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

Vattila · Oct 6, 2019

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!

eek2121 · Jun 10, 2022

So uh…SMT4 on Zen 5? 😉

Abwx · Jun 10, 2022

Karnak said:
Do you know what you're talking about? Or you know what AMD did use to make that statement?

Pretty sure it'll be something like a 64C Milan EPYC vs. an upcoming 96C Genoa EPYC. But without knowing it's pointless. ofc it's not the process itself. For that we know everything directly from TSMC already.

The slide is not about the products, it s explicitely about their custom N5 process performance, it state 2x the efficency or 1.25x the performance, either it is right or if not they can be sued by their stock holders at the rate of bns..

This mean that a 5nm shrinked 5950X would consume half the power or be clocked 25% higher at same power.

For Zen 4 this mean that at same TDP it can be theorically clocked 20% higher and that would put a 142W PPT 7950X at 4.8GHz all cores in Cinebench since a stock 5950X run this bench at 4GHz, eventually 170W PPT to have some margin but certainly not 230W.

lobz · Jun 10, 2022

eek2121 said:
So uh…SMT4 on Zen 5? 😉

I'm begging you. If we were ever to discuss SMT4(+...) again on an AMD architecture, please let's make a completely separate thread for it, ideally on the dark web or in another galaxy.

DrMrLordX · Jun 10, 2022

nicalandia said:
The entire Zen4 stack is design to compete favorably against ADL/RPL and MTL.

See comment below by turtile, he beat me to it!

JasonLD said:
Now I wish AMD does it with RDNA3 or future architecture and make 300W+ Big APU. It will be a great high-end system with small footprint like M1 Ultra system.

Apple has demonstrated that they can sell a "big APU" (more like "big SoC") to the fans of their software ecosystem. It's not entirely clear that AMD could do the same thing since they don't have control over the software stack they'll be running. There were rumours of enterprise-class APUs that never came to fruition, so odds of them going in that direction seem low.

turtile said:
All that matters for AMD is the efficiency of the core for servers.

Boom. You win a cookie. Raphael is an overwrought afterthought. Don't get me wrong! I'll be buying one, perhaps a 7950X this time around . . . but I still know that I'm buying scraps.

eek2121 said:
So uh…SMT4 on Zen 5? 😉

Parents can be so cruel.

Timmah! · Jun 10, 2022

All of this new stuff is cool, but it has one caveat - everything, i would be interested the most (Zen 4 Threadripper, V-cache products, maybe Ryzens with more than 16 cores), will seemingly come next year. The problem is, i wanted to build new rig this year, alongside RTX40
Now Zen 4 /7950x actually looks good, i am actually swayed by it more than i thought i would be...but i just cant shake the feeling this will be 2016 all over again. That year i built x99 machine with i7 6850k, only for it to become obsolete less than year later by first Threadrippers/Skylake-X. And i regretted not waiting. If i go 16C Zen4 route this year, i fear i might regret again.

Ajay · Jun 10, 2022

uzzi38 said:
Btw, I don't think it was mentionned here, but in the Q&A AMD clarified that both N4 and N3 Zen 5 based products will launch by the end of 2024. Also: "You'll see a diversity in nodes for most products going forwards".

External chiplets with UCIe is looking like a 2025 deal at best, really needs a second generation standard.

Just a couple of interesting snippets that have been overlooked.

Thanks. Thats very interesting news. I’ll have to stop myself from ruminating on the possible product line up between N3 and N4. Off to look for that Q&A. Also, more node diversity == more wafer starts available per quarter. Good news!

nicalandia · Jun 10, 2022

This is a Solid Zen4 Stack AMD is building here for EPYC, Intel Sapphire Rapids and Emerald Rapids have a daunting task ahead.

Just look at this "All Star" Line Up.

Genoa:

Genoa-X:

Bergamo:

AMD EPYC Siena

RnR_au · Jun 10, 2022

Poor Intel...

nicalandia · Jun 10, 2022

RnR_au said:
Poor Intel...

Just Updated my post.

uzzi38 · Jun 10, 2022

Ajay said:
Thanks. Thats very interesting news. I’ll have to stop myself from ruminating on the possible product line up between N3 and N4. Off to look for that Q&A. Also, more node diversity == more wafer starts available per quarter. Good news!

I don't think there's a write up of the Q&A, so you'll have to just open up the webcast again and watch it.

Ajay · Jun 10, 2022

Hmm, going over one of Ryan's write ups...

From AMD: 8-10% IPC gains + clock gains (~ 10%). That equals an aggregate 18-20% performance uplift. That’s nothing to sneeze at. And, MT will be even higher. Why is everyone so salty. Mike Clark and Papermaster already pointed out that they were especially excited about Zen5. Everything seems in order to me.

eek2121 · Jun 10, 2022

lobz said:
I'm begging you. If we were ever to discuss SMT4(+...) again on an AMD architecture, please let's make a completely separate thread for it, ideally on the dark web or in another galaxy.

You know I just had to! 😉

DrMrLordX said:
Apple has demonstrated that they can sell a "big APU" (more like "big SoC") to the fans of their software ecosystem. It's not entirely clear that AMD could do the same thing since they don't have control over the software stack they'll be running. There were rumours of enterprise-class APUs that never came to fruition, so odds of them going in that direction seem low.

Apples and oranges. You cannot currently install a GPU in an ARM based Mac, except possibly an external one.

That being said, if AMD wanted, they could definitely put a beefy GPU on a Ryzen product.

trexfromouterspace · Jun 10, 2022

Stop trying to make SMT4 a thing, it's not happening. And for good reasons too.

gdansk · Jun 10, 2022

Oh so they're skipping to SMT8? Dare I say

Timmah! · Jun 10, 2022

So is Siena going to be the base for that Zen4 Threadripper then?

nicalandia · Jun 10, 2022

Timmah! said:
So is Siena going to be the base for that Zen4 Threadripper then?

No, ThreadRipper will be bussy ripping appart Sapphire Rapids-X HEDT with it's 98C/196T....

AMD Confirms Ryzen Threadripper 7000 HEDT CPUs With Zen 4 Core Architecture In 2023

AMD's Ryzen Threadripper 7000 HEDT CPUs have been confirmed to launch in 2023 with the latest Zen 4 core architecture.

wccftech.com

Poor Sapphire Rapids-X....

rtxtwt · Jun 10, 2022

Det0x said:
More singlethread napkin performance math

ZEN 3:
CB: +13% IPC.
Average: +19% IPC.
Clock Speed: +4.5%.

Average ST Zen3 performance increase over Zen2 =
1 * 1.19 * 1.045 = a ~24% ST performance uplift

ZEN 4:
CB: +8-10% IPC.
Average: +11.5-14.5% IPC.
Clock Speed: +12.5-19%.

Average ST Zen4 performance increase over Zen3 =
1 *1.115 * 1.125 = a ~25% ST performance uplift on the low-end of the scale
1 *1.145 * 1.19 = a ~32% ST performance uplift on the high-end of the scale

Multithread performance increase is likely higher, hence the "greater than + 35% overall performance"

We'd better cautious if we start calculating performance by using PPT slide numbers especially manufacturers cherry pick results and with ambiguous claim.

a good example is Zen3 vs Zen2, normalize the result you can see the average "IPC" difference is only slightly above 10%, some uplift is as miserable as 1-2% or close to none, and different version of same software could vary with a huge gap like Cinebench R15/R20

lobz · Jun 10, 2022

Ajay said:
Hmm, going over one of Ryan's write ups...

From AMD: 8-10% IPC gains + clock gains (~ 10%). That equals an aggregate 18-20% performance uplift. That’s nothing to sneeze at. And, MT will be even higher. Why is everyone so salty. Mike Clark and Papermaster already pointed out that they were especially excited about Zen5. Everything seems in order to me.

there will be plenty of sneezing at launch day 😊😊

igor_kavinski · Jun 10, 2022

cortexa99 said:
View attachment 62890

Which website is this from?

Vattila · Jun 10, 2022

gdansk said:
Oh so they're skipping to SMT8?

Well, they are really going for density and thread count with "Bergamo" to fend off ARM.

Has anyone else noticed that "Bergamo" is an anagram of "bego ARM"?

bego

Definition, Synonyms, Translations of bego by The Free Dictionary

www.thefreedictionary.com

biostud · Jun 10, 2022

eek2121 said:
You know I just had to! 😉

Apples and oranges. You cannot currently install a GPU in an ARM based Mac, except possibly an external one.

That being said, if AMD wanted, they could definitely put a beefy GPU on a Ryzen product.

Isn't their big APU what powers Xbox and PS?

biostud · Jun 10, 2022

uzzi38 said:
I wouldn't count on Zen4C coming to desktop. Not ruling it out entirely, but I'd say 95% chance it doesn't happen.

After looking at the slides again, I think you're right, it only mentions zen4 in the 7000 series, not 4c. But let's see what Intel brings to the table. The good thing for AMD is they have a flexible product stack.

Ajay · Jun 10, 2022

eek2121 said:
You know I just had to! 😉

Apples and oranges. You cannot currently install a GPU in an ARM based Mac, except possibly an external one.

That being said, if AMD wanted, they could definitely put a beefy GPU on a Ryzen product.

Can't use external GPUs with M-series Macs. Not sure if that will ever change.

Anhiel · Jun 10, 2022

Looking at the slides for Zen4... with 8-10% IPC AMD basically confirmed it being in the same range as Rembrandt. So it doesn't seem a lot got improved here. Updating my graph based on AMD's publications and filling the still non-published (cherry picked) data points...
Desktop perf/Watt is exactly at 3x (+25% perf/Watt statement). The upper limit would be below 5x for mobile (referring to Rembrandt). AMD might push it up here a bit for mobile but I expect it to be marginal and insignificant. In any case the changes nearly exactly match my old projections I posted last time. So things look legit.
In red are my calculated numbers which likely will differ somewhat from what AMD will be publishing but shouldn't be too far off.

Anyhow, since Zen5 will have a wider front end. It looks like it'll be staying ahead of the curve.

nicalandia · Jun 10, 2022

Anhiel said:
Looking at the slides for Zen4... with 8-10% IPC AMD basically confirmed it being in the same range as Rembrandt.

Rembrandt it's getting a performance boost of about 7-10% Combined with IPC and speed.... Zen4 it's getting about 25-30% performance boost combined.....

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

Senior member

Platinum Member

Lifer

Platinum Member

Lifer

Golden Member

Lifer

Diamond Member

Golden Member

Diamond Member

Platinum Member

Lifer

Platinum Member

Member

Platinum Member

Golden Member

Diamond Member

Senior member

Platinum Member

Lifer

Senior member

Lifer

Lifer

Lifer

Member

Attachments

Diamond Member