Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 200 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Vattila

Senior member
Oct 22, 2004
799
1,351
136
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).

Untitled2.png


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! :)
 
Last edited:
  • Like
Reactions: richardllewis_01

SteveGrabowski

Diamond Member
Oct 20, 2014
6,764
5,725
136
That was mostly the basis of my claim when I said Intel is better for emulation.

Apparently the extra Zen3D cache helps for Yuzu but not for Ryujinx, but Switch emulation and common games should run ok in any modern CPU, so they should not guide your purchasing decision.

But RPCS3 games like The Last of Us won't always be above 30fps even in a 12900k, so that's what's important to consider if you plan to use the emulator a lot.

Can you define running ok? I have a real Switch but the 30 fps and 720p resolution feel pretty rough to play for games like Shin Megami Tensei V, Xenoblade Chronicles 2, and Dragon Quest XI S for example, and I was hoping to play them at 40-60 fps (which is my monitor's FreeSync range) and 1440p resolution on PC instead. Switch emulation is way more important to me than PS3 emulation. Do you think Yuzu on a midrange cpu like 12400F / 5600X is usually a better than original hardware experience?
 

eek2121

Platinum Member
Aug 2, 2005
2,883
3,859
136
Can you define running ok? I have a real Switch but the 30 fps and 720p resolution feel pretty rough to play for games like Shin Megami Tensei V, Xenoblade Chronicles 2, and Dragon Quest XI S for example, and I was hoping to play them at 40-60 fps (which is my monitor's FreeSync range) and 1440p resolution on PC instead. Switch emulation is way more important to me than PS3 emulation. Do you think Yuzu on a midrange cpu like 12400F / 5600X is usually a better than original hardware experience?

Most of the switch games I've played play at 60fps, probably more if there is an FPS unlock. I haven't done it in a while, however. I actually have an oled switch, so I usually end up using that.
 

jpiniero

Lifer
Oct 1, 2010
14,487
5,150
136

Videocardz apparently got one of the slides from tomorrow... only room for 2 chiplets and release is fall. Says "Over 15% uplift" in single thread but Videocardz don't know what it is comparing to.
 

eek2121

Platinum Member
Aug 2, 2005
2,883
3,859
136

Videocardz apparently got one of the slides from tomorrow... only room for 2 chiplets and release is fall. Says "Over 15% uplift" in single thread but Videocardz don't know what it is comparing to.

Likely over Zen 3. Trends with what rumors have reported recently: 15-24% uplift, 5+ ghz boost clocks. An absolute monster. If what is being said about multicore frequencies is true, I'm going to have to upgrade from my 5950x...😭

EDIT: 6nm IO die (with GPU) seemingly confirmed.

Based on the slides, my suspicion is we may not get concrete models listed tomorrow. I hope I'm wrong. I also hope the rest of the slides leak so I don't have to stay up until 1am.
 
Last edited:

leoneazzurro

Senior member
Jul 26, 2016
901
1,426
136
Over 15% in ST seems a bit low if it's compared to Zen3, especially if we are looking at frequency uplift of 10% (or more). With a lot more transistors, it seems strange. It will also depend on how that ">15%" is calculated.
 

eek2121

Platinum Member
Aug 2, 2005
2,883
3,859
136
Over 15% in ST seems a bit low if it's compared to Zen3, especially if we are looking at frequency uplift of 10% (or more). With a lot more transistors, it seems strange. It will also depend on how that ">15%" is calculated.

It's going to depend on workload. AMD focused a lot on FPU performance (adding AVX-512, etc.) and doubling L2 cache this time around. It is likely some workloads will be significantly faster, and others won't.

It also appears these chips are launching this fall, possibly September (if rumors are accurate). This may be to give OEMs time to get back to school models ready. It also launches close to Rocket Lake, so I suspect AMD is going to keep relatively quiet about performance (beyond a couple of workloads) until the chips are almost ready to drop. I could be wrong, however.
 

Panino Manino

Senior member
Jan 28, 2017
813
1,010
136
Zen 4 is going to outclass Intel chips in any emulation workload. Calling it now. Not that Zen 3 was lacking, however. As a Dolphin and Yuzu user, I can tell you that you heard wrong. Also, Emulators are becoming increasingly optimized for AMD thanks to stuff like the Steam Deck.

Are you really sure?
Zen 3 can't even match older Intel skylake refreshes and Ice Lake, I wouldn't be so sure that Zen 4 will be such a winner.


Videocardz apparently got one of the slides from tomorrow... only room for 2 chiplets and release is fall. Says "Over 15% uplift" in single thread but Videocardz don't know what it is comparing to.

"Advanced low power architecture"?
So this means that AMD is not aiming for the maximum performance possible and is aiming to keep the "performance per watt" advantage? Will leave to compete on sucking power like Intel to later?
 
Last edited:

eek2121

Platinum Member
Aug 2, 2005
2,883
3,859
136

This is surprisingly bad, or could it be faked from AMD to prevent early leaks?

How is it surprisingly bad? This puts AMD ahead of Alder Lake, even in multicore performance. It also says > 15%, not 15%. The rumors are 15-24%. Take a look at this: CPU 2021 Benchmarks - Compare Products on AnandTech and add (or subtract) 15%.

Are you really sure?
Zen 3 can't even match older Intel skylake refreshes and Ice Lake, I wouldn't be so sure that Zen 4 will be such a winner.

Which emulator? I

CPU 2021 Benchmarks - Compare Products on AnandTech has a dolphin test. The 5950x is about 15% behind the 12900k. For other emulators, Zen 4 has a number of changes that will help with emulation, such as the increase in L2 and some floating-point improvements. Also, unless the emulator is open source, there is no way to know if it is an issue with the way the emulator is written (optimizations for Intel, for example) vs. an actual issue with the chip. I'm a developer by trade, have written (older) emulators (also I've contributed to certain open-source operating systems and drivers), and I can tell you that certain design decisions can absolutely make or break performance on a given architecture.
 

deasd

Senior member
Dec 31, 2013
508
711
136
'5Ghz+' is already a sandbaging if the processor is running towards 5.4Ghz+ though......

Over 15% in ST seems a bit low if it's compared to Zen3, especially if we are looking at frequency uplift of 10% (or more). With a lot more transistors, it seems strange. It will also depend on how that ">15%" is calculated.

It would be almost no IPC uplift if it's just +15% ST with ~5.4Ghz+..... strange

edit: ouch, well, it's OVER 15%..... and if it's launch this fall I think it's still too early to see any real performance number during Computex though.
 
Last edited:

Asterox

Golden Member
May 15, 2012
1,026
1,775
136

This is surprisingly bad, or could it be faked from AMD to prevent early leaks?

Why, maybe because someone mentioned 25% higher IPC somewhere in the woods? :laughing:

In reality Zen 4 is pretty bad, even old Athlon II X2 240 has 1mb L2 Cache per core.:grinning:


The fact that now every Zen 4 CPU has a VCN hardwer, it is a very welcome and very useful hardware.

 

coercitiv

Diamond Member
Jan 24, 2014
6,137
11,615
136
How is it surprisingly bad?
He's reading the vague wording as ST performance uplift. You're looking at both bullet-points and conclude the first must mean IPC uplift, while the second indicates frequency uplift.

Problem is the fragment of that slide is not enough to attribute clear meaning to the numbers. ST performance uplift is still anywhere from 15% to 35%. The only difference is VCZ can now claim to have a leak on Zen4, only they don't know what the leak means :tearsofjoy:
 

mikk

Diamond Member
May 15, 2012
4,108
2,099
136
How is it surprisingly bad? This puts AMD ahead of Alder Lake, even in multicore performance. It also says > 15%, not 15%. The rumors are 15-24%. Take a look at this: CPU 2021 Benchmarks - Compare Products on AnandTech and add (or subtract) 15%.


Because the expectations were much higher like 20% more IPC (if not more) and on top of this a higher ST clock speed. MLID claimed 28-37% improved ST.

They don't say over 15% on a marketing slide if it's more than 20% or 25%, you have to be realistic. Also ADL-S is roughly 20% faster in ST, it isn't ahead with a little more than 15%.
 
  • Like
Reactions: Tlh97

exquisitechar

Senior member
Apr 18, 2017
654
859
136
Significantly more than 20% IPC uplift was always unlikely, we know Zen 4 is not the next significant core update. However, 15% or so ST performance uplift with a rumored 400MHz clock speed increase would be genuinely terrible and would make me worry about AMD's ability to compete in the near future. That wouldn't even be enough to match Golden Cove. Something went majorly wrong if that's the case, how were all those transistors even put to use?
 
  • Haha
  • Like
Reactions: ftt and Thibsie

Timmah!

Golden Member
Jul 24, 2010
1,389
596
136
Wait, no V-cache by default?
No 24 cores is bummer, but not surprised, it was unfounded rumor. Anyway, no Zen 4 for me it seems.

Intel, you are my only hope :-D
 

CakeMonster

Golden Member
Nov 22, 2012
1,382
474
136
Neither V-Cache nor 24c was very likely.

I suspect 5nm and clock increases would give more than 15% on average, so this number might either indicate the clock increases are less, or simply that they made some other priorities that are yet unknown (except L2 and AVX512).
 

Hougy

Member
Jan 13, 2021
77
60
61
Can you define running ok? I have a real Switch but the 30 fps and 720p resolution feel pretty rough to play for games like Shin Megami Tensei V, Xenoblade Chronicles 2, and Dragon Quest XI S for example, and I was hoping to play them at 40-60 fps (which is my monitor's FreeSync range) and 1440p resolution on PC instead. Switch emulation is way more important to me than PS3 emulation. Do you think Yuzu on a midrange cpu like 12400F / 5600X is usually a better than original hardware experience?
By "run ok" I mean have a framerate at least as good as on the real Switch. If you want to emulate a 30fps game at twice that framarate then you might need an insanely powerful CPU. I'm not up too up to date on Switch emulation. Check Youtube videos of people running the games you're targeting on different CPUs or ask people in the emulator communities.
 

yuri69

Senior member
Jul 16, 2013
366
561
136
So it is "single-thread" not "IPC" uplift now. The previous Zen gens were reported as average IPC uplifts.

Given the 5.4GHz reports it's a lower gen-to-gen IPC gain than Zen 2's. Did AVX512 waste the trn budget?

So far it seems pretty good for Intel since Zen 4 is not a monster - it just scales better in servers.
 

Hougy

Member
Jan 13, 2021
77
60
61
Will the integrated GPU be on the IO die or the compute dies? I'm guessing the IO die because the compute dies will probably also be used for servers, that don't need the GPUs, and RDNA2 can't be divided in multiple dies, it would have to be RDNA3