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!

 

[Hitman928]

It absolutely is unheard of for a 13600k to use 100W difference based on the motherboard. Nor would that explain why the gap doesn't seem consistent across tests.

I didn't say the motherboard caused the CPU to use 100W more, I said the choice of motherboard lead to 40 - 70 W system power usage. It all depends on features, power delivery, default bios settings, etc. Comparing between completely different platforms, who knows how much the difference could be.

Quantity, not quality.

This is a different argument than calling them lazy. Also, you need both and having quantity doesn't mean you give up quality. Having a perfectly accurate and reproducible single test is not of much use in a complex system.

Yes, it absolutely is their responsibility to ensure that their results adequately reflect the target platform.

Maybe it does? How do you know it doesn't without being able to identify a flaw in their testing or trying to reproduce their results?

If no one else can reproduce their data, then it clearly doesn't reflect what a buyer can expect out of the system, regardless of how they arrived at those numbers.

Who has tried to reproduce their results?

There're a thousand ways to get bad data. We could blindly speculate, but only they have the tools necessary to actually conclude which one it is.

So why label their data as bad or untrustworthy when you don't know either way?

Yes, if only they are seeing this supposed bug, then it doesn't represent the actual hardware, and thus makes for an invalid result.

Again, who has tried to reproduce their results? That involves matching as close as possible their system, setup (os version, drivers, everything), and test. As I mentioned, GN has also mentioned seeing strange behavior at times (it was either ADL or RPL) in performance that they think is due to the hybrid architecture, they just haven't seen it as a consistently as HWUB. Even if it is a bug or an issue with this game, they tried multiple times to retest and found the same results and gave the average without the game included. This is a reasonable way to address the issue. Hopefully they have time to maybe work with Intel and figure out what the issue is, but that shouldn't stop them from publishing in the manner they did.

 

[Carfax83]

There are too many variables with no controls and no data to make this type of conclusion and there are reasonable explanations and data from another reviewer showing that there is, or at least can be, a significant difference in platform power between the two which is not accounted for in the CPU alone.

Well I definitely agree with you about this. We don't have enough data to really come to a final conclusion about what happened. In the end it doesn't really matter. I've already unsubscribed from HWUB and if anyone asks me whether they are a quality reviewer I will tell them no. Quality reviewers don't publish anomalous results without investigating what, why, how and when until they come up with a reasonable explanation.

CapFrameX's twitter account posted performance numbers for Plague Requiem and they found poor performance on the 7950x compared to the 13900K in a populated area of the game. And since I own the game, I posted screenshots with MSI Afterburner of what I thought was the area in question, but it turns out the area that CapFrameX was referring to was in a much later chapter. The area that I investigated was in the early part of the game and I found out that it had occlusion culling issues that caused low GPU usage without spiking CPU usage, which ruled out a CPU bottleneck. I'm sure it was occlusion culling related as the camera angle had a strong impact on the framerate.

Anyway, I think I'm in the chapter now that CapFrameX was alluding to so when I get to the particular area, I will do a framerate analysis and see if there is any shenanigans going on that could explain the low performance on the 7950x.

https://twitter.com/x/status/1591802741265166336

 

[Carfax83]

I mean I usually play games like ArmA where the engine indeed does spend a lot of time actively sabotaging itself via the CPU. The spawned threads could be thought of redundant more so than useless. Or even, a quirk of drivers as this test was using a 4090 after all. It could be that the E cores feed the GPU better and the Nvidia driver asks for extra threads even though the game is bottlenecked somewhere else. HUB saw the 13600k system consume more power in every game, the closest gap being 40w. Keeping in mind its a total system power measurement, the difference could be entirely due to something like the intel motherboard having different power delivery, or default bios settings being overly aggressive from the factory.

I own the game and the game is only moderately CPU intensive in specific sections, ie towns or populated areas and whenever large amounts of rats are on screen. Everywhere else it's heavily GPU bound.

Also I'm pretty sure Windows 11 doesn't use E cores in this game unless it's for low priority background tasks like asset decompression, asynchronous shader compilation. When it comes to actual game performance and setting up frames, P cores should be used exclusively. This is based on my observation of thread usage in the game.

 

[Carfax83]

Because there s no need to investigate.

The 13900K take something like 35-40W more than the 7950X in games, assuming the 13900K has 5% better perf this imply that the GPU , wich is say at 390W with the 7950X, will madate 15% more power for thoses 5% better GPU perf, that s about 100W more GPU + CPU power and something like 130W more at the main.

According to Computerbase.de, the 7950x's average power consumption during games is 118w, and the 13900K was 144w.

 

[Carfax83]

In the few games tested, at stock settings, by Computerbase, the 13600K use 50W more than the 7600X in the most demanding games, you have at least 50% being explained, dunno what are the games tested by the decried site, if they have overclocked RAM/IMC and what are exactly the tested scenes...

FTR the 7600X hoover at 60-66W depending of the game...

Computerbase.de never tested the 7600x's power consumption in games. They tested the 7700x and the 7950x. The 7700x had an average power consumption of 77w and the 13600K had an average power consumption of 88w.

 

[Abwx]

Computerbase.de never tested the 7600x's power consumption in games.

They did, in the 13900K review, 60W on average and 66W at most.

Core i9-13900K, i7-13700K & i5-13600K: Gaming-Könige im Test: Leistungsaufnahme und Effizienz

Intel Raptor Lake im Test: Leistungsaufnahme und Effizienz / Leistungsaufnahme in Anwendungen ab Werk / Der Verbrauch steigt

www.computerbase.de

 

[nicalandia]

Over 50 games tested the 7600X is faster than the 5800X3D.

https://twitter.com/x/status/1601518366837342208

 

[eek2121]

Well I definitely agree with you about this. We don't have enough data to really come to a final conclusion about what happened. In the end it doesn't really matter. I've already unsubscribed from HWUB and if anyone asks me whether they are a quality reviewer I will tell them no. Quality reviewers don't publish anomalous results without investigating what, why, how and when until they come up with a reasonable explanation.

CapFrameX's twitter account posted performance numbers for Plague Requiem and they found poor performance on the 7950x compared to the 13900K in a populated area of the game. And since I own the game, I posted screenshots with MSI Afterburner of what I thought was the area in question, but it turns out the area that CapFrameX was referring to was in a much later chapter. The area that I investigated was in the early part of the game and I found out that it had occlusion culling issues that caused low GPU usage without spiking CPU usage, which ruled out a CPU bottleneck. I'm sure it was occlusion culling related as the camera angle had a strong impact on the framerate.

Anyway, I think I'm in the chapter now that CapFrameX was alluding to so when I get to the particular area, I will do a framerate analysis and see if there is any shenanigans going on that could explain the low performance on the 7950x.

https://twitter.com/x/status/1591802741265166336

I would not believe a word that comes out of that account. He/she claimed forever that they could not get DDR5 6000 working with the 7950x. They also claimed to have tried multiple kits. I brought up the fact that I had a 64gb kit working just fine, not only at DDR5 6000, but with tightened timings. Nothing except silence. Why does this matter? They clearly did something wrong on their end and failed to correct it, instead blaming AMD (they actually did blame AMD). That is when they lost credibility with me. One thing to note is that there are several options in the BIOS that can improve gaming performance depending on your setup, AGESA version, and whether you are single/dual CCD. For example, enabling NUMA mode for L3 on dual CCD helped raise my framerates (minimums and averages) by 15% or more in the games I've tested (one game went from 110-120fps with dips into the 80s to 190-210fps with the minimums at 135) Regardless of that, they've already started a negative RDNA3 spin by yelling at AMD for not providing telemetry for an unreleased card.

I'm not keen to buy that game, but if I do, I can run tests if anyone wants.

I am not saying they are wrong btw, but to me the credibility for that account is MLID or Adored level: less than zero. Shoot, they appear to be biased toward Intel (vs being neutral), so I will place them on an even lower rung on the ladder.

 

[Markfw]

I would not believe a word that comes out of that account. He/she claimed forever that they could not get DDR5 6000 working with the 7950x. They also claimed to have tried multiple kits. I brought up the fact that I had a 64gb kit working just fine, not only at DDR5 6000, but with tightened timings. Nothing except silence. Why does this matter? They clearly did something wrong on their end and failed to correct it, instead blaming AMD (they actually did blame AMD). That is when they lost credibility with me. One thing to note is that there are several options in the BIOS that can improve gaming performance depending on your setup, AGESA version, and whether you are single/dual CCD. For example, enabling NUMA mode for L3 on dual CCD helped raise my framerates (minimums and averages) by 15% or more in the games I've tested (one game went from 110-120fps with dips into the 80s to 190-210fps with the minimums at 135) Regardless of that, they've already started a negative RDNA3 spin by yelling at AMD for not providing telemetry for an unreleased card.

I'm not keen to buy that game, but if I do, I can run tests if anyone wants.

I am not saying they are wrong btw, but to me the credibility for that account is MLID or Adored level: less than zero. Shoot, they appear to be biased toward Intel (vs being neutral), so I will place them on an even lower rung on the ladder.

anybody who says they can't get ddr5 6000 working with a 7950x is an idiot or complete moron. I have three 7950x's , all 3 using ddr 6000 cl30 expo, never touched them after install working perfectly.

 

[Carfax83]

I'm not keen to buy that game, but if I do, I can run tests if anyone wants.

Well I definitely don't think you should buy the game just so you can confirm or deny CapFrameX's report, but I can definitely recommend the game. I don't like stealth oriented type gameplay, but this game's graphics and story are so good that it doesn't really bother me. Plus, you can kill enemies as well using a variety of tactics rather than just sneaking your way through them.

It's a legitimate next gen title with no last gen influence, and the DX12 implementation is very good with the exception of that one market area in the early part of the game that had some occlusion culling issues. HDR is also the best I've ever seen as well if your monitor or T.V supports it.

 

[DrMrLordX]

anybody who says they can't get ddr5 6000 working with a 7950x is an idiot or complete moron. I have three 7950x's , all 3 using ddr 6000 cl30 expo, never touched them after install working perfectly.

Or maybe they just bought non-EXPO memory kits.

 

[Markfw]

Or maybe they just bought non-EXPO memory kits.

I had 6600 exp kits that worked, just only 5800 speed. They said they could not get them to work, (did not specify that at any speed ??)

 

[DrMrLordX]

I had 6600 exp kits that worked, just only 5800 speed. They said they could not get them to work, (did not specify that at any speed ??)

Most people are surprised when memory kits will not run at their advertised speed. Well, most people who are not accustomed to building AMD rigs anyway.

 

[igor_kavinski]

Most people are surprised when memory kits will not run at their advertised speed. Well, most people who are not accustomed to building AMD rigs anyway.

That's why AMD had to create EXPO. I haven't heard anyone have a bad experience with one so far.

 

[Markfw]

Most people are surprised when memory kits will not run at their advertised speed. Well, most people who are not accustomed to building AMD rigs anyway.

I would like to point out that Intel and AMD have different memory controllers. The fact that EXPO kit DO work at advertised speeds just means that whatever is different about AMD's memory controller has been addressed by compatible memory. Being different is not wrong. And Intel is not the only game in town. Its like diesel and gasoline. They are different and each has its strengths, but they are not the same.

And as I said, EXP memory DID work, just not as the default speed. And if I was better or wanted to spend more time, maybe it world have worked. I just like "plug and play"

 

[nicalandia]

Whoever doubts HWUB numbers are clearly Biased and out of touch with Reality.

 

[DrMrLordX]

I would like to point out that Intel and AMD have different memory controllers.

It's been like this for years. At least we have some alternatives to XMP 2.0 now.

 

[eek2121]

Or maybe they just bought non-EXPO memory kits.

My kit was Intel XPM only and is (to this day) only certified for use on Intel’s latest motherboards. I selected the XMP profile and it worked fine.

I have been tightening timings when I have time.

 

[nicalandia]

Anyone with enough $$ to make a monster workstation with 192 Water Cooled Cores?

 

[Carfax83]

Finally got to that area that CapFrameX posted on their twitter account in Plague Tale Requiem. I checked the area and found no performance dips at all that would explain the low GPU usage he got with the Zen 4 rig. I thought it may be like that weird occlusion culling bug in Chapter 2 in the town market in Chapter 2, but I only encountered that problem once and this new area doesn't have that bug.

GPU usage is at 99-100% and CPU usage is fairly moderate at native 4K. So whatever caused that performance drop, is likely a bug or a configuration issue with CapFrameX's Zen 4 rig. One thing that stands out is that his rig is using a lot less memory than mine in this particular area; not VRAM but system memory.

 

[nicalandia]

Some interesting info coming from MLD

 

[Joe NYC]

It seems that if addition of V-Cache gave Zen 3 10-15% performance and if it really adds 30% performance to Zen 4, even though there are some gains from higher bandwidth and lower latency on Zen 4 L3, it seems there would also have to be bigger size of the V-Cache than 64 MB in order to achieve this sort of performance improvement. Maybe double...

 

[RnR_au]

VCache on mobile would be a bigger win I think. Mobile ram tends to be slow.

 

[coercitiv]

Some interesting info coming from MLD

When MLID says something it's never news, just speculation. That includes looking outside and reporting a full moon.

 

[Exist50]

Anyone with enough $$ to make a monster workstation with 192 Water Cooled Cores? View attachment 72785

Is that a 19" blade? Looks like if they add any more memory channels, that form factor's no longer going to work.