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]

So Nobody Will Acknowledge that I was Right About AVX512? So far Geekbench5 seems to like it because the Crypto performance is Off The Chart.

 

[Det0x]

Check ir.amd.com. I'm on mobile right now.

Thanks i found it

 

[Shamrock]

The endnotes not being part of the video is pretty stupid. Will I still get to all the fact checking before going to sleep?

Just goto the website. It has footnotes, that ALL PCs (including the Intel) were using DDR5-6000 CL30

https://www.amd.com/en/processors/ryzen

Testing as of 15 August, 2022, by AMD Performance Labs using the following hardware: AMD Socket AM5 Reference Motherboard with AMD Ryzen™ 9 7950X, Ryzen™ 9 7900X, Ryzen™ 5 7600X and G.Skill DDR5-6000C30 (F5-6000J3038F16GX2-TZ5N) with AMD EXPO™; versus AMD Socket AM4 Reference Motherboard with Ryzen™ 9 5950X, Ryzen™ 9 5900X, Ryzen™ 5 5600X; versus ROG Maximus Z690 Hero with Core i9-12900K and G.Skill DDR5-6000C30 (F5-6000J3038F16GX2-TZ5N) with AMD EXPO™ loaded. ALL SYSTEMS configured with NXZT Kraken X63, open air test bench, Radeon™ RX 6950XT (driver 22.7.1 Optional), Windows® 11 22000.856, AMD Smart Access Memory/PCIe® Resizable Base Address Register (“ReBAR”) ON, Virtualization-Based Security (VBS) OFF. All games tested at 1920x1080 with HIGH in-game preset and the chronologically newest graphics industry API available within the game’s rendering engine (e.g. Vulkan® over OpenGL™, DirectX® 12 over DirectX® 11). Results may vary. RPL-007

 

[Markfw]

Zen 4 mobile will DESTROY Intel in efficiency and AMD laptops will have great battery life

my 6800u will be an example (when it ships)

 

[Saylick]

Zen 4 mobile will DESTROY Intel in efficiency and AMD laptops will have great battery life

Yeah, that "up to 75%" perf/W gain at 65W is bonkers. What's it going to be at 35W? Double the efficiency?


I don't know how Intel can compete in the mobile space without at least node parity.

 

[inf64]

The only gripe I have with the endnotes is no power consumption numbers. Maybe they will add it later?

 

[shady28]

Just goto the website. It has footnotes, that ALL PCs (including the Intel) were using DDR5-6000 CL30

https://www.amd.com/en/processors/ryzen

Why would they use AMD EXPO memory timings on Alder Lake?

 

[desrever]

Yeah, that "up to 75%" perf/W gain at 65W is bonkers. What's it going to be at 35W? Double the efficiency?
View attachment 66780

I don't know how Intel can compete in the mobile space without at least node parity.

Dragon range will bring that to notebooks. Will be a slaughter against w/e intel can do.

 

[Joe NYC]

Zen 5 confirmed 3N+4N

Interesting, does that mean Zen 5 on N4 and Zen 5c on N3?

 

[Saylick]

Dragon range will bring that to notebooks. Will be a slaughter against w/e intel can do.

I think I know what they'll do...

Spoiler

 

[tamz_msc]

Why would they use AMD EXPO memory timings on Alder Lake?

Yeah, I hope that is a typo otherwise it would seem weird.

 

[moinmoin]

The only gripe I have with the endnotes is no power consumption numbers. Maybe they will add it later?

In an ideal world every single benchmark would list joules consumed along everything else.

 

[FangBLade]

I must quote our friend reb0rn here:



As can be seen, you sir have failed. Better luck next time, and I suggest that you use your brain .

I guess if he starts to use brain then his nick will have real meaning.

 

[Abwx]

The only gripe I have with the endnotes is no power consumption numbers. Maybe they will add it later?

It s likely that the CB R23 and such benches were done at close to 230W, 47% better perf/watt than a 12900K is not much, and close to 10% higher score than what was announced cant be had for free.

They surely wanted to make sure that the competition novelty would be DOA and couldnt claim any possible crown using the same TDP trick as ADL.

 

[Joe NYC]

Where did he say "end of 2024" ?

Papermaster said "on track for 2024"

 

[poke01]

We can finally have a proper node comparison with the TSMC 5nm family of nodes.

Ok the Ryzen 5 7600X scores around 2175 in Geekbench 5. M2 is around 1930 single core.

The base freq of 7600X is 4.7Ghz with a turbo of 5.3Ghz and M2 is 3.5Ghz. I would say Apple is still in the lead in PPW.
Man, Apple has really wide designs.

The 7600X is 13% faster in ST(Geekbench only) vs M2.

Intel won't even catch up to AMD in PPW until Lunar Lake.

 

[Zucker2k]

It s likely that the CB R23 and such benches were done at close to 230W, 47% better perf/watt than a 12900K is not much, and close to 10% higher score than what was announced cant be had for free.

They surely wanted to make sure that the competition novelty would be DOA and couldnt claim any possible crown using the same TDP trick as ADL.

Well, at least now we're speaking the same language. Let's also hope the particular CB R23 bench is the very short standard test, and not some looping that'll force ADL to revert to 125w.

 

[Joe NYC]

Same little birdie told me Nov 15th for Rdna 3 We'll see.

Did the bridie say anything about the V-Cache launch?

 

[Abwx]

Well, at least now we're speaking the same language. Let's also hope the particular CB R23 bench is the very short standard test, and not some looping that'll force ADL to revert to 125w.

According to the 47% perf/watt advantage the 12900K was benched at 240W, the CPU hold this level during 56s according to Computerbase, that s way enough for a CB run.

 

[desrever]

We can finally have a proper node comparison with the TSMC 5nm family of nodes.

Ok the Ryzen 5 7600X scores around 2175 in Geekbench 5. M2 is around 1930 single core.

The base freq of 7600X is 4.7Ghz with a turbo of 5.3Ghz and M2 is 3.5Ghz. I would say Apple is still in the lead in PPW.
Man, Apple has really wide designs.

The 7600X is 13% faster in ST(Geekbench only) vs M2.

Intel won't even catch up to AMD in PPW until Lunar Lake.

You are comparing a desktop CPU to a mobile one. The 7800U or w/e would make a much better comparison to the M1/M2. The IO die + chiplets approach costs power and latency.

 

[poke01]

You are comparing a desktop CPU to a mobile one. The 7800U or w/e would make a much better comparison to the M1/M2. The IO die + chiplets approach costs power and latency.

yeah true. Well that time won't come till next year.

 

[FangBLade]

I'm curious, where is Henry Swagger? he is inactive for 1 month. I'm still waiting to hear his optimistic comments about Intel revenues, Zen 4 performance etc.

 

[Markfw]

According to the 47% perf/watt advantage the 12900K was benched at 240W, the CPU hold this level during 56s according to Computerbase, that s way enough for a CB run.

Actually, I would like to see sustained runs of ANY benchmark that is stressful, for 10 minutes or more. I want to see sustained performance and efficiency. This is just a short run.

 

[Kaluan]

Games love high clk

IPC chart for 8c/16t@4GHz, how high those architectures clock is irrelevant there.
Also, clocks obviously don't scale linearly in games.

More importantly, ~35% overall performance (as indicated in the slide) over Zen 3 @170w @Abwx

Also:
- ~15% average gaming performance over Zen 3 (which other chip is 15% average over Zen 3? 5800x3d, which happens to trade blows with ADL + DDR5). I'm optimistic for RPL. No wonder the X3D is looking to make an almost immediate showing this time around.
- 170w TDP/230w PPT (to allow Zen 4 chips to leg room to do their thing). So, the chip power consumption debate should end tonight.

Their comparison are vs SKUs, not architectures. 5800X3D (+15%) was showcased vs 5800X, 7600X (+5%) vs 12900K (DDR5-6000CL30 on both) and 7950X (+15%) vs 5950X. It would be a lie to say the 2-5% 5950X has over 5800X are statistically irrelevant. It's pretty obvious they want to imply Zen4 SKUs edge out 5800X3D.

 

[Vope45]

IPC chart 8c/16t@4GHz, how high those architectures clock is irrelevant there.
Also, clocks obviously don't scale linearly in games.

Their comparison are vs SKUs, not architectures. 5800X3D (+15%) was showcased vs 5800X, 7600X (+5%) vs 12900K (DDR5-6000CL30 on both) and 7950X (+15%) vs 5950X. It would be a lie to say the 2-5% 5950X has over 5800X are statistically irrelevant. It's pretty obvious they want to imply Zen4 SKUs edge out 5800X3D.

You're right. .What interests me is that amd chose to include more games this in the geo mean ipc calculation. We'll see if the actual ipc uplift is 13% in a wide range of apps and games, which I hope it should be.