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]

Remember how OpenVINO is the Absolute best app to take advantage of AVX-512 and AMX?

Here is a nice comparison between recent Desktop CPUs, 7950X with and without AVX-512, 13900K(No AVX), 12900K(No AVX-512), 5950X, 3950X, 11900K(AVX-512) and 7700X(Default)




AVX-512 Benchmark

 

[nicalandia]

A single Genoa 9654 posting World record MT GB5 Score

https://browser.geekbench.com/v5

 

[nicalandia]

I posted the Wrong Link.

GIGABYTE MZ33-AR0-000 - Geekbench

Benchmark results for a GIGABYTE MZ33-AR0-000 with an AMD EPYC 9654 processor.

browser.geekbench.com

Thats the most powerful(MT) Single CPU in the World(17% slower than the 9654 2S System pisted).

Geekbench Browser

Threadripper will just Curbstomp every thing in its path

 

[Saylick]

I posted the Wrong Link.

GIGABYTE MZ33-AR0-000 - Geekbench

Benchmark results for a GIGABYTE MZ33-AR0-000 with an AMD EPYC 9654 processor.

browser.geekbench.com

Thats the most powerful(MT) Single CPU in the World(17% slower than the 9654 2S System pisted).

View attachment 76310

Geekbench Browser

Threadripper will just Curbstomp every thing in its path

Dude... We could totally use something like this at my workplace. We have a workload that is primarily single threaded and only utilizes SSE3 but we run multiple independent threads in parallel. The license model is 24 cores per license per computer, so it's in our favor to install the fastest MT processor that is a multiple of 24 cores.

Currently, we run the workload on Intel Xeon Scalable (1st Gen, so Skylake) but those processors are only 3.2 GHz all core boost.

 

[bsp2020]

Dude... We could totally use something like this at my workplace. We have a workload that is primarily single threaded and only utilizes SSE3 but we run multiple independent threads in parallel. The license model is 24 cores per license per computer, so it's in our favor to install the fastest MT processor that is a multiple of 24 cores.

Currently, we run the workload on Intel Xeon Scalable (1st Gen, so Skylake) but those processors are only 3.2 GHz all core boost.

The more you buy, the more you save.
Just buy it.

 

[Saylick]

The more you buy, the more you save.
Just buy it.

I'll see what I can do to convince the IT Department to upgrade. Unfortunately, since we're tied to Dell, I think it means our options are limited to Intel. Yes, I know that Dell offers EPYC on some of their server line, but I'm not sure the rest of the department will want to deviate from what they are used to. Being able to get some 48C Genoa would be sweet though. Worst case, we get Sapphire Rapids.

 

[DrMrLordX]

Worst case, we get Sapphire Rapids.

No, worst-case you get IceLake-SP since SPR is probably still low volume.

 

[biostud]

Well, sometimes the data is faulty.

https://twitter.com/x/status/1624740626406289408

 

[Saylick]

No, worst-case you get IceLake-SP since SPR is probably still low volume.

*Gags*

Yeah, I'd rather tell the IT Department to wait. Icelake-SP only offers an IPC increase but not much of a clock speed increase. Considering that we refresh our servers every 3 years, that's such a paltry upgrade given the amount of time that's elapsed, largely due to Intel dropping the ball, I know.

 

[Hitman928]

Well, sometimes the data is faulty.

https://twitter.com/x/status/1624740626406289408

He’s also overclocked and highly tuned his 13900k. Pretty sure he hasn’t done any such effort on his AMD system.

 

[BorisTheBlade82]

*Gags*

Yeah, I'd rather tell the IT Department to wait. Icelake-SP only offers an IPC increase but not much of a clock speed increase. Considering that we refresh our servers every 3 years, that's such a paltry upgrade given the amount of time that's elapsed, largely due to Intel dropping the ball, I know.

I am with you. Needed to order a new Server a couple of weeks ago. Still no AMD, still the Xeon 8280 Platinum is the Top Dog in our catalogue 😔

 

[DrMrLordX]

*Gags*

Yeah, I'd rather tell the IT Department to wait. Icelake-SP only offers an IPC increase but not much of a clock speed increase. Considering that we refresh our servers every 3 years, that's such a paltry upgrade given the amount of time that's elapsed, largely due to Intel dropping the ball, I know.

I am with you. Needed to order a new Server a couple of weeks ago. Still no AMD, still the Xeon 8280 Platinum is the Top Dog in our catalogue 😔

I weep for both of you (not joking. Well okay maybe I'm not quite moved to tears, but still, it IS sad).

 

[Saylick]

I weep for both of you (not joking. Well okay maybe I'm not quite moved to tears, but still, it IS sad).

But seriously. Just sad in general that the size of the company with the significantly inferior product is what's protecting it from losing market share. You'd think that in an industry where performance metrics matter that it would reward the technically superior product way faster, but nope. How can we expect technology to improve at the fastest rate possible when incumbents have such a big advantage?

 

[leoneazzurro]

Well, sometimes the data is faulty.

https://twitter.com/x/status/1624740626406289408

Considering the source, not surprising at all.

 

[Timmah!]

Well, sometimes the data is faulty.

https://twitter.com/x/status/1624740626406289408

So 7950x is 77 FPS instead of 65? On different motherboard? What the hell?
Then 12900 uses 6600 RAM and gets 5 FPS more, color me surprised game doing more FPS with faster RAM.

I have to admit i havent played the game myself, perhaps its pretty good gameplay-wise. But the visuals seem to be rather cartoony, i havent seen anything to warrant this kind of performance hog. Cyberpunk, Control, Metro...sure, with their kind of photorealistic approach, but this one, i dont see it. I realize you can bump-up even cartoony visuals to wannabe photoreal look with RT, but this does not seem to be the case? Or is it different when you actually play it?

 

[biostud]

So 7950x is 77 FPS instead of 65? On different motherboard? What the hell?
Then 12900 uses 6600 RAM and gets 5 FPS more, color me surprised game doing more FPS with faster RAM.

I have to admit i havent played the game myself, perhaps its pretty good gameplay-wise. But the visuals seem to be rather cartoony, i havent seen anything to warrant this kind of performance hog. Cyberpunk, Control, Metro...sure, with their kind of photorealistic approach, but this one, i dont see it. I realize you can bump-up even cartoony visuals to wannabe photoreal look with RT, but this does not seem to be the case? Or is it different when you actually play it?

I have no idea, the most recent triple A game I've played is Borderlands 3

 

[Geddagod]

The Zen 4c chiplet, according to AMD, is built on an HPC variant of TSMC N5. This aims at denser logic and denser cache, likely at the expense of high-end frequency. AMD says that this process offers 2x density, 2x power efficiency, and >1.25x silicon performance over the regular N7 it uses.

Quote from AnandTech, by Ian Cutress. This doesn't really make sense. Either he meant "built on an HD variant of TSMC N5" or his explanation is completely wrong. If we get a confirmation from Ian, it could be confirmation or not about whether Zen 4C uses denser libs than its Zen 4 counter part.

 

[Markfw]

Quote from AnandTech, by Ian Cutress. This doesn't really make sense. Either he meant "built on an HD variant of TSMC N5" or his explanation is completely wrong. If we get a confirmation from Ian, it could be confirmation or not about whether Zen 4C uses denser libs than its Zen 4 counter part.

I different quote from Ian said a different "mask".

 

[Geddagod]

I different quote from Ian said a different "mask".

idk. Ian sounds like a great guy, but sometimes he misquotes stuff that creates misleading info. Maybe someone shoot him a tweet or something about what he meant.

 

[Markfw]

idk. Ian sounds like a great guy, but sometimes he misquotes stuff that creates misleading info. Maybe someone shoot him a tweet or something about what he meant.

Here is the exact quote:
"This means that these two cores, while functionally identical, will use different sets of masks. The Zen 4c core has been redesigned to have the same functionality but offer more density combined with a different power/performance point on the voltage frequency curve. This is going to affect power consumption, efficiency, and likely the range of frequencies available. "

 

[Exist50]

I mean, obviously Zen 4c uses different masks. It also requires different base layers. That's just an observation of the obvious that Zen 4c isn't just an underclocked Zen 4.

 

[NostaSeronx]

The Papermaster quotes and the Charlie-summed comments were posted earlier in this thread:
"We’ll talk about Bergamo, our dense core that goes head-to-head with smaller arm cores where you just need throughput processing." - Mark Papermaster

"We are adding in first half of this year, what we call Bergamo, which will be with our Zen 4c, we increased staffing to our CPU team, and we added a version of Zen 4. It’s still Zen 4. It runs a code just like Genoa, but it’s half the size." - Mark Papermaster

"And that competes head-to-head with Graviton and ARM-based solutions where you don’t need the peak frequency. You’re running workloads like Java workloads, throughput workloads that don’t have to run peak frequency, but you need a lot of cores" - Mark Papermaster

"Bergamo can run in non-SMT mode, which helps with per-thread performance. On a thread vs thread basis, 128C/128T Bergamo is about 60% more performant than 64C/128T Milan." - Charlie Summerized1

"That jaw stuff comes when you turn off threading or what AMD calls SMT Off mode. In this mode, Bergamo has 2x the cores as Milan but the same thread count, not exactly impressive. That said AMD claims it will beat a 64C/128T 7763 by *60%*! Let me repeat that, a Bergamo running in SMT off mode will have the same thread count as the current top Milan but be overall 60% faster." - Charlie Summerized2

 

[nicalandia]

When Phoronix posted the AVX512 On/Off comparison between The 60C/120T 8490H and the 192C/384T Genoa. I believed they should have included the Lower Core 64C/128T 9564 CPU.

I put together that comparison and also included Genoa 9654 with only 6 Channel Memory.

This is the Geometric Mean.



Genoa vs Sapphire Rapids AVX512 Comparison

 

[Exist50]

"Bergamo can run in non-SMT mode, which helps with per-thread performance. On a thread vs thread basis, 128C/128T Bergamo is about 60% more performant than 64C/128T Milan." - Charlie Summerized1

"That jaw stuff comes when you turn off threading or what AMD calls SMT Off mode. In this mode, Bergamo has 2x the cores as Milan but the same thread count, not exactly impressive. That said AMD claims it will beat a 64C/128T 7763 by *60%*! Let me repeat that, a Bergamo running in SMT off mode will have the same thread count as the current top Milan but be overall 60% faster." - Charlie Summerized2

And this is a good example of the current issue with SMT. If you turn it on for the throughput, your per thread performance falls drastically. But leave it off, and something like Bergamo can still deliver great throughput with respectable per-thread performance. Ironically, Bergamo with SMT disabled is almost certainly faster than Genoa with it left on, per thread.

 

[Markfw]

And this is a good example of the current issue with SMT. If you turn it on for the throughput, your per thread performance falls drastically. But leave it off, and something like Bergamo can still deliver great throughput with respectable per-thread performance. Ironically, Bergamo with SMT disabled is almost certainly faster than Genoa with it left on, per thread.

That really depends on the workload. Wait until benchmarks come out. AMD is not stupid enough to make 2 products where one always wins in the same area and price bracket. I.E. Servers One is for the cloud, and the other for servers that have a different type workload.