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!

 

[moinmoin]

Just recently AMD was non existent on servers and suddenly now it equips the absolute fastest machines.
It's absurd how fast this happened and I can't imagine how people on Intel may feel.

Even better Frontier is the one project that arguably brought out AMD from depth of bankruptcy back toward offering high end tech. Game consoles ensured AMD's day to day business could continue. DARPA's Forward sponsoring over the years ensured AMD could put R&D into planing and creating the hardware that eventually cumulated in Zen 3 and CDNA2 as used in Frontier.

 

[nicalandia]

If 1300 at 3,7Ghz, then 2000 at 5,7?

The 1T keeps jumping on 12 different CCD. Desktop could reach 2200 points

 

[uzzi38]

So TDP should be pretty tame. Like less than 400W.

400W is max cTDP (c = configurable) for Genoa. This is per-socket FYI.

Servers run hard power limits that are set in BIOS, there's none of the traditional boosting of high power limit -> sustained power limit or any of that.

 

[jamescox]

Latest Cinebench R23 ST Run on an ES sample. Cinebench can only detect 256 Threads, but this is a 384T CPU. If CBR23 could take advantage of all cores, we might be looking past 160,000 points.

View attachment 66110

ST: 1300 Points at 3.7 Ghz
MT(256T)110,000 Points at 2.15 Ghz

So it is using 64-cores per processor and the clock speed is only 2.15 GHz? That isn’t a very impressive clock speed considering the 7763 runs 2.45 GHz. I guess it might be an earlier, low clocked ES part. Curious as to what clock speed they will be able to hit with the full 96-cores. It is funny that the number of cores in systems has outpaced the software support.

 

[moinmoin]

It is funny that the number of cores in systems has outpaced the software support.

That's nothing new though and will be repeated as long as developers put in short sighted limits. OS support is most important in any case. Linux stops at 8192 cores at the moment, that's plenty room left lol.

 

[nicalandia]

I posted here first folks.... waited to see if WCCFTECH Would pick it up..... I had to ask YuuKi_Ans to post it and now WCCFTECH posted it...

Dual AMD EPYC Genoa 96-Core CPUs Score Over 110K Points In Cinebench R23 With Only 66% of Their Active Threads

Dual AMD EPYC Genoa CPUs rocking 96 cores each have been benchmarked in Cinebench R23 and show some insane performance numbers.

wccftech.com

 

[Abwx]

I posted here first folks.... waited to see if WCCFTECH Would pick it up..... I had to ask YuuKi_Ans to post it and now WCCFTECH posted it...

Dual AMD EPYC Genoa 96-Core CPUs Score Over 110K Points In Cinebench R23 With Only 66% of Their Active Threads

Dual AMD EPYC Genoa CPUs rocking 96 cores each have been benchmarked in Cinebench R23 and show some insane performance numbers.

wccftech.com

Someone pointed that the displayed 3.76GHz on bottom of the pic has a 3 wich is not at a right scale, i link your pic :

 

[nicalandia]

Someone pointed that the displayed 3.76GHz on bottom of the pic has a 3 wich is not at a right scale, i link your pic :

Yes. I got a few blurred sections on the original picture. I tried my best to clean it with Paint. Got with YuuKi_Ans, he reviewed, like it and posted on his Twitter Account

 

[Saylick]

I posted here first folks.... waited to see if WCCFTECH Would pick it up..... I had to ask YuuKi_Ans to post it and now WCCFTECH posted it...

Dual AMD EPYC Genoa 96-Core CPUs Score Over 110K Points In Cinebench R23 With Only 66% of Their Active Threads

Dual AMD EPYC Genoa CPUs rocking 96 cores each have been benchmarked in Cinebench R23 and show some insane performance numbers.

wccftech.com

I, too, am surprised they didn't find this thread and then immediately post an "exclusive" article saying their "sources" sent them Genoa benchmarks.

 

[LightningZ71]

Yes. I got a few blurred sections on the original picture. I tried my best to clean it with Paint. Got with YuuKi_Ans, he reviewed, like it and posted on his Twitter Account

Wow, you aren't winning many friends on the reddit thread.

 

[nicalandia]

Wow, you aren't winning many friends on the reddit thread.

Yeah, I deleted the post, people freaking out due to my bad paint skills...

This is the Original.

Can Anyone Please review it and see if it looks legit before I make a fool of myself?

 

[Markfw]

Yeah, I deleted the post, people freaking out due to my bad paint skills...

This is the Original.

Can Anyone Please review it and see if it looks legit before I make a fool of myself?

View attachment 66161

OK, first, 96 core cpus that are strong ? Yes, I can believe they are 3.76 based on task manager. Its like somebody edited the (red)76. Or the 2.15 in cinebench. I really don't know. I think its time to wait for more reputable benchmarks.

 

[nicalandia]

OK, first, 96 core cpus that are strong ? Yes, I can believe they are 3.76 based on task manager. Its like somebody edited the (red)76. Or the 2.15 in cinebench. I really don't know. I think its time to wait for more reputable benchmarks.

The 2.15 is the Base Clock. The chip is capable of boosting to 3.7 Ghz

 

[jamescox]

That's nothing new though and will be repeated as long as developers put in short sighted limits. OS support is most important in any case. Linux stops at 8192 cores at the moment, that's plenty room left lol.

Apparently they only put support for 256 threads; very short sighted… At least it functions with more than 256 available.

Linux has been running such large, single system image machines for over 20 years:

NASA Ames Installs World's First Altix 512-Processor Supercomputer

I think they had a 1024 core machine in 2001, but I don't think they used standard Linux until 2006. Those machines were spread across many cabinets but they were still a single system image. We might have 1024 threads in a dual socket system with Zen 5, so running a single thread will be close to 0.1 % utilization. I look forward to gaming benchmarks from LTT.

 

[moinmoin]

Even better Frontier is the one project that arguably brought out AMD from depth of bankruptcy back toward offering high end tech. Game consoles ensured AMD's day to day business could continue. DARPA's Forward sponsoring over the years ensured AMD could put R&D into planing and creating the hardware that eventually cumulated in Zen 3 and CDNA2 as used in Frontier.

Frontier Exascale Unveiling: 'Breathtaking..., a Huge Leap Forward for Science, for Our Country" - High-Performance Computing News Analysis | insideHPC

It was a day of high pride at Oak Ridge National Laboratory, a day to stop and take stock of the successful completion of a project based on an idea – exascale computing – that was begun more than 20 years ago. We’re talking about the official unveiling of Frontier, the country’s and the world’s […]

insidehpc.com

Article quotes Lisa Su:
"We are a … company with 25,000 employees, and I can tell you, personally, there was no project or company that was more important than delivering Frontier."

I wish to hear a post mortem about Frontier by Mark Papermaster who has been in the same position from the very beginning to the very end, so he should be able to tell all the stories regarding the team building along with the first Forward grant applications back in 2012 etc.

 

[nicalandia]

What are the chances of the AMD MI300 Instinct showing up on Geekbench. Some of the runs I have been able to find look like they are SOC APU?

 

[inf64]

I was a bit bored due to slow news and played around with the charts and specs for Raptor lake and Ryzen 7000. Below is my "final" guesswork/perf. estimate for both lineups (ST and MT; no gaming). Data is based on the known and rumored specs:



Edit:
Here is how I got to the numbers above.

 

[Thibsie]

Thanks. But what is a 7900E ?

 

[inf64]

Thanks. But what is a 7900E ?

It's a part that I suspect AMD will use to fill in the large performance gap between 7900X and 7800X. If it launches, it will likely be called 7900 (no E). I expect it to have lower TDP and therefore lower boost (ST and all core). That is the only thing that makes sense, as the "3D" versions will not affect the application performance that much, as seen in 5800X3D benchmarks.

 

[Rigg]

I was a bit bored due to slow news and played around with the charts and specs for Raptor lake and Ryzen 7000. Below is my "final" guesswork/perf. estimate for both lineups (ST and MT; no gaming). Data is based on the known and rumored specs:

*Edited for terrible math

5950x - 7950x 30% sc 38% mc
5900x - 7900x 29% sc 26% mc
5800x - 7800x 25% sc 27% mc


Why the discrepancy in multi-core performance gain for the 5950x vs the others?

 

[inf64]

5950x - 7950x +27% sc +63% mc
5900x - 7900x +26% sc +36% mc
5800x - 7800x +22% sc +27% mc

Why the massive discrepancy in multi-core performance gain for the 5950x vs the others?

Hmm, how did you get the 63% MT gain VS 5950X? 7950X is 228% while 5950X is 165%, that is 228/165~=1.38 or 38% gain.

 

[Rigg]

Hmm, how did you get the 63% MT gain VS 5950X? 7950X is 228% while 5950X is 165%, that is 228/165~=1.38 or 38% gain.

I guess i'm bad at math...LOL

 

[Saylick]

I was a bit bored due to slow news and played around with the charts and specs for Raptor lake and Ryzen 7000. Below is my "final" guesswork/perf. estimate for both lineups (ST and MT; no gaming). Data is based on the known and rumored specs:

View attachment 66209View attachment 66210

Now do a plot for power draw!

But in all seriousness, gotta give it to Intel for maintaining competitiveness on an older, likely still cheaper, node.

 

[inf64]

Now do a plot for power draw!

But in all seriousness, gotta give it to Intel for maintaining competitiveness on an older, likely still cheaper, node.

I have an excel chart with all the mental gymnastics, so if anyone is interested here it is: https://docs.google.com/spreadsheet...ouid=117779538430217342497&rtpof=true&sd=true

 

[reb0rn]

@inf64
graph made by AMD fanboy as many here still promote and FUD intel for their own reason, there is <1% chance 16/32 7950x is faster then 24/32 13900k






Use of word fanboy in the tech forums is prohibited.


esquared
Anandtech Forum Director