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! 🙂

 

[wilds]

This is the kind of stuff that will be possible with Zen 4 3D but Zen 3 is not dead yet.
View attachment 69224

Still not sure if this was real or not. Saw a few claims it was fake.

 

[Hans Gruber]

Still not sure if this was real or not. Saw a few claims it was fake.

The reason why the 5800x3D was limited to 4.5ghz was because of the v-cache memory. AMD believed they were voltage limited on Zen 3. The Zen4 3D v-cache will not have the name voltage limit issues.

If the 5800x3d is golden silicon and the performance limits were removed. I could see a 5.2ghz fully functional 5800x3D CPU.

 

[Det0x]

Still not sure if this was real or not. Saw a few claims it was fake.

It is alive doing benches atleast 🙂


More benches with that chip can be found here

 

[nicalandia]

It is alive doing benches atleast 🙂
View attachment 69258View attachment 69259

More benches with that chip can be found here

Any gaming benchmarks?

 

[Kaluan]

Still not sure if this was real or not. Saw a few claims it was fake.

Yeah... If we're talking about the same thing... the claims were made by Intel slanded people, and not even tech-savvy ones, basically coping people.

I wouldn't put any weight in what such people say, unless they have valid proof.

 

[nicalandia]

Some info on Cooling...

 

[nicalandia]

Some Spec numbers with Low end Air Cooling



"Note: The results of the 7950X in the picture are taken from my new ITX machine, which uses AXP90-X47 low-power down-pressure heat sink, so it significantly affects the single-core maximum frequency (the higher IPC sub-item is only around 5.4 GHz)"

SPEC CPU 2017 单线程整数性能测试与总结 (2022)

简单分享最近一段时间收集的SPECint2017 rate-1单线程性能数据。由于个人时间精力限制，2022年我应该只会测这些CPU了，所以发一篇总结。 测试环境编译器统一为 GCC 12经过对比确认，GCC 12.1与12.2几乎不存在性能…

zhuanlan.zhihu.com

 

[igor_kavinski]

AMD Cuts Down Ryzen 7000 "Zen 4" Production As Demand Drops Like a Rock

AMD reportedly scaled down production of its Ryzen 7000 series desktop processors in response to bleak demand across the PC hardware industry. Wccftech claims to have read an internal company document calling for reduced supply to the channel as market response to the Ryzen 7000-series is weak...

www.techpowerup.com

What will AMD do next? Launch Zen 4 on AM4?

 

[nicalandia]

What will AMD do next? Launch Zen 4 on AM4?

Highly unlikely.. What's more profitable? Saving 8C/16T CCDs for Genoa or downgrade Zen4 performance so a few can upgrade?

 

[inf64]

AMD Cuts Down Ryzen 7000 "Zen 4" Production As Demand Drops Like a Rock

AMD reportedly scaled down production of its Ryzen 7000 series desktop processors in response to bleak demand across the PC hardware industry. Wccftech claims to have read an internal company document calling for reduced supply to the channel as market response to the Ryzen 7000-series is weak...

www.techpowerup.com

What will AMD do next? Launch Zen 4 on AM4?

There is zero % chance of that happening. AM4 is fine as it is, AM5 just needs more time and a bit better economic circumstances. Intel will face the same problem once they launch their next platform.

 

[HurleyBird]

There is zero % chance of that happening. AM4 is fine as it is, AM5 just needs more time and a bit better economic circumstances. Intel will face the same problem once they launch their next platform.

It's probably non-zero, but it really depends how much AMD hedged their bets in advance. There was talk of Zen 4 AM4 prototypes, but how far did development get? If AMD knew a year ago what they do now about economic conditions I have no doubt they would also plan to launch Zen 4 on AM4. The problem is that, if there's six months of work left to get Zen 4 AM4 ready to launch, by the time it does DDR5 is likely much cheaper, non-vcache Zen 4 is going to be bottlenecked by DDR4 to some extent, and you've pulled resources from other areas for a prolonged period of time.

That said, if Nvidia can swallow their pride and unlaunch the fake 4080, AMD should be able to swallow theirs and lower prices of at least the 7600X, 7700X, as well as the B650, and X670 chipsets (assuming that chipset price hikes are one of the major reasons for absurd motherboard prices.)

 

[deasd]

AMD Cuts Down Ryzen 7000 "Zen 4" Production As Demand Drops Like a Rock

AMD reportedly scaled down production of its Ryzen 7000 series desktop processors in response to bleak demand across the PC hardware industry. Wccftech claims to have read an internal company document calling for reduced supply to the channel as market response to the Ryzen 7000-series is weak...

www.techpowerup.com

What will AMD do next? Launch Zen 4 on AM4?

Next? Just launch 125$ B650 and even cheaper A620.

Also, don't trust WCCF.

 

[igor_kavinski]

Question - Windows 11 hobbling Ryzen 7000 performance

It looks like Windows 11 is hobbling Ryzen 7000 gaming performance currently.... for more details, read more at PcGamer: Hobbled Ryzen 7000

forums.anandtech.com

Intel gamers already in the habit of turning off E-cores and now AMD gamers may want to turn off both CCD and SMT on the 7950X for performance boost.

EDIT: Sorry, already posted few pages back.

 

[nicalandia]

Okay, took me a while but I was able to compile the most extensive Linux Benchmarks one could find on these CPUs

13900K(DDR4/DDR5) vs 12900K, 5950X, 7950X

.CPU is 13900K DDR4





It's good that The Mining Craze just collapsed.

AMD Ryzen 7900 Series Linux CPU Benchmarks Performance - OpenBenchmarking.org

openbenchmarking.org

 

[Markfw]

Okay, took me a while but I was able to compile the most extensive Linux Benchmarks one could find on these CPUs

13900K(DDR4/DDR5) vs 12900K, 5950X, 7950X

.CPU is 13900K DDR4
View attachment 69335

View attachment 69336


It's good that The Mining Craze just collapsed.
View attachment 69337

AMD Ryzen 7900 Series Linux CPU Benchmarks Performance - OpenBenchmarking.org

openbenchmarking.org

Total domination, as I suspected.

 

[nicalandia]

Total domination, as I suspected.

The 7950X is an Equal Opportunity(indiscriminate) executioner it just tramples the 5950X and 12900K As Well.

 

[eek2121]

It's probably non-zero, but it really depends how much AMD hedged their bets in advance. There was talk of Zen 4 AM4 prototypes, but how far did development get? If AMD knew a year ago what they do now about economic conditions I have no doubt they would also plan to launch Zen 4 on AM4. The problem is that, if there's six months of work left to get Zen 4 AM4 ready to launch, by the time it does DDR5 is likely much cheaper, non-vcache Zen 4 is going to be bottlenecked by DDR4 to some extent, and you've pulled resources from other areas for a prolonged period of time.

That said, if Nvidia can swallow their pride and unlaunch the fake 4080, AMD should be able to swallow theirs and lower prices of at least the 7600X, 7700X, as well as the B650, and X670 chipsets (assuming that chipset price hikes are one of the major reasons for absurd motherboard prices.)

The only thing outselling Zen 4 is Zen 3. I doubt AMD cares much.

Overall demand is weaker across all of big tech.

 

[Abwx]

5950X/7950X/5995WX HPC comparison at Puget.

AMD Ryzen 7950X Scientific Computing Performance - 7 Optimized Applications

This post presents scientific application performance testing on the new AMD Ryzen 7950X. I am impressed! Seven applications that are heavy parallel numerical compute workloads were tested. The 7950X outperformed the Ryzen 5950X by as much as 25-40%. For some of the applications it provided...

www.pugetsystems.com

64 cores are 64 cores, even lowly clocked and from previous uarch.

 

[Markfw]

5950X/7950X/5995WX HPC comparison at Puget.

AMD Ryzen 7950X Scientific Computing Performance - 7 Optimized Applications

This post presents scientific application performance testing on the new AMD Ryzen 7950X. I am impressed! Seven applications that are heavy parallel numerical compute workloads were tested. The 7950X outperformed the Ryzen 5950X by as much as 25-40%. For some of the applications it provided...

www.pugetsystems.com

64 cores are 64 cores, even lowly clocked and from previous uarch.

They use 4800 for the 7950x. While not offcially spec, even AMD says use 6000 cl30. The other 2 are using 3200 which is NOT spec.

Obviously flawed setup, so the results are half useless.

 

[moinmoin]

They use 4800 for the 7950x. While not offcially spec, even AMD says use 6000 cl30. The other 2 are using 3200 which is NOT spec.

Obviously flawed setup, so the results are half useless.

I disagree. The areas Puget Systems tend to test are ones where using ECC memory makes sense which seldom go significantly beyond JEDEC specs. So while not matching typical mainstream DIY specs, it's still plenty useful.

 

[Markfw]

I disagree. The areas Puget Systems tend to test are ones where using ECC memory makes sense which seldom go significantly beyond JEDEC specs. So while not matching typical mainstream DIY specs, it's still plenty useful.

You just put a 25% handicap on the 7950x. And they did not use avx512, which a few applications also do. And 16 cores compared to 64 ? whats that about.

 

[nicalandia]

I disagree. The areas Puget Systems tend to test are ones where using ECC memory makes sense which seldom go significantly beyond JEDEC specs. So while not matching typical mainstream DIY specs, it's still plenty useful.

Even Genoa ES CPUs are being tested with 5200 DDR5. Why limit the Ryzen 7000 CPUs to 4800?

 

[moinmoin]

Even Genoa ES CPUs are being tested with 5200 DDR5. Why limit the Ryzen 7000 CPUs to 4800?

That I don't know. Maybe Puget Systems only sells them with 4800? 🤷

 

[Abwx]

They use 4800 for the 7950x. While not offcially spec, even AMD says use 6000 cl30. The other 2 are using 3200 which is NOT spec.

Obviously flawed setup, so the results are half useless.

If official spec is 5200 non ECC then it looks right that ECC will mandate one bin lower frequency, dunno if 5200 is also supported by the 7950X when it s with ECC.

Anyway it shouldnt have a big influence in things like Linpack, possibly in large compilations.

 

[nicalandia]

By the Way, I just got banned from Techpowerup Forums, because of the 5800X vs 12900k 4090 GPU Benchmarks.