Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[biostud]

I don't even get this continued talk of niche. As if only a few 1000 people own or play these titles. MSFS has over 10 million players total. That means there are millions playing on PC. Hardly a small niche title. Next, the goalpost gets moved to "It's hard.". Professionals being paid to test games, shouldn't include games that are hard to test? LOL Nonsense. If they can run around Hogsmeade, then they can save a file flying around NY City.

Factorio disappeared from testing suites. I was seeing some reviewers featuring that one before. Odd how both favor 3D cache and are no longer tested. I'd be very interested to see how Zen 4 3D does in both. Hopefully at least one reviewer will include these titles again.

Totally agree, probably also a lot more people play these games than 3Dmark.

 

[biostud]

Mate, that's as off topic as it gets, but if you want your wife happy, put a yearly reminder on your calendar, preferably a day before, to buy at least a flower

Nah, she’s cool. When we got married it was just us and our parents going to the town hall, afterwards having lunch at a café, picking up the cake from the baker and kids from kindergarten on our way home. But yes, I’ll stop derailing the thread.

 

[Kaluan]

I personally think that reviewers keep testing CS:GO to see which CPU hits the 1,000 fps barrier at 1080p.

Reminds me of back when outlets still benched Quake III Arena (@1024x768) for years. It got well over 1.000 fps with 9.0c and 64bit era GPUs and CPUs.

 

[Timmah!]

Do we know the exact time of day for the embargo lift tomorrow?

 

[Timorous]

It isn't a claim, it's just a fact. Stuff like CS:GO gets featured because more people play it than other games. I'm sure someone playing 4x games would like to see their titles featured as well, but

a). it isn't quick and easy to benchmark that stuff in a relevant fashion and
b). reviewers only reach a small audience by featuring those titles.

Anyone interested in knowing how well Stellaris runs on a 7950X3d will just have to buy the hardware and do a comparison on their own. Ideally published for the benefit of their community on a forum somewhere.

All paradox grand strategy games use the same engine so testing one is enough to get picture. Further there are more concurrent players of those games by far than stuff like cp2077, tomb raider, hitman 3, watchdogs legion, borderlands 3 and other frequently used titles.

The fact is the audience for these titles is far higher than almost any other single player game. Sure hogwarts right now might have more players but give it a month and it will drop down the list but Civ 6, Stellaris, football manager etc will still be in the top 100 and maybe the new AAA hotness will be around.

EDIT To add figures.

Civ 6 current is 61,995
Civ 5 current is 23,681
FM 2023 is 69,320
FM 2022 is 17,025
Stellaris, EU4, HOI4 and CK3 combined are ~ 99,000 - all use the same engine.
Path of Exile is 22,982
Cities Skylines is 21,479

Borderlands 3 is 14,997
Hitman 3 is not on the list
Tomb Raider is not on the list
No Assasins creed game is on the list
CP2077 is 16,955
Elden Ring is 60,775
Hogwarts has 317,185 now but if that joins the benchmark suite in a few months time it will be like CP2077 or maybe Elden Ring.

Bottom line is that while a new AAA game will have a burst of players the in between release games like Civ, Stellaris, Factorio etc are actually really popular and they have legs. Civ 6 is nearly 7 years old, Stellaris is approaching its 7th birthday. the only AAA games that are that old and still in the top 100 that are not multiplayer 1st are Skyrim, Fallout 4 and The Witcher 3.

Other than that everything else is either much newer still or has a huge online following like GTA V which still gets benchmarked quite often and rightly so, same with RDR2.

Stuff like Tomb Raider, Assasins Creed, Hitman 3, Borderlands 3 etc though should all drop off the small benchmark suites for the main review and be replaced by other more popular and more relevant games when it comes to benching whatever component is for test. For GPUs it would be newer AAA stuff and for CPUs it would probably be titles with non FPS primary performance metrics.

EDIT2: So having new AAA titles come along and have their own performance articles does make perfect sense. OTOH adding them to a benchmark suite not so much unless they have reasonable legs a few months down the line because it becomes a worthless title pretty quickly. If you are doing a 50 title benchmark run though then obviously you can add them in there but for the 8-10 select titles in a review, I really don't see the point unless maybe it is the 1st title to use a new engine or API.

 

[Timmah!]

I wish someone would test Sins of a Solar Empire II, thats really only game that would IMO be good candidate for benefiting strongly from the v-cache, that i would care about. But let me hazard a guess, nobody will. And surprisingly, devs on discord/official boards werent quite outgoing about this either, when asked about the performance in relevance to 5800x3D. The response i got from Frogboy was: "i dont know, i am playing it on 3990x and it runs just fine" :-D Extremely useful piece of info.

 

[Det0x]

I personally think that reviewers keep testing CS:GO to see which CPU hits the 1,000 fps barrier at 1080p.

Have done that long time ago

1080p high = 1004 fps

Spoiler: settings

FHD 1920x1080
Globalbe Schattenqualität Hoch
Modell-/Texturdetails Hoch
Effektdetails Hoch
Shader-Detail Sehr Hoch
Multicore-Rendering Aktiviert
Multisampling-Antialiasing-Modus 4x MSAA
FXAA-Antialiasing Deaktiviert
Texturfiltermodus Anisotropisch 16x
VSync Deaktiviert
Bewegungsunschärfe Deaktiviert

1080p low = 1111fps

 

[DrMrLordX]

If you're already testing 50 games, 20 of which all the other sites also tests, then it would be obvious to attract the "niche" crowd by change a couple of games to differentiate yourself.

Yeah someone could specialize and get an audience. Not everyone sees every potential market.

MSFS has over 10 million players total.

I was talking 4x strategy. I think most reviewers gave up on Microsoft Flight Simulator but I'm not sure why . . .

All paradox grand strategy games use the same engine so testing one is enough to get picture.

Fair enough, but:

EDIT To add figures.

Civ 6 current is 61,995
Civ 5 current is 23,681
FM 2023 is 69,320
FM 2022 is 17,025
Stellaris, EU4, HOI4 and CK3 combined are ~ 99,000 - all use the same engine.
Path of Exile is 22,982
Cities Skylines is 21,479

Counter-Strike 2 - Steam Charts

An ongoing analysis of Steam's player numbers, seeing what's been played the most.

steamcharts.com

CS:GO alone has all of those beat.

I think in the case of other bits of software (like Cyberpunk 2077 or Hogwarts Legacy) they buy a hot license and then sit on it for awhile.

 

[lopri]

This is what happens when you pair Zen 4 with slow ram.

I watched that video and I do not think the data backs up what he says. There just is not a clear trend among the data points he collected, maybe except for DDR5-5200 and lower. But even before that, he does not tell you where the CPU frequencies were when those memory sticks were tested. Only way to see memory scaling is by fixing CPU frequency at X GHz. If he simply tested different sticks with "default" out-of-the-box mobo setting or default PBO/Echo mode, that means the CPU frequencies was all over the places, and run-to-run variances on top of the already trivial difference memory sticks make his data basically a garbage.

Until AMD reveals how data exchanges hands between IO die and Infinity Fabric, this type of test is a massive waste of time.

 

[Schmide]

I watched that video and I do not think the data backs up what he says. There just is not a clear trend among the data points he collected, maybe except for DDR5-5200 and lower. But even before that, he does not tell you where the CPU frequencies were when those memory sticks were tested. Only way to see memory scaling is by fixing CPU frequency at X GHz. If he simply tested different sticks with "default" out-of-the-box mobo setting or default PBO/Echo mode, that means the CPU frequencies was all over the places, and run-to-run variances on top of the already trivial difference memory sticks make his data basically a garbage.

Until AMD reveals how data exchanges hands between IO die and Infinity Fabric, this type of test is a massive waste of time.

It's a multi core OOP that runs to a power target. Artificially fixing the frequency would govern performance. The variance is the memory, everything else should be a black box.

 

[Timorous]

CS:GO alone has all of those beat.

I think in the case of other bits of software (like Cyberpunk 2077 or Hogwarts Legacy) they buy a hot license and then sit on it for awhile.

As does DOTA 2 and other free to play multiplayer titles.

In a 10 game suite you could test CS:GO and COD as your 1st person MP games, One of Total War, Riftbreaker, AoE 4 as your scale RTS game then there is room for 4 of the latest / currently popular AAA games and room for 3 games where FPS is not the primary metric, probably Civ 6 turn time, HoI 4 sim speed and then one other, Factorio maybe because that has a built in benchmark.

That suite would be balanced across genres, popularity, and game engines.

 

[Det0x]

Will be interesing to see how much overhead there is with single CCD vs dual CCD (easy to test with a 7950x3d)

AMD chipset drivers feature special game optimizations for Ryzen 9 7000X3D CPUs - VideoCardz.com

AMD Ryzen 9 7900X3D Game Optimizations AMD discloses new features in the new chipset drivers. AMD Ryzen 7000X3D processors equipped with 3D V-Cache will automatically optimize performance depending on a workload. The newest chipset driver will be able to dynamically allocate which core is to be...

videocardz.com

 

[PJVol]

In other words, if the workload is cache-sensitive, the operating system will be modified to prioritize cores with 3D V-Cache

What the ... is that?

 

[leoneazzurro]

What the ... is that?

Probably there are some words missing, i.e. it should be "Standard Operating System thread allocation" or something like this.

 

[biostud]

I have a couple friends that only play/use MSFS and even built their systems purposefully for that program.
The concept of average is a mathematical construct that often has no real world reality. 2.2 children/family? How many earns the average salary?

Exactly. If a processor performs well in large variety of games you will have to assume it is more likely to perform well in the games you play, but there is no guarantee. Where my old computer has struggled the most is Company of Heroes 2, with 500 pop cap co-op vs 4 computers. After 30 minutes gameplay, I'm so far behind in game time that it is unplayable. I'm very likely to play a lot of Company of Heroes 3, so the CPU performance in that game would be important to me, and 95% of the games in the review suites I'm never going to play.

 

[PJVol]

Probably there are some words missing, i.e. it should be "Standard Operating System thread allocation" or something like this.

Yeah, or something like "notifed" instead of "modifed"

 

[Abwx]

7950X3D review at computerbase :

Ryzen 9 7950X3D im Test gegen 5800X3D und 13900K(S)

Zen 4 3D ist da. Im Test erweist sich AMDs Ryzen 9 7950X3D gegen 7950X, 5800X3D und Core i9-13900K(S) als schnell und effizient.

www.computerbase.de

 

[PJVol]

7950X3D review at computerbase :

Learning new terms - frequency chiplet, 3D chiplet.
Aaand...

Ryzen 7000, whether with 3D cache or without, were driven with DDR5-5200 32-32-32-61, Intel Core 13000 on the other hand with DDR5-5600 32-32-32-61

 

[biostud]

Learning new terms - frequency chiplet, 3D chiplet.
Aaand...

So faster on average than the 13900KS, with less than optimal memory for both systems.

 

[PJVol]

So faster on average than the 13900KS

Yeah, kinda convincingly )
Though TPU found some scheduling issues

 

[coercitiv]

So faster on average than the 13900KS, with less than optimal memory for both systems.

It's faster with fast memory as well, you can check the Zen4 3D review thread for more results.

 

[moinmoin]

Phoronix with some tests:

AMD Ryzen 9 7950X3D Linux Performance Review - Phoronix

www.phoronix.com

Right away a test result showing that the BIOS toggle for cache vs. frequency by itself has no effect but needs a driver counterpart in the OS to work (nowhere to be seen in current Linux 6.2 Git).

 

[PJVol]

This is funny )

 

[PJVol]

"XBox Game Bar" and "Balanced power plan" are prerequisites for the proper CCD scheduling?
GN review

 

[moinmoin]

"XBox Game Bar" and "Balanced power plan" are prerequisites for the proper CCD scheduling?
GN review

Sounds like AMD essentially reuses some tools available within Windows instead implementing the desired behaviour in the driver on their own. Not a bad thing, per se.