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!

 

[LightningZ71]

Prices were recently cut on 5xxx. 58x3d is going for $329 and the 5800x has been spotted for $199.

 

[biostud]

Prices were recently cut on 5xxx. 58x3d is going for $329 and the 5800x has been spotted for $199.

That is what I was referring to.

 

[moinmoin]

Woops, there goes Greymon. Did anyone get a screenie of that tweet before he axed it?

It was nothing special really. He stated that 7000X3D HVM started in 3rd week of October and that going by the timeline when 5800X3D HVM and launch occurred 7000X3D should launch around February/March.

Personally I'd have much preferred some youtubers to completely delete themselves before him, ah well. It's rumors any way so no loss really.

 

[DrMrLordX]

It was nothing special really. He stated that 7000X3D HVM started in 3rd week of October and that going by the timeline when 5800X3D HVM and launch occurred 7000X3D should launch around February/March.

Personally I'd have much preferred some youtubers to completely delete themselves before him, ah well. It's rumors any way so no loss really.

Guess we'll find out? Not that anyone will care much about the fate of Greymon after this particular rumour proves out (or doesn't).

 

[Gideon]

Is it certain X3DX parts are coming in spring? Considering Zen 5 is a 2024 producit, I wouldn't be surprised if AMD times these to the second half of 2023
(i hope not, but wouldn't be surprised)

 

[biostud]

Is it certain X3DX parts are coming in spring? Considering Zen 3 is 2024 i wouldn't be surprised if AMD times these to second half of 2023 :/
(i hope not, but wouldn't be surprised)

zen 5?

 

[Gideon]

zen 5?

yes, obviously. A late night typo, edited it out

 

[RnR_au]

Is it certain X3DX parts are coming in spring? Considering Zen 5 is a 2024 producit, I wouldn't be surprised if AMD times these to the second half of 2023
(i hope not, but wouldn't be surprised)

Makes some sense to release it early given the pentup demand for it. If its not mentioned in the Jan 4 CES presentation, it'll probably be a 2H release.

It'll be interesting to see what stacking tech they'll be using. N5-on-N5 or N6-on-N5.

 

[A///]

CES for sure. It'll kill the announcement for Intel's 13900KS announcement. my memory isn't what it used to be and I'd spent a long time looking up announcements without getting anywhere. Su showed off the 5900X with v-cache at computex 2021. 5800x v cache came out earlier this year. I'd say amd was supply constrained in the latter half of 2021 and early 2021 and didn't see a point in releasing more than one product along with higher costs, especially so close to raphael's launch and knowing how much of a nuisance the 5800x 3d would be to them not to mention intel. imo we'll see more than one 3d model this generation. am5 and 13th gen sales don't seem hot going off of one online retailer but can't tell how things will look until Intel's next quarterly. They themselves mentioned alderlake didn't sell well. if apple is raking back orders on phones then things are certainly bad, because every tiktok star needs the latest and greatest i device and other materialistic bs for their stupid little bubble.

 

[nicalandia]

AMD Ryzen 7600 non-x makes it's debut on Sisoftware

Details for Component AMD Ryzen 5 7600 6-Core

: SiSoftware Official Benchmark Ranker

ranker.sisoftware.co.uk

 

[nicalandia]

AMD Ryzen 7700 non-X makes it's debut on Sisoftware

: SiSoftware Official Benchmark Ranker

ranker.sisoftware.co.uk

 

[eek2121]

AVX-512 Performance

7700X vs 11900K, 8C/16T performance.

"While initially leary of AMD Zen 4's "double pumped" approach for supporting AVX-512 using a 256-bit data path, it's proven to be very efficient for performance and yield great results without negative clock impairments or wreaking havoc on the power consumption"

AMD Ryzen 7 7700X vs. Core i9 11900K AVX-512 Performance Analysis Review - Phoronix

www.phoronix.com

FYI: running an AVX-512 stress test on my machine absolutely dropped the clocks. 4.46ghz slowly down to 3.8ghz as the chip warmed up. The same workload without AVX-512 was around 5.0-5.1ghz for the same time period.

 

[nicalandia]

FYI: running an AVX-512 stress test on my machine absolutely dropped the clocks. 4.46ghz slowly down to 3.8ghz as the chip warmed up. The same workload without AVX-512 was around 5.0-5.1ghz for the same time period.

Really? Phoronix did not notice any clock gating while doing their benchmarks.. What is the name of the stress test? Do you run on Water or Air?

 

[igor_kavinski]

FYI: running an AVX-512 stress test on my machine absolutely dropped the clocks. 4.46ghz slowly down to 3.8ghz as the chip warmed up. The same workload without AVX-512 was around 5.0-5.1ghz for the same time period.

You are using the 7950X which would generate more heat than a 7700X used by Phoronix.

 

[Kaluan]

AMD Ryzen 7600 non-x makes it's debut on Sisoftware

Details for Component AMD Ryzen 5 7600 6-Core
View attachment 70559

: SiSoftware Official Benchmark Ranker

ranker.sisoftware.co.uk

AMD Ryzen 7700 non-X makes it's debut on Sisoftware
View attachment 70639

: SiSoftware Official Benchmark Ranker

ranker.sisoftware.co.uk

3,2GHz IMC? Someone testing these early samples with DDR5-6400 straight out of the gate? Or am I misunderstanding that?

 

[nicalandia]

3,2GHz IMC? Someone testing these early samples with DDR5-6400 straight out of the gate? Or am I misunderstanding that?

Most likely a faulty reading from Sisoftware

 

[Det0x]

Interesting article:

AMD’s Zen 4, Part 2: Memory Subsystem and Conclusion

Please go through part 1 of our Zen 4 coverage, if you haven’t done so already. This article picks up where the previous one left off. To summarize, Zen 4 has made several moves in the fronte…

chipsandcheese.com

I did some numbers on my own maxed out 7950x system with the same memory benchmark:

Latency:


Dram multicore read bandwidth:

 

[nicalandia]

Interesting article:

AMD’s Zen 4, Part 2: Memory Subsystem and Conclusion

Please go through part 1 of our Zen 4 coverage, if you haven’t done so already. This article picks up where the previous one left off. To summarize, Zen 4 has made several moves in the fronte…

chipsandcheese.com

I did some numbers on my own maxed out 7950x system with the same memory benchmark:

Latency:

Yellow square is L1
Orange square is L2
Red square is L3
Dark red square is System RAM

 

[Carfax83]

View attachment 70670

Interesting. Now that I have my new system built, I've been doing some Aida64 runs and one thing I noticed is that my memory latency scores are higher than what I thought they would be using DDR5 7400. The lowest latency score I've had so far is 61.7ns. This version I am using is a more recent build so they cannot truly be compared, but I definitely think that AMD has the advantage in memory latency due to being able to use Gear 1 with DDR5. If whoever did that run with CL30 DDR5 6000 memory, he would have gotten into the 5xns range for sure.

But Zen 4's read, write and copy scores leave much to be desired to be honest. Not sure what's up with that.

 

[Kaluan]

You don't find the pairing of 7900 XTX and 7950X ridiculous? AMD is telling gamers that this is the premium gaming experience. Typical gamer sees the marketing and ads and buys it. Few months later, AMD releases X3D. Now gamer sees his "premium" gaming machine getting beaten by relatively cheaper DIY desktops with X3D. Do you really expect the gamer to not utter an expletive at that point? It's just a really bad move by AMD.

They don't market the Advantage Desktop program just for gamers, do they? I specifically remember them touting 30% or such faster encoding times when pairing Ryzen 7000 + Radeon 7000 (iGPU + dGPU encoding?), "Smart Access Video". They marketed this in tandem with the Advantage rigs.

Ironically enough, when Raphael-X arrives, I expect them to also put some emphasis on higher productivity this time, not just higher frames. 😅

Anyway, weird hangup TBH, this is nothing new. This always happens to the average consumer, but especially to early adopters.

What were they supposed to do? Not talk
about it/release it at all? Or tease Zen4 V-Cache so they can Osbourne Effect vanilla Raphael even harder? 😂

 

[BorisTheBlade82]

@Det0x
Your numbers are a bit better as the ones from CnC. But they again show quite some lack of bandwidth compared to RTL.

@mmaenpaa
So you have this shiny new 7700X. Might you give the Micro Bench posted by @Det0x a try? 😉 I am really interested how a single CCD ZEN4 might turn out bandwidth wise.

 

[BorisTheBlade82]

But Zen 4's read, write and copy scores leave much to be desired to be honest. Not sure what's up with that.

The assumption is that the IFoP (the interconnect between IOD and CCD) is the bottleneck. It is limited to 64 GByte/s read and 32 GByte/s write. The CnC article points to this as well.
Now, there are some details to be worked out:

• CnC showed write being limited to 64 GByte/s - as expected.
• That could either mean, that a 2 CCD SKU only uses one IFoP port per CCD (although they have two each).
• Or the above mentioned numbers apply to both ports in total.
• Or only on a 1 CCD SKU both ports are used.

This is why @mmaenpaa could help here.

 

[eek2121]

Really? Phoronix did not notice any clock gating while doing their benchmarks.. What is the name of the stress test? Do you run on Water or Air?

It is the prime95 build with AVX-512 support (you have to grab this from the forum or elsewhere, last I checked they don't have it available for download, it is a separate EXE currently).

You are using the 7950X which would generate more heat than a 7700X used by Phoronix.

Yes, it was implied that AVX-512 does not require additional cooling or require additional power consumption. That is clearly not the case. If it were my 7950x would run at the same speeds, AVX-512 or not.

 

[BorisTheBlade82]

@eek2121
Well, you need to take performance into account as well. If the AVX test is much faster even downclocked, then total energy consumption for a given workload very likely is smaller. The main reason for the behaviour should be that the occupancy of execution units is much higher with AVX.

 

[eek2121]

@eek2121
Well, you need to take performance into account as well. If the AVX test is much faster even downclocked, then total energy consumption for a given workload very likely is smaller. The main reason for the behaviour should be that the occupancy of execution units is much higher with AVX.

I agree, however, I feel like it was claimed there was absolutely no overhead. There is. The chip may very well be much faster, regardless.