is why the 64MB L3 cache on the 5900X/5950X did not seem to make any difference in these tests vs. the non-X3D 5800X (32MB cache). The way I understand it, the 64MB should be shared between both CCDs (as opposed to 32MB per CCD), so even if games don't use multiple cores, shouldn't they still benefit from the larger cache size?
This kind of video is the most valuable where we really learn something new!
Problem with these infopacked videos is the fact that i get more questions to ask from them. My 5600X and 6750XT is looking a bit slow on my ultra-wide running games like Hogwarts and Cyberpunk in 3440x1440... so i am looking to upgrade and as i see it, my road forward is X3D and 7800XT. What i would like to see is a video showing what happens when you go X3D and switch-around on graphics. What would be my best way forward? Stay AMD or go nVidia? Too many questions :-)
I like, that you "just wanted to" do this test. Also it is still relevant and interesting.
It would be interesting to do a deep dive into this subject and trace the CPU operations to see how many times L3 cache is hit and how many times main memory during gaming. That would give you conclusive information.
_Ca$h really is king._
Kinda shows how I/O bound CPUs are. You can go from dirt cheap sticks to premium ones and get 10% more performance, but add a few dozen mb of cache and you get huge boosts. Hopefully DDR ram is gonna die soon and we'll see something groundbreaking-ly fast. Modern CPUs need that.
Also interested in Zen 4 version of these tests, those have doubled the L2 cache comapred to Zen 3.
Wild idea.
Imagine getting RAM onboard the CPU and cutting out those slots entirety.
I'm sure the skus would explode and that would lead to an overall increase in size and heat concentration, and of course less RGB performance, but it does make me wonder.
Years ago, when I had 2 work PC and 1 gaming PC, your argument about cores was king.
Question: is the gap between 3D variant and non-3D one the same on 4k/1440p Gaming? In terms of percentage. Asking these too see if is worth it getting the 3D variant for the high res gaming. In my country 7900X is cheaper than 7800X3D. So I'm considering the 12 core over the 8core if the gaming is not affected that much on high resolutions.
I would take the 7800X3D if gaming is your focus. Will scale way better in the long run with higher powered GPU's and will keep its resale value. Most new applications are GPU accelerated anyways.
Great video, hope AMD and Intel will use this to make more cache for gaming CPU, because it is interesting how far is scale.
tbh i always thought that increase in intel performance going up the stack was due to clocks:
10600K - 4.1 GHz (4.8 GHz Turbo)
10900K - 3.7 GHz (5.3 GHz Turbo)
6:10 the point of the video. Thank you for making it.
I have the 5700G but I'm also running 4k resolution. I understand the 1080p testing reasons but it would be nice to see other resolutions even if it was for just 1 bar graph chart...I am fairly certain there's little to no fps difference between a 5700 and 5800x3d at 4k
Depends on your graphics card of course. The 4090 will be CPU-limited in 4k with a 5700, not sure how much though.
@@Syssn3ck true i'm using an RX6800 which plays my games at 4k 60fps no issue. I have no interest in owning a 4090 for 1080p either...
These are CPU tests, that's why there's no 4k -- We're trying to see what the CPUs are capable of doing. At 4k, you're just benchmarking the GPU. Keep in mind these tests are for comparison purposes of the specific components, and not intended as a "This is the performance you can expect" sort of review content.
Memory talks, memory talks, Dirty cache I want you, dirty cache I need you oh.
Ah Stevie V. Taking me back to my late teens
I would appreciate a follow-up to this video covering emulation performance, RPCS3 and Yuzu specifically.
3:00 … 😂 … "I don't really have a good reason, I just wanted to. And I found this data interesting and so you're _going_ to find it interesting as well" (or else!) 😂
🤕
Welcome back to hardware on box indeed.
Shocked you didn't test The Riftbreaker. Talk about a game that loves v-cache....
I agree with his findings. But games like Transport Fever where as you progress and build more assets, more cores will improve performance. I played the game on 4, 6, 8, 12, and 16 cores
In a span of like 4-5 years i been steadily uprading the basic used PC I bought back then.
Had an FX-6100 and a 1060 3GB. Horrible but enough for me at that time to get away from Laptop gaming and over to PC gaming.
Went from the FX-6100 (fried a board trying to OC it lmao ) to a FX-8300 ( black edition it was i think ), then R5 2600 with a nice all-core overclock and now a 5600x. You could really feel the jump from FX to Ryzen and then from the 2600 to 5600x. It's incredible how much more performance CPUs, even in the low and middle class, have compared to some years ago.
Using the PBO + CO method for the 5600x which works flawlessly. Really nice and easy ( even though i love the normal overclocking via bios ).
For my 1080p high refresh gaming it's a nice combination with a RTX 3070 as GPU. Not having watched the video, i am thinking that while it depends on the title,. more Cache might have a bigger impact than Core-Count.
Next thing I could see me upgrade to, is a 5600/5800x3D. They offer an additional amazing performance jump and I dont need to change motherboard and stuff, cause legendary AM4.
Too bad the 5600x3D is a microcenter only part again. It's a damn nice chip. Should get some new Ram as well. current one is not really impressive at all. xD
2:25 - Why is Steve taking credit for putting out the knowledge that cache improves performance? Admittedly I am an old person, but the performance gains were obvious even back in the 386/486 days when motherboards would offer sockets to add more cache to your system
Can you guys test the difference in frame timings between X3D cpus and intel cpus?
Just curious.
I really like this episode. Can you guys do another episode comparing l3 caches versus speed of ram?
I recently switched from a 12600k to 7800x3D and the difference in gaming is staggering - I was NOT used to seeing this much performance gain from just the CPU. Typically I expect around 20% increases when I switch, but in most games I get close to 40-50% which is insanity.
Especially given how the 12600 is hardly even old and was extremely competitive at launch
This is especially helpful for low-end buyers, whose options right now are a R3 4500 and R5 3600. The major difference is cache.
Really great idea for a video
Very helpful definitely 🙏🏻
Love the content like always
I'm curious to know if you locked the CPU Core clocks of the 5700G/5800X to be identical (and the same for the 5600G/X) just how much the cache itself makes a difference between those two. I mention this, as the cache clocks on Zen chips is tied to the CPU Core clocks.
I.e. CPU Freq of 4GHz means the cache is also at 4GHz.
Great content though!
I hope for future memory technologies, they work more on reducing latency rather than increasing bandwidth. That would likely have more effect on performance at this point.
I really love this analysis. Great job!
"Cache really is king"
The delivery on that was flawless