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!

 

[nicalandia]

its possible that their test uses the e-cores from the 12900k for the single core test?

Absolutely No Chance of that...

a single e-core does not have the grunt to beat a single Zen3 Core.

 

[Kedas]

If true, Userbenchmark will really need to update their test software again for some reason.

Anyway the game score is what will mainly matter, imagine that AMD doesn't even have to release the +64MB version to beat or match the 13900K....
Then the intel CEO better gets used to cold showers

unfortunately the prices will probably reflect those wins.

 

[nicalandia]

unfortunately the prices will probably reflect those wins.

And Intel will raise prices so it's a lose lose for gamers? Thankfully pair with a 3080 all you need is a CPU from the last two years.

 

[Makaveli]

If true, Userbenchmark will really need to update their test software again for some reason.

Anyway the game score is what will mainly matter, imagine that AMD doesn't even have to release the +64MB version to beat or match the 13900K....
Then the intel CEO better gets used to cold showers

unfortunately the prices will probably reflect those wins.

For for me I do more than just play games on my PC so gaming performance it not the main metric.

 

[DiogoDX]

well..... I don't know what's going on there, especially it's Userbenchmark

AMD Ryzen 5 7600X 6 Core "Zen 4" CPU Sample Destroys The Intel Core i9-12900K, 20% Faster Single-Core Performance In Leaked Benchmark

AMD's Ryzen 5 7600X 6-Core "Zen 4" CPU has been benchmarked once again & it destroys Intel's Core i9-12900K in single-threaded tests.

wccftech.com

Since is a new cpu they still did not have the chance to make Intel beat AMD. If I remember well by the time that Zen 1 and 2 were competing with better multicore they lowererd the weight of the multicore part of the test in the final score. I bet now that Intel will have better multicore they will increase the weight of that part on the test.

 

[Saylick]

Since is a new cpu they still did not have the chance to make Intel beat AMD. If I remember well by the time that Zen 1 and 2 were competing with better multicore they lowererd the weight of the multicore part of the test in the final score. I bet now that Intel will have better multicore they will increase the weight of that part on the test.

Maybe Userbenchmark will give points for every W of power consumption so that Intel dominates the charts. 🥲

 

[Exist50]

At some point Userbenchmark is going to give points for every W of power consumption so that Intel dominates the charts. 🥲

"Space heater multiplier"

 

[Saylick]

"Space heater multiplier"

*Netburst has entered the chat*
"Look who's at the top of the charts now?!"

 

[Kryohi]

Absolutely No Chance of that...

a single e-core does not have the grunt to beat a single Zen3 Core.

View attachment 65144

Highly likely that userbenchmark uses AVX512 in many tests, must have been useful during the days of Rocket Lake, but with zen4 this is the result...

 

[nicalandia]

Highly likely that userbenchmark uses AVX512 in many tests, must have been useful during the days of Rocket Lake, but with zen4 this is the result...

That would mean the 11900K would match the 12900K due to AVX-512 Making up for the IPC deficit... But the 12900K is about 23% ahead of the AVX-512 11900K..... So no AVX-512 is not part of the deal here

 

[DrMrLordX]

My only concern is the source UB is utter trash and you cannot really trust anything from them.

Same. Do not trust their results. They've never been useful.

 

[desrever]

Userbenchmarks probably has code to detect AMD CPUs and lower the score which didn't detect the new CPUs.

 

[Det0x]

well..... I don't know what's going on there, especially it's Userbenchmark

AMD Ryzen 5 7600X 6 Core "Zen 4" CPU Sample Destroys The Intel Core i9-12900K, 20% Faster Single-Core Performance In Leaked Benchmark

AMD's Ryzen 5 7600X 6-Core "Zen 4" CPU has been benchmarked once again & it destroys Intel's Core i9-12900K in single-threaded tests.

wccftech.com

Wow that was unexpected..



For comparison, this is my maxed 5950x (pretty much 100th percentile on every component)


Knowing userbenchmark history, how long until they change the benchmark weighting to keep Alderlake and soon to be Raptorlake on top again ?
(as we know, they have already done it before )

 

[nicalandia]

Userbenchmarks probably has code to detect AMD CPUs and lower the score which didn't detect the new CPUs.

Userbenchmark has never been kind to AMD even in ES samples, and the time the Ryzens started to compete, they go and change the algorithim to favor Intel. Let see how the do that now.

 

[Det0x]

Seems like its userbenchmark use of alittle avx512 to keep rocket lake ahead of Zen3 that's backfiring now..

Here are 8 Alderlake P cores with avx512 disabled


Here are 8 Alderlake P cores with avx512 enabled

 

[Hans de Vries]

From APISAK as well:

https://twitter.com/x/status/1553038437191536641

 

[nicalandia]

Seems like its userbenchmark use of alittle avx512 to keep rocket lake ahead of Zen3 that's backfiring now..

Here are 8 Alderlake P cores with avx512 disabled
View attachment 65148

Here are 8 Alderlake P cores with avx512 enabled
View attachment 65149

Are you serious?? 1T it's only 3.5% performance boost and that can be attributed to other factors.

Zen4 is not beating Alder Lake that much due to AVX-512. Let's be clear about that.

 

[Det0x]

Are you serious?? 1T it's only 3.5% performance boost and that can be attributed to other factors.

Zen4 is not beating Alder Lake that much due to AVX-512. Let's be clear about that.

I dont know, just thinking out loud..
At what clockspeed does Alderlake P-cores run avx512 instructions vs non-avx512 instructions by "stock settings" ? (--> what's the avx512 offset by default?)

AVX negative ratio offsets are very useful to achieve maximum performance for both SSE and AVX workloads. While the function carried over in principle from Rocket Lake, there are some big differences in the implementation of AVX ratio offset between Alder Lake and Rocket Lake. First, on Alder Lake, the AVX2 negative ratio offset is only applied to the P-cores. The E-core frequency is unaffected. Second, by default, the maximum ratio during an AVX workload is the Turbo Boost 2.0 ratio. If you want an offset of 0, so AVX workload doesn’t reduce the frequency, you’ll need to manually set 0. Note that on some BIOSes programming 0 means programming “default”, in which case you’ll be limited to a maximum 51X for the AVX ratio on the 12900K.

Theoretical numbers pulled from my behind:
Alderlake 12900k ST avx512 workload -> run at 4900mhz or 4700mhz? (maybe even lower?)
Alderlake 12900k ST non-avx512 workload -> run at 5200mhz
Zen4 ST avx512 workload -> run at 5ghz (?)

Just trying to find a explanation for these too good to be true numbers..

 

[nicalandia]

I dont know, just thinking out loud..


Just trying to find a explanation for these too good to be true numbers..

Both the Non-AVX and the AVX showed average boost to 4.9 on both runs. Cinebench has been confirmed to not allow AVX because the clocks get so low that it just negates any Boost in performance. That should also apply to AMD so any performance gain by AVX-512 would be nullified by the lower clocks and that Zen4 score would have been about even with Zen3, Not 40%+ above.. Unless AMD has sprinkled fairy dust on their Zen4 and their CPU do not down clock or get overheated by AVX-512...

 

[HurleyBird]

Unless AMD has sprinkled fairy dust on their Zen4 and their CPU do not down clock or get overheated by AVX-512...

AMD has the benefit of a much better process, and we don't know how their implementation compares to Intel's anyway.

That said, part of the reason that 1T is so much better than 2T here could be down to AVX-512 running at or near full speed in 1T, while clocking down >=2T.

 

[dr1337]

Unless AMD has sprinkled fairy dust on their Zen4 and their CPU do not down clock or get overheated by AVX-512...

The 5.5ghz all core demo'd at computex is already much more efficient than what intel can do, granted without considering FP. Just an assumption but perhaps AMD is actually able to run AVX512 much faster than intel seeing as their engineering samples can also clock much higher in games/other workloads?

 

[nicalandia]

AMD has the benefit of a much better process, and we don't know how their implementation compares to Intel's anyway.

If that's the case. Then Intel Is DONE at the Top level HPC/AI since AMD will have a 96 Core monster that will plain old murder 60C at AVX-512

 

[FangBLade]

Can't wait to see Henry swagger spit on userbenchmark.

 

[Exist50]

If that's the case. Then Intel Is DONE at the Top level HPC/AI since AMD will have a 96 Core monster that will plain old murder 60C at AVX-512

AMX (and perhaps one or two other bells and whistles) might buy them some room for inference, but yeah, I'm not seeing anyone seriously claim Intel will be competitive for more general purpose workloads.

 

[FangBLade]

If that's the case. Then Intel Is DONE at the Top level HPC/AI since AMD will have a 96 Core monster that will plain old murder 60C at AVX-512

And best of all is that Zen 5 will be the first one that fully utilize Xilinx AI hardware accelerators or something like that, and it is clear that AMD saved all resources and IPC gain for Zen 5, Intel is in real trouble if they can't beat Zen 4 that is only evolution of Zen 3, while Zen 5 will be revolution, complete redesign architecture with Xilinx product.