Thanks for mentioning Stellaris. I know the AAA games are more popular, but I play a lot of CPU limited games that no one seems to mention. Awesome review.
Factorio also sees massive gains. I just don’t like how the obscurity of the title affects when or if AMD will whitelist it for the V-cache treatment. For instance, HUB found Factorio operates at half the performance until HUB disabled the right CCD in BIOS. If I have to babysit this CPU, then I have to judge it based on what happens if I don’t babysit it, because I won’t.
I play a lot of stellaris and football manager. Football manager has some guys on their forums collecting performance data and put it into a spreadsheet. Unfortunately links don’t work on RU-vid, but if you google for “FM 23 performance benchmarking” there’s a link to a thread on the SI forums with the testing methodology as well as the results. Right now it seems like the X3D doesn’t help that game at all, as they show similar benchmarks to the base 5800x. But the 7950x holds the title, being significantly faster than the 13900k. Always interesting looking at games that entirely CPU dependent. And stellaris still needs a built in benchmark, it’s just so hard to compare systems with it.
@@djsnowpdx honestly they’ll figure out which side is better over time with chipset drivers it’s not like they’re optimizing code paths, just changing the scheduling
I would love to see 8-core X-3D cpu with powerful igpu, like Steam Deck CPU in steroids for super small SFF builds with nice 1080p competitive gaming abilities.
That opening laugh is promising I wonder what pleasant madness we'll be led down to After watching the video edit: I'd love to see if the 7900x3d has a 6 core or 8 core v-cache complex And ofc how much of a 1% lows impact you get in games between these two top dog CPUs
@@MrStoffzor Kinda interested in games that are more single core dependent while prefers huge cache compared to high clock speeds.. Since the 6core vcache is going to have even more cache per core than the 7950x3d which has to share the cache between 8 cores, it might even beat that. lol
@@MrStoffzor i mean think of it like this, intel has big cores and little cores, amd has cache cores and no cache cores or you could have all no cache, you get lots of options i just really wish we get zen4c 32/24 cores variations for the am5 socket (wish amd allowed more ram too and quad channel, make it the X670XE chipset or something, like a threadripper sucessor but the am5 socket, surely amd will get bad yield zen4c, slap in some v cache and 4ghz charge 1000-2000 for 32 cores, people would be happy lol)
Please test more grand strategy games like Stellaris. Every mainstream channel test stuff like cyberpunk and completely disregard ACTUAL cache limited games. Your 5800x3d "obscure games" video had a good selection.
I got the 5800X3D, not on fire sale, but definitely much cheaper than when it launched and I paired it with a 6950 XT so I should be set for many of years🙂
I do CAD work in AutoCAD, Fusion 360 and Revit. With a little photo editing in Affinity Photo. I used to use Photoshop but their pricing model is stupid for what little I need. I am on a Ryzen 9 3950X and it works pretty well other than AutoCAD which is my main money maker! If I could bump up my speed in AutoCAD it would be worth moving to the 7950X3D, but I am not seeing anything dedicated to my needs tested. I guess I am just not a use case anyone cares about. Next year I will be spending $4k to $5k on a new system and have started learning what I should go with. Threadripper would be an option if it is in that price range. I have used AutoCAD on XEON, i7, Ryzen 9 and a couple other systems and they all run about the same. Also ran on GTX, RTX, Quadro, cards and see little difference in speed. It does seem to do better if you have a lot of RAM and VRAM overhead, but doesn't seem to care about speed. What am I missing here? Is it just that AutoCAD sucks? Probably just AutoCAD sucks. Unfortunately it is the best option for my specific use case.
I think you might be the only reviewer who is actively using Optane, so for me at least, I think it would be very interesting to see how newer games benchmark with the P5800X vs. a top-end NVMe SSD, what with DirectStorage and Unreal Engine 5 entering the picture... and to what degree the X3D vs. non-X3D parts relate to that. I'd also like to see a focus on games that might really benefit from X3D, such as Stellaris (you mentioned it, but then didn't compare to other CPUs), Civ 6 turn times, Factorio, and even something like Microsoft FS2020 which is constantly pulling in world data.
@@snowwsquire I hope so too, but I doubt they will. There will be some similar tech that'll become available eventually, but for now, most of us are stuck with "fast" NAND Flash.
A small correction: Cinebench does not fit in cache, but it is not latency sensitive, since it has a smart prefetcher for all the work that threads do. And this is precisely why Intel was right about it not being a real world performance benchmark. It is too optimized unlike other software. Software that is optimized to the level of synthetic benchmarks are synthetic benchmarks.
I use a 5950X at home for gaming, heavy web browsing, Excel, Google Sheets, and occasional highly multi-threaded applications like fast downloads and building test chambers in Portal 2. I want my 16 cores, but I certainly do not want to babysit my CPU - change settings as I change tasks. I play Factorio, Cities: Skylines, and other simulators and am occasionally CPU-bottlenecked at sub-par performance levels, particularly in Cities. I was shocked at how well the Raptor Lake i9 keeps up with the 7950X3D outside of games like Factorio and Stellaris, but I’m also a big advocate for low-power gaming, running my 5950X in eco mode, and my room has very little ability to soak heat from my computer, so the i9 holds no attraction for me. Guess I’ll skip this gen, and hope AMD offers a 12- or 16-core Zen 5 CPU with V-Cache you don’t have to babysit.
Maybe I missed it mentioned, but PCIE lanes are often a big deal for workstation candidates. AM5 has 24x pcie5 lanes plus chipset ersatz lanes. Intel have 16x CPU PCIE% lanes plus chipset lanes. I am not surprised AM5 mobos are dearer.
I would like to see how this performes in productivitiy apps, especially audio apps like DAWs. I read some reports about the 5800x3d that the cache is also a game change in audio performance regarding how many tracks you could run. 7950x3d should be much better.
@@joemarais7683no they are not, not even close. DDR5 has some rudimentary error correction cause at those speeds it will produce errors but that is NOT the same as ECC memory
I prefer the 7950x. A generally good processor is better than one designed for video game frames. I always convert my old pcs into servers when I retire them. Much better for that.
I've been thoroughly surprised with Zen. My first Zen experience, personally, was an R3 2300u Laptop with Vega 6 graphics and I could not believe how smooth Windows 10 ran on it. I have been disappointed though with Microsoft regarding this APU. It has TPM 2.0 and Secure Boot. It's a Quad Core processor, It has 16GB of memory and runs both a 256GB M.2 NVMe SSD and a 500GB SATA SSD, yet Windows 11 keeps telling me it's not supported. It literally makes ZERO sense. But gripes aside, I could not believe how smooth and snappy the OS was. In comparison to Bulldozer this chip stomps all over it and it's just 4 cores. My 4.7 Ghz FX 8350 can't even beat it in multi core! And it's a mobile part. Seeing it pef at 3.9 Ghz all core is fun too. It's nice to know it has some chops and it has pretty good single core (Broadwell) so emulators and anything else with a single thread affinity runs well on it too. I was very afraid to try out Ryzen. I didn't care what the reviews said. I'm disappointed in myself for not giving it a try sooner than Zen+.
Vendel, can you please do a video on gaming + streaming? 'cos from what others have said, Windows "parks" cpu cores when gaming. If that means, if effectively kills those cores and the stream encode would run on the 3d cores, that would be bad, big time
The default Task Manager can manually assign which CCD to certain apps or games so you can always know what app running on what cores. But if you want a more powerful app to assign 3DV-Cache to make it only run the games, and then Screen recording or Livestream on the other CCD, then use the official Microsoft Sysinternals Process Explorer.
lots of people have been parroting claims about the parking being a complete disabling of the ccd; it's not. it's still available for other applications like background or streaming/capture.
Gotta admit I'm pretty happy that I decided up build a new computer this gen. Went from a 8700k and 1070 to 7800x3d and 7900xtx. On the CPU side im ecstatic, the 7800x3d is incredible for gaming. The 7900xtx is no slouch either but it didnt have the generational gains loke the 7800x3d had. Not to mention the 7800x3d is pretty much the top cpu for gaming despite costing 250ish less than the flagships.
I would like to see physics simulation and virtual machine workloads.. more of a compare against epyc chips with 16 cores on memory bandwidth sensitive workloads.
Wendel is there any chance you can do testing to see how the performance changes if you don't rely on the CPU-Xbox game bar connection to park the non-3D CCD, and instead utilize something like Process Lasso to manually direct games to the 3D CCD without shutting down the second CCD? I really genuinely don't like the idea of half of the CPU shutting down when a game is detected, I typically have tons of background tasks going on even when gaming, and I'm trying to figure out how this will play out. So far no reviews have gone into the details on different situations and solutions. I'm primarily interested in how game performance might look if you restrict it's affinity to the first 8 cores, but still have the second CCD online with other processes. Love the channel you guys do great work.
@@jetrpg22 absolutely, the results in games that benefit from extra cache is fantastic, but if the CPU is only a 7800X3D when gaming, I would rather just buy a 7800X3D and keep my 5950X as a second system.
Do you think the upcoming threadripper CPU's with 3D V-cache will have 3D stacked cache on all CCD's or only on selected ones ? How did AMD do it with Milan-X Epyc CPU's & 3D V-cache, do all of the CCDs have 3D cache ? Really enjoyed the review, thanks a lot !
the Milan-X and upcoming Genoa-X have cache on all its cores. On server stuff it's different because there are actually workloads that benefit from extra cache. Those CPUs are meant ONLY for those workloads and nothing else.
19:00 The drivers usually don't work on Windows Server 2022 always gotta manually install everything I wonder if this special one may just make it even less compatible for us that refuse to use 10/11/desktop distros of windows..
It would make more sense if the 7900x3d was a 8core with 3dvcache & high frequency 4core. The four core would have very quick performance for streaming & internet use. While the 3d ccx would be incredible for games. It would also explain why no four core CPUs have released if AMD is binning for four core ccx.
Actually I have to think for gaming, the optimal would be the main game thread runs on the CCD with the Vcache and other threads are thrown wherever though I have to think there could be some fine optimizations, and this would vary. If ALL game threads need access to the same data pool then yes they would run faster on the one CCD, but if not, since the other CCD clocks faster, it would finish its work and pass whatever on to the main game thread or another main thread that's running. I have to think over time this will become a pretty interesting thing to see with these really high core count CPUs, which for desktop I image in the next couple generations the thread count is going to start getting up to 48, and then 64. I have to think that these parts once again give reason for people to buy into AM5. These are the parts many people wanted, either creators that wanted an AMD lower power system that also games at top levels, or people who just game who want the 7800X3D. I'm waiting for Zen 6 personally because the 5800X3D is great at gaming and will be for many years and paired with X570 it's a nice system.
Did anything change with regards to the memory controller? Can you actually get 128gb now without absolutely demolishing the frequency? Or is it the same controller as in the 7950X?
@@KrumpetKruncher yes but that's not what he asked. He asked if this CPU has better memory controller, not if there are firmware updates that make all 7000 better. An Agesa update would work the same on all 7000 because they use the same IO die
So far the parking works in the way that it detects a game (or similar profiled app) it turns off the none vcache cores. they are off you dont get to use them for other apps or anything. the 7950x3d is doa. wait for 7900x3d at least, maybe even a 7800x3d, there is a world in which they preforms as well or even better if the hz are normalized.
I feel like this cpu would be super interesting to use as a small vm server (so using a mb with 10gb eth), 2c:host, 1c:opnsense/firewall of ur choice, 1c:pihole, 4c: linux dev system and the ccx with the cache as the windows dev/gaming vm, but obv thats not everyone's use case
I'm curious about the result of ignoring the optimization that the chipset drivers does and manually pinning game threads to the 3d chiplet with Process Lasso.
Personally, I've been picking up 11th Gen Intel stuff. I've seen some pretty steep discounts on that stuff lately. It might have been a terrible value at launch, but with some of the discounts and combos I've seen in the last few months, it's hard for me to pass up.
I guess it makes sense that the X3D is more efficient since only one of the 2 CCDs is being thrashed with power and the other one is operating in a much better position on the efficiency performance curve
@Menti Capti The non X3D chips have an additional cache die that forces the clockspeeds down? We're talking about factory behaviour here, not behaviour after playing with curve optimiser.
@Menti Capti I thought you were just misunderstanding me but it turns out you're just factually wrong. The non X3D parts do not have 3D cache on any CCD. They use a 2D chiplet design - there is an interposer but there's only a single layer of active silicon on top of that. This is distinct from 3D Vcache where an additional die containing cache is attached on top of the CCD, producing multiple layers of active silicon. This is a really important distinction here because the 3D Vcache is in between the other active silicon and the cooler. All active silicon on non X3D processors is *directly* attached to the heat spreader, whereas the 3D cache equipped CCD on the X3D parts has to push its heat through an additional layer of silicon and another thermal interface. The result of this is that the Vcache equipped CCD is *hard limited* to a lower clockspeed because its cooling is fundamentally limited. The bonus side effect of this is that it's actually more efficient at that clockspeed. This is in opposition to the non X3D parts, which as proven here are less efficient, and a big part of that is because both CCDs, having more direct cooling, are being more aggressively clocked and therefore run less efficiently (but marginally faster).
@Menti Capti Oh, and if you do consider the interposer to make the chip "3D", then I'm not sure why you're singling out the dual CCD chips, since the single CCD chips are still chiplet based designs with a separate IO die. The 7800x is literally a 7950x with one less CCD, it still has the interposer, the IO die and the CCD has exactly the same cache configuration.
@Menti Capti Who said anything about 3D cache? I did. In the first comment in the thread you're replying to. That's the entire discussion - how 3D cache affects efficiency in all core workloads, not in gaming, not in chips without 3D cache. You can't park a CCD if all 16 cores are working. Also, that behaviour isn't even limited to AMD processors, let alone dual CCD chips - Intel and AMD chips can independently clock cores, which means that on a 4 core workload you can park up to 12 cores instead of a mere 8. And you have the gall to call me dense.
My big problem with the current Intel platform is that we all know that the 13900k or KS has zero upgrade potential. Intel historically changes sockets very frequently. AMD, on the other hand, has historically not changed sockets so frequently. Therefore, they have a much better upgrade path than Intel. So x670 will be relevant for several years longer than the Intel alternative.
Cache makes the world go round. I would love some thorough IGP testing after seeing the results that PCMag got with just four titles all increasing the performance by incredible 3-4x.
that got debunked. The iGPU is in the I/O die, not in the CCDs, so it doesn't really benefit from the cache on the CCDs (it doesn't use that cache at all). They had some kind of error in their testing where the normal 7950X had 3-4 lower iGPU performance than usual (so the 7950X3D iGPU was performing normally, hence the big difference in results)
Thanks for comparison. and a bit of test with Stellaris end game scenario performance. Is that applies with more and also older games that are 'modable' or have a lot of content (like end game scenario with Stellaris) like SimCity4 , Cities Skylines , Sims4 and other that are hitting few CPU core very hard, but not uses good enough parallel computing and use all available threads. There would be amazing to found somebody that can give some comparison not only for modern titles, but also for some good old games and for the modders...
It'd be cool if you could incorporate the DAWbench 2021 and DAWbench VI benchmark suites in your testing. Would really ingratiate you with the DAW crowd.
You kind of mentioned that the bios, drivers, and game bar do more than just parking the "Fast CCD" as other reviews have said. Have done any looking at what happens with a large background workload while gaming? I'd like to know, since I do have workloads that run for days sometimes, and also want to game while they are going. I know there will be clearly a performance hit to the game. But if the fast cores genuinely get parked even with a large background workload, that would be the worst case. I'd have to disable auto features and set things manually.
From looking at all these reviews, it seems like when "it detects a game running" it forces the other CCD to "park" meaning forces all the threads running (I think it includes background apps) to one CCD. If that's the case, then this is worse than the Intel situation: you have 8 P cores dedicated to the active game but also another up to 16 E cores that games can also use, but also use for background processing to smooth out gameplay.
@@jetrpg22 they are not Parked 100% ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-qq13-xxAVDE.html&ab_channel=PCWorld @ look at 8:18 you will see that windows start to use the parked cors
@@mohamedalmuhairi324 Yeah but thats not showing background app offloading , but either leakage or maybe a universal system use (IE windows offloading, as Microsoft can be sure of it). For all performance purposes, they are parked/not in use.
As a 3d artist looking to bring in more Unreal Engine real time renders into my workflow this is great new but I would LOVE to be able to use the newly announce Micron 48x2 Gig Dimms totalling 96 gig of memory with this CPU, and Expo timing were promised but could Level1Techs do a video on them and what is going on with them? Are there gotchas there? Will they run on system like this with set timings at promised speed? The memory controler is what worries me. Upgrading finally from a threadripper 1950x and don't want to make the wrong choice. I am favoring the AM5 over the new TR or Intel's new WS for now due to the single core speeds and the ability to get somewhat parity to the old system that is showing it's age.
untested opinion here, i think it should work because their memory controller struggles more with multiple sticks, but not high capacity afaik. i’m running 2x32 on a 7950x
I'm curious as the performance of the 7950x3d in games with the non-3d cache CCD being shut off in the bios but leaving the rest of the CPU alone. Does it change or enhance the performance in gaming? Is it a wash? Would you be better waiting for the 7800x3d at that time?
7950X3D seems great for people like me who want to play games and do heavy multitasking. I'm on a 7950x so I don't think I'll be swapping it out. But I do wonder if X3D is a valid way to also run 128GB of ram due to it being less sensitive to memory bandwidth?