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]

Yep, kinda like in a recent 5600, 5700x.
My office 5700g has similar cores.

You are not going to get any Quality CCD from a 7600X, at that stage you are really reaching almost bottom of the barrel quality.

Top Quality goes to Top of the Line Genoa and 7950X.

 

[Det0x]

AMD Ryzen 7700X/7600X rumored Cinebench R23 scores are in

Greymon claims that in single-core test, Ryzen 5 7600X scores 19XX points, while Ryzen 7 7700X up to 20XX range. Multicore benchmark result were not revealed. If only ST cores were to be considered, 7600X would score around Core i7-12700K, while 7700X reaches i9-12900K performance. But again, those are only single-core tests, which may go higher or lower as AMD implements further optimizations and leakers get their hands on final silicon.

https://twitter.com/x/status/1562751802059399168

Those wondering why we have not seen CPU-Z leaks featuring Ryzen 7000 CPUs yet, is actually simple. In its current state, Ryzen CPUs reportedly have problem fishing the full benchmark. This might be solved soon as AMD releases an updated BIOS, however for now there are CPU-Z scores available.
AMD 7000-series CPUs either do not run the benchmark in full (stuck at 80%) or show lower performance gains compared to Cinebench. In any case, it is probably better to wait for someone to verify this issue with the latest BIOS or using different samples.

https://twitter.com/x/status/1562751758673137664

Interestingly, there are no benchmark results for Ryzen 9 series 7900X/7950X yet, however the mid-range series are already in many hands. Some of those CPUs even ended up on auction sites.
AMD will unveil the Ryzen 7000 series next week. The company will most likely end the speculation by sharing official performance figures and hopefully prices for each of the four expected SKUs.

 

[Asterox]

I see no way that AMD can ask the same price for 7700X (16T) as intel is asking for 13700K, the performance will not be there.

Yes it can, especially if they have the same gaming performance + R7 has significantly lower consumption.Now every Zen 4 CPU has iGPU/VCN hardware, which can be twice as fast compared to classic CPU only encoding.

Intel main problem is Desktop platform, or 1700 socket is dead no more future CPU upgrades.

- if you already have Alder Lake CPU, hm upgrade to Raptor lake it doesn't make any sense

- if you buy an Intel 1700 socket today+ i7 13700K as an example, the buyer must be aware of the fact that this is the last cpu series for that socket

AM5 is new, and you can expect at least three cpu generations.

Red facts are very problematic for intel, and that's exactly why as a new PC(if you have money)people will prefer AM5 socket no doubt.

We all now, "what happened to the unfortunate people who voluntarily chose AM4 socket five years ago".

 

[nicalandia]

Those wondering why we have not seen CPU-Z leaks featuring Ryzen 7000 CPUs yet, is actually simple. In its current state, Ryzen CPUs reportedly have problem fishing the full benchmark. This might be solved soon as AMD releases an updated BIOS, however for now there are CPU-Z scores available.
AMD 7000-series CPUs either do not run the benchmark in full (stuck at 80%) or show lower performance gains compared to Cinebench. In any case, it is probably better to wait for someone to verify this issue with the latest BIOS or using different samples.

Interesting. Genoa showed very good performance on CPU-Z Standard and AVX512 benchmarks

 

[Zucker2k]

A

That s not at isopower, the CPU provide >35% better perf at a given power such that it has >25% better perf/watt than a 5950X.

At equal perf/watt it would consume 1.35 x 142 = 191.7W

But since it has 25% better perf/watt it consume 191.7/1.25 = 153.36W.

Or,
It'll roughly consume 106w at the same perf of Zen 3 at 142w.
Or,
Zen 3 would roughly consume 210w to equal perf of Zen 4 at 170w.

Who knows how AMD derived this 35% "overall performance" gain for Zen 4?

Edited for clarity.

 

[Abwx]

A

Or,
It'll roughly consume 106w at the same perf of Zen 3 at 142w.
Or,
Zen 3 would roughly consume 210w to equal perf of Zen 4 at 170w

Who knows which of the above translates into 35% "overall performance" gain for Zen 4?

Better to stick with the litteral sentences from AMD, they are clear on the conditions and they match all others deductions that can be made from previous technical tips, that is, 0.5x the power at isofrequency and 1.25x the frequency at isopower for the N5P process.

The fact that they give both perf and perf/watt improvement makes no doubt about the conclusions, the chip perform at least 1.35x better and its comsumption is 1.35/1.25 = 1.08 x higher at most.

 

[Kaluan]

ANY benchmark that makes results public will have this issue. It isn’t the benchmark that is the problem, but rather the users.

Want me to show my CBr23 results while doing a blender render?

Sorry, but nothing is as bad as GB at this, the fact that they don't curate results and jist auto-publish everything IS part of the criticism of it, not part it's excuses. So that's a bit backwards.

Also you didn't adress why GB5 shows 5800X as 10-20% slower than 12600K when meta results from reputable review outlets place it ~5% in front of the i5. Or why anyone with half a sense would ever think a 30s benchmark can be a reliable metric of sustained performance in anything (the results are even more misleading for mobile devices)

I don't get, why 7600X and not 7900X?
There are 12 clocks in that tweet - hence 12 cores. Fmax for the 7900X rumored to be 5.6Ghz.
If proper cooling applied it very well could hit Fmax at all cores, being unrestricted on PPT/TDC

I think those are advertised boost clocks. Fmax for 7900X is likely 5,75GHz (+150MHz), if Ryzen 5000 logic still applies.

So that's about what Alder Lake does. Wouldn't that be a little underwhelming, esp at similar clock speeds?

No? For starters, how would lower/mid range Ryzen 7000 scoring like 12th gen i9s be underwhelming? Secondly, it's likely CB and other memory/cache agnostic workloads will not be Zen4's strong point. Other workloads might fare a lot better.

Not sure what you meant by similar clocks tho.

 

[xilli_fiberbit]

There are a new AGESA 1.0.0.1H :

https://twitter.com/x/status/1562774288729378820

 

[bakyt115]

https://twitter.com/x/status/1562831148904230912

seems like cb23 mt scores for 7600x and 7700x

 

[jpiniero]

seems like cb23 mt scores for 7600x and 7700x

That would be less than the 12600K and then 12700K, let alone the E Core spam that is Raptor Lake.

 

[Karnak]

seems like cb23 mt scores for 7600x and 7700x

Impressive considering it's cinebench. Gaming performance will be great and better than expected.

 

[inf64]

That would be less than the 12600K and then 12700K, let alone the E Core spam that is Raptor Lake.

Why is that surprising? 12600K is 16T part and is slower than 16T 7700X. Makes perfect sense.

Edit: 12T i5 12500 is a lot slower than 12T 7600X : 12974pts vs 15100pts . That's 16% difference for two CPUs with 6 big cores with SMT.

 

[TESKATLIPOKA]

	5600x 65W	5700x 65W	5800x 105W	7600x ?W	7700x ?W
CB R23	11225 (100%)	13474 (+20%)	15758 (+40%)	15100 (+35%)	19800 (+76%)

R23 scores are from TPU R7 5700x review.
If 7600x and 7700x are still 65W, then It would be great.

 

[Abwx]

That would be less than the 12600K and then 12700K, let alone the E Core spam that is Raptor Lake.

12600K score 17626 pts, 12700K is at 22861.

5600X score 11564 pts and 5800X is at 15805 pts.

Intel Core i9-12900K, i7-12700K & i5-12600K im Test: Benchmarks in Anwendungen

Intel Alder Lake im Test: Benchmarks in Anwendungen / Ein rekordverdächtiger Testumfang / Windows 10 vs. Windows 11

www.computerbase.de

 

[Det0x]

https://twitter.com/x/status/1562831148904230912

seems like cb23 mt scores for 7600x and 7700x

Then we have the following:

• 7700x = 20xx ST and 198xx MT
• 7600x = 19xx ST and 151xx MT

Looks like a good fight is brewing between:

• 13900k(32t) vs 7950x(32t)
• 13700k(24t) vs 7900x(24t)
• 13600k(20t) vs 7700x(16t)
• 13500 (20t)
• 13400(16) vs 7600x (12t)

This could possibly mean that its on the low-end intel will do the best vs Zen4 this generation, if you only play Cinebench 24/7
Pricing will decide the winner below the 13700k i guess

 

[jpiniero]

Why is that surprising? 12600K is 16T part and is slower than 16T 7700X. Makes perfect sense.

I was kind of expecting AMD to charge 13600K prices for the 7600X and likewise for the 7700X.

This could possibly mean that its on the low-end intel will do the best vs Zen4 this generation, if you only play Cinebench 24/7
Pricing will decide the winner below the 13600k i guess

Don't you think the real competition is AM4? I wonder if that 5600X3D with half vcache is real.

 

[Saylick]

That would be less than the 12600K and then 12700K, let alone the E Core spam that is Raptor Lake.

Sorry, had to.

 

[StinkyPinky]

Am I reading that correctly? The 13700k has two more p cores than the previous gen version? I've been out the loop a bit.

 

[jpiniero]

Am I reading that correctly? The 13700k has two more p cores than the previous gen version? I've been out the loop a bit.

It goes from 8+4 to 8+8.

 

[Det0x]

Am I reading that correctly? The 13700k has two more p cores than the previous gen version? I've been out the loop a bit.

Do note 13600, 13500 and 13400 is based on (refreshed) Alder Lake silicon, not the new Raptor lake

 

[nicalandia]

Am I reading that correctly? The 13700k has two more p cores than the previous gen version? I've been out the loop a bit.

4 More e cores. So basically a 12900K/S

 

[Timmah!]

https://twitter.com/x/status/1562831148904230912

seems like cb23 mt scores for 7600x and 7700x

Well, as expected then, is it not?
7950x was rumored to be ~38K since the reveal back in May, so 7700x with half the cores scoring 19000+ sounds about right.

 

[Ranulf]

View attachment 66588

Sorry, had to.

 

[Zucker2k]

Then we have the following:

• 7700x = 20xx ST and 198xx MT
• 7600x = 19xx ST and 151xx MT

Looks like a good fight in brewing between:

• 13900k(32t) vs 7950x(32t)
• 13700k(24t) vs 7900x(24t)
• 13600k(20t) vs 7700x(16t)
• 13400(16) vs 7600x (12t)

This could possibly mean that its on the low-end intel will do the best vs Zen4 this generation, if you only play Cinebench 24/7
Pricing will decide the winner below the 13600k i guess

That 7700x is pushing very high multithreaded clocks in this test, compared to the 7600x. You can see this from the delta between 1t and nt.

Having said that, it's interesting to see the changing narrative with regard to the upcoming chip matchups. I mean, AMD purposely slotted themselves into the i-moniker categories with their own r-moniker, and pushed chips with superior core counts into these categories:
i9 = R9
i7 = R7
i5 = R5
i3 = R3

Here's the msrp of these chips at release:
i9 / R9
1. $799 = 5950x
2. $589 = 12900k
3. $564 = 12900kf
4. $549 = 5900x
i7 / R7
5. $449 = 5800x
6. $409 = 12700k
7. $384 = 12700kf
i5 / R5
$299 = 5600x
$289 = 12600k
$264 = 12600kf
$202 = 12500
$164 = 12400f

Just looking at the categorization and pricing, where do y'all think these chips were meant to compete at? None of those pushing thread for thread comparisons did so when the core balance favored AMD, and most saw that as a good thing because AMD offered consumers more value for their money. Well, take a look at the list above. What has changed?

 

[v.strix]

Better to stick with the litteral sentences from AMD, they are clear on the conditions and they match all others deductions that can be made from previous technical tips, that is, 0.5x the power at isofrequency and 1.25x the frequency at isopower for the N5P process.

The fact that they give both perf and perf/watt improvement makes no doubt about the conclusions, the chip perform at least 1.35x better and its comsumption is 1.35/1.25 = 1.08 x higher at most.

Dude, pipe down on that whole thing for your own sake, seriously.
You have precisely as much grounds for your "calculations" as that other guy who "desperately hope" for it to be wrong.

First of all, AMD wouldn't give out nearly accurate performance estimate of a product more than a quarter prior to its release!

Those figures are ambiguous for a reason.
Was that >25% achieved at the same power or at the same performance? Maybe neither was equal? You don't know. Because it's not stated anywhere in those slides.
All those are but points on the power-performance curve. You just picked an arbitrary one basing it on what exactly? Personal preferences?

The second figure is even worse unfortunately.
You're assuming that it refers to the same test as the first one. But there are 2 endnotes describing 2 different tests performed at different dates. First one indeed is about CB R23 nT between 5950X and "pre-production 7000 series 16-core". The other one, however, is about CB R23 1T single-thread test and there's no mention of power measurement here unlike in the first one.
Does that mean that >35% figure is actually referring to a single-thread improvement, despite it's not explicitly stated as such on the slide? Or is it just someone at AMD mixed something up and this second endtote is actually irrelevant to anything in the entire presentation (it's not referenced from any other page either)?

Again, we don't know. You're just filling unknown variables with whatever you please to arrive at those "conclusions". And while you're doing so, one thing i don't understand is why are you so modest? Let us just assume the power efficiency figure is about 7950X at its full might. A 16-core CPU at heavy multi-threaded load would surely be capped by power. So with accordingly raised TDP/PPT we have:

170/105*1.25 = 2.02

Whoo hoo, look at this! Zen 4 actually doubles the performance in CB R23 nT test!
AMD revealed this all the way back in june but nobody noticed...