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!

 

[Panino Manino]

Generational improvements (Older results are all from SiSoft Sandra articles)
Still waiting for the Ryzen 9 7950X...

View attachment 67589
Made with Excel (plus some image editing)

Looking just at the 12 core, the improvement from Zen 2 to Zen 3 was "small". The reason why Zen 4 is showing these absurd gains it because Zen 3 was power constrained? At the top of my ignorance looks like the new TDP is allowing Zen to give all it got.

 

[Tuna-Fish]

He most likely has a P5800X drive

Right now he's actually using a M2 Macbook Air, iirc. He's been asking for a good powerful ARM desktop or laptop for a long time so he can work self-hosted on the arm branch, so when Asahi Linux got good enough to be barely usable he switched over to Mac hardware.

 

[Kaluan]

Looking just at the 12 core, the improvement from Zen 2 to Zen 3 was "small". The reason why Zen 4 is showing these absurd gains it because Zen 3 was power constrained? At the top of my ignorance looks like the new TDP is allowing Zen to give all it got.

Well, might be just me, but I don't think you're supposed to get nearly 55% gains in one test (non-AVX512) from "just" 20-25% higher clocks.


BTW the 7900X and 7700X Sandra results are from different systems and different firmwares, just a small FYI.

 

[igor_kavinski]

so when Asahi Linux got good enough to be barely usable he switched over to Mac hardware.

Wow! Torvalds Modified Fedora Linux to Run on his Apple M2 Macbook (itsfoss.com)

Fedora now. I guess he loves the low heat buildup and good battery life. And there's the looking cool factor too. You can't really expect him to be objective. He's pretty opinionated. If computing power was his only requirement, he would be using a Ryzen 5995WX workstation.

 

[coercitiv]

If computing power was his only requirement, he would be using a Ryzen 5995WX workstation.

No reason he can't have both.

Linus Torvalds drops Intel and adopts 32-core AMD Ryzen Threadripper on personal PC

‘My 'allmodconfig' test builds are now three times faster than they used to be’ says Linux overlord

www.theregister.com

In fact, the biggest excitement this week for me was just that I upgraded my main machine, and for the first time in about 15 years, my desktop isn't Intel-based. No, I didn't switch to ARM yet, but I'm now rocking an AMD Threadripper 3970x. My 'allmodconfig' test builds are now three times faster than they used to be, which doesn't matter so much right now during the calming down period, but I will most definitely notice the upgrade during the next merge window.

 

[igor_kavinski]

I wish he would write about the performance difference between his 3970X and M2 MBP.

 

[LightningZ71]

Looking just at the 12 core, the improvement from Zen 2 to Zen 3 was "small". The reason why Zen 4 is showing these absurd gains it because Zen 3 was power constrained? At the top of my ignorance looks like the new TDP is allowing Zen to give all it got.

The 5900X and 5950X were constrained in two ways, which is evident when you compared the all-core MT load clock speeds between the 5800x and the two dual CCD parts. The dual CCD parts would often run up against limitations in memory throughput having the same memory setup as the 5800x. They were also constrained by total socket power, having notably lower all core frequencies than the 5800x, unless you messed with the power limits. Once the power limits were relaxed, you then ran up against the limits of extracting heat from the total package. N5 as used in Zen4 and the new DDR5 memory system address all three of those limits.

 

[Hans Gruber]

The 5900X and 5950X were constrained in two ways, which is evident when you compared the all-core MT load clock speeds between the 5800x and the two dual CCD parts. The dual CCD parts would often run up against limitations in memory throughput having the same memory setup as the 5800x. They were also constrained by total socket power, having notably lower all core frequencies than the 5800x, unless you messed with the power limits. Once the power limits were relaxed, you then ran up against the limits of extracting heat from the total package. N5 as used in Zen4 and the new DDR5 memory system address all three of those limits.

DDR5 is currently constrained by very high latencies in the DDR5 chips.

 

[maddie]

DDR5 is currently constrained by very high latencies in the DDR5 chips.

Is it true that cache can mitigate latencies. albeit not always, but can never increase absolute bandwidth.

 

[eek2121]

DDR5 is currently constrained by very high latencies in the DDR5 chips.

Meh, not as much as you may think. Just because a DDR5 6400 kit has more than double the latency of a DDR4 3200 kit, it doesn't necessarily mean things are worse.

 

[Hans Gruber]

Meh, not as much as you may think. Just because a DDR5 6400 kit has more than double the latency of a DDR4 3200 kit, it doesn't necessarily mean things are worse.

If you take DDR3 vs DDR4. The latencies are basically doubled. So DDR5 vs. DDR4 would be DDR5 latencies down the road (best case scenario) 28 across the board. So in a year or two. DDR5 6000mhz would be 28-32 latencies.

The speeds right out of the box are very fast for DDR5. The latencies are a bit high.

 

[nicalandia]

Zen4 vs Golden Cove, core vs core.

 

[igor_kavinski]

RAM Latency Calculator (notkyon.moe)



That's the DDR5 we need to reach just DDR4-3200 CL14.



You can see how expensive you have to go on DDR5 side to match or exceed relatively affordable DDR4.

 

[igor_kavinski]

View attachment 67597

Crap! Core i5-12600K gets absolutely slaughtered here!

 

[Asterox]

RAM Latency Calculator (notkyon.moe)

View attachment 67595

That's the DDR5 we need to reach just DDR4-3200 CL14.

View attachment 67599

You can see how expensive you have to go on DDR5 side to match or exceed relatively affordable DDR4.

Main point is, what can do very affordable DDR4.

Local prices for 2x8 ddr4 or ddr5 combination, or 60euro ddr4 vs 102euro ddr5.

Protis - Kingston Fury Beast Black 16GB DDR5 5200-40, (2x8GB), KF552C40BBK2-16

www.protis.hr

Protis - Memorija Team Vulcan Z Grey 16GB DDR4 K2 (2x8GB) 3600MHz, TLZGD416G3600HC18JDC01

www.protis.hr

In short, for AM5 platform ddr5 memory prices they are no longer a problem.

Cheep AM5 motherboards, that's a problem because locally we can expect no less than 160 euros for the most affordable model.

 

[Timmah!]

Main point is, what can do very affordable DDR4.

Local prices for 2x8 ddr4 or ddr5 combination, or 60euro ddr4 vs 102euro ddr5.

Protis - Kingston Fury Beast Black 16GB DDR5 5200-40, (2x8GB), KF552C40BBK2-16

www.protis.hr

Protis - Memorija Team Vulcan Z Grey 16GB DDR4 K2 (2x8GB) 3600MHz, TLZGD416G3600HC18JDC01

www.protis.hr

View attachment 67603

In short, for AM5 platform ddr5 memory prices they are no longer a problem.

Cheep AM5 motherboards, that's a problem because locally we can expect no less than 160 euros for the most affordable model.

Whats the problem around here, where i live, is availibility of 64GBkits with CL higher than 5200. I was looking at GSkill 64GB CL30 kit, which for a while seemed to be "on road" on local main HW e-shop, but now they list it as "unavailable".

 

[Hans Gruber]

The problem is AMD says AM5 Zen 4 sweet spot is 6000mhz. The way to overcome latency is with high speed ram. Does this mean the same memory divider problem will exist with Zen 4 as Zen 3 faces with speeds above 3800-4000mhz? The 1:1 memory divider issue.

I am not trying to jump the gun here but I am hearing 8000mhz DDR5 speeds have been achieved. There is DDR4 memory that can go above 5000mhz.

 

[TESKATLIPOKA]

RAM Latency Calculator (notkyon.moe)

View attachment 67595

That's the DDR5 we need to reach just DDR4-3200 CL14.

View attachment 67599

You can see how expensive you have to go on DDR5 side to match or exceed relatively affordable DDR4.

Latency is the same, true, but DDR5-6400 has also 2x higher BW.
I don't think even with slower DDR5 you would lose a lot of performance in games, but reviews will tell.
It would be interesting to compare Zen4 with 3d cache + slow DDR5 vs Zen4 without 3d cache + fast DDR5.

 

[igor_kavinski]

I was looking at GSkill 64GB CL30 kit, which for a while seemed to be "on road" on local main HW e-shop, but now they list it as "unavailable".

I did say in this thread or some other that tasty DDR5 kits are gonna be scooped up by "interested" parties closer to the AM5 launch

Next time, pay heed to my ramblings. MUAHAHAHA!

 

[igor_kavinski]

Latency is the same, true, but DDR5-6400 has also 2x higher BW.
I don't think even with slower DDR5 you would lose a lot of performance in games, but reviews will tell.

Depends on the game. Some games love lower latency. Those with good multicore scalability and penchant for crunching through gobs of data love bandwidth.

 

[nicalandia]

Zen4 has amazing Multi-Media performance too.

Here is a 7700X vs a 12900KF(8P + 8e) Recently published data(As in Today)


AMD Ryzen 7 7700X


Intel i9 12900KF


Overall the 7700X is
Multi-media Integer: 20% Higher
Multi-media Long-int: 5% Higher
Multi-media Single-float: about even
Multi-media Double-float: about even

The 12900KF is 11% Higher than the 7700X in Multi-media Quad-Float

 

[igor_kavinski]

Woah. 12900KF rank sucks! Looks like lots of overclockers kicked it down to hell.

 

[AMDK11]

Zen4 vs Golden Cove, core vs core.

View attachment 67597

AVX512?

 

[nicalandia]

AVX512?

No AV-X512 just pure Raw unadulterated Integer Performance... If you check the Floating Point performance they are about even on that(At least on the 8C/16T parts since they have similar speed and IPC)

 

[LightningZ71]

On ram latency, realize that, with healthy L3 caches and the increased L2 cache size, dram latency is reduced in importance. Yes, there will always be some impact, but we're splitting hairs here. First word latency is larger for DDR5 today, but, for a transfer of any appreciable size, DDR5 will complete the memory transfer before DDR4 does, even if it gets a few dozen cycles head start.