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!

 

[gdansk]

As 8004 series has shown us, the V/F curve of PHX2 might have been true. No matter the TDP, even 32C cores are not clocked above 3Ghz. They have like 2x watts per core headroom vs Bergamo, yet stay on 3Ghz max.
Even more is given away by base/boost clocks, not really going anywhere with increasing watts per core.

So pretty much the sweet spot is ~2.6Ghz and not only Zen4C can't clock 4-5Ghz like You've claimed above, it is not really efficient above 3Ghz either.

What portion of the TDP is IOd related? But yes it is apparent 3GHz is AMD's preferred max clock rate for Zen4C.

 

[Abwx]

As 8004 series has shown us, the V/F curve of PHX2 might have been true. No matter the TDP, even 32C cores are not clocked above 3Ghz. They have like 2x watts per core headroom vs Bergamo, yet stay on 3Ghz max.

And why should the 32C clock higher..?.
That would negate the 2x perf/watt advantage they are claiming in respect of the compertition offerings.

Even more is given away by base/boost clocks, not really going anywhere with increasing watts per core.

So pretty much the sweet spot is ~2.6Ghz and not only Zen4C can't clock 4-5Ghz like You've claimed above, it is not really efficient above 3Ghz either.

We ll know once there s some silicon at hand for testers, if a chip is that limited above 3GHz it means that it has already mediocre perf/watt at 3GHz, it s not like the perf/watt is optimal up to given frequency and then fall abruptly, for the chip to be efficient at 3GHz it must work at least up to 4GHz so the curve at 3GHz is still in a very efficient segment.

 

[Markfw]

And why should the 32C clock higher..?.
That would negate the 2x perf/watt advantage they are claiming in respect of the compertition offerings.

This may or may not be the case for Sienna, but case in point. The 9654 clocks quite a bit different under load than the 9554, 2.7 vs 3.5. I am talking about 100% load on all logical cores.

 

[Joe NYC]

This may or may not be the case for Sienna, but case in point. The 9654 clocks quite a bit different under load than the 9554, 2.7 vs 3.5. I am talking about 100% load on all logical cores.

If you want to name a baby girl, Sienna would be a fine name, but the name of the city in Italy (and the platform name) is spelled Siena.

 

[Abwx]

This may or may not be the case for Sienna, but case in point. The 9654 clocks quite a bit different under load than the 9554, 2.7 vs 3.5. I am talking about 100% load on all logical cores.

Genoa is for high perfs while Siena is tailored for efficency, so increasing clocks of the 32C respectively to the 64C would reduce perf/watt and hence competitivity in respect of Intel s equivalent offerings.

If they have 2x the perf/watt of intel for the 64C vs 60C that wouldnt be the case for the 32C vs 32C if this latter s offerings have frequency increased significantly, not counting that the IOD power doesnt scale to 0.5x for half the core count.

 

[Markfw]

If you want to name a baby girl, Sienna would be a fine name, but the name of the city in Italy (and the platform name) is spelled Siena.

Then why do they also spell this Sienna ?

AMD launches EPYC 8004 Sienna CPU series, up to 64 Zen4c cores and SP6 socket - VideoCardz.com

EPYC 8004 features up to 64 Zen4c cores AMD is launching a new series of its Zen4 processors. AMD EPYC Genoa, Genoa-X, Bergamo, and Sienna all share a common feature: they are powered by AMD Zen4 architecture and are strategically tailored to cater to the high-core count market segment within...

videocardz.com

 

[Abwx]

Then why do they also spell this Sienna ?

AMD launches EPYC 8004 Sienna CPU series, up to 64 Zen4c cores and SP6 socket - VideoCardz.com

EPYC 8004 features up to 64 Zen4c cores AMD is launching a new series of its Zen4 processors. AMD EPYC Genoa, Genoa-X, Bergamo, and Sienna all share a common feature: they are powered by AMD Zen4 architecture and are strategically tailored to cater to the high-core count market segment within...

videocardz.com

That s a misspelling from Videocardz.

AMD Epyc 8004: Siena sorgt im Sockel SP6 doch noch für eine Überraschung

AMD machte bis zur letzten Minute ein großes Geheimnis aus Siena, nun ist bekannt warum: Die Server-CPU setzt doch nicht auf Zen-4-Kerne.

www.computerbase.de

 

[Joe NYC]

Then why do they also spell this Sienna ?

AMD launches EPYC 8004 Sienna CPU series, up to 64 Zen4c cores and SP6 socket - VideoCardz.com

EPYC 8004 features up to 64 Zen4c cores AMD is launching a new series of its Zen4 processors. AMD EPYC Genoa, Genoa-X, Bergamo, and Sienna all share a common feature: they are powered by AMD Zen4 architecture and are strategically tailored to cater to the high-core count market segment within...

videocardz.com

Between Videocardz and AMD official roadmap, I am going with the official roadmap:

 

[rainy]

AMD needs more tools for mobile tweaking. I can’t do anything with my Cezanne laptop, not even eco mode.

Maybe you should try one of those tools?
https://amdaputuningutility.com/
https://ryzencontroller.com/

 

[BorisTheBlade82]

As 8004 series has shown us, the V/F curve of PHX2 might have been true. No matter the TDP, even 32C cores are not clocked above 3Ghz. They have like 2x watts per core headroom vs Bergamo, yet stay on 3Ghz max.
Even more is given away by base/boost clocks, not really going anywhere with increasing watts per core.

So pretty much the sweet spot is ~2.6Ghz and not only Zen4C can't clock 4-5Ghz like You've claimed above, it is not really efficient above 3Ghz either.

I am not having trouble with the v/f curve above 3 GHz - it is everything below that, I find underwhelming. Besides a very small range around 1.5 GHz it is worse than Zen4.
So Bergamo/Siena is really just about area savings and being able to cram more cores on a package.
Hoping for Phoronix et al. with ISO core, ISO frequency package consumption comparisons between Siena and Genoa.

 

[Thibsie]

I am not having trouble with the v/f curve above 3 GHz - it is everything below that, I find underwhelming. Besides a very small range around 1.5 GHz it is worse than Zen4.
So Bergamo/Siena is really just about area savings and being able to cram more cores on a package.
Hoping for Phoronix et al. with ISO core, ISO frequency package consumption comparisons between Siena and Genoa.

Price might be the real incentive....we'll see.

 

[adroc_thurston]

I am not having trouble with the v/f curve above 3 GHz - it is everything below that, I find underwhelming. Besides a very small range around 1.5 GHz it is worse than Zen4.
So Bergamo/Siena is really just about area savings and being able to cram more cores on a package.

Do proper power instrumentation before making statements like these.

 

[Abwx]

Performance figures for AMD s 96C HEDT :

AMD Threadripper PRO 7000 coming this fall with 20% faster performance than 5000 series, 96 cores confirmed - VideoCardz.com

AMD Threadripper 7000 launching this fall with up to 96 cores A series of slides from AMD Threadripper PRO 7000 presentation have been shared by Underfox. AMD Threadripper PRO, Source: Underfox/AMD These slides corroborate essential details regarding the upcoming “Storm Peak” Threadripper...

videocardz.com

 

[Markfw]

Performance figures for AMD s 96C HEDT :

AMD Threadripper PRO 7000 coming this fall with 20% faster performance than 5000 series, 96 cores confirmed - VideoCardz.com

AMD Threadripper 7000 launching this fall with up to 96 cores A series of slides from AMD Threadripper PRO 7000 presentation have been shared by Underfox. AMD Threadripper PRO, Source: Underfox/AMD These slides corroborate essential details regarding the upcoming “Storm Peak” Threadripper...

videocardz.com

It does not say, but I bet its on socket SP5, and then I bet Noctua will have a good heatsink on it that I can use for my 9554's and 9654's

 

[jpiniero]

It does not say, but I bet its on socket SP5

There might be Threadrippers also released for Siena's socket. We'll have to see if it gets released and what they call it.

 

[naukkis]

I am not having trouble with the v/f curve above 3 GHz - it is everything below that, I find underwhelming. Besides a very small range around 1.5 GHz it is worse than Zen4.

Power efficiency is more than just v/f-curve. Zen4c static and dynamic capacitance is about 2/3 of that regular Zen4 core which means that c--core need to have 22% more voltage to consume same amount energy as regular core at given frequency. When it can operate with less than that 22% more voltage than regular core it will be more efficient. Calculations are approximations as real capacitance figures aren't known so don't took example calculation too seriously. Point is that c-core is far more power efficient than what you seem to believe.

 

[BorisTheBlade82]

Do proper power instrumentation before making statements like these.

I am basing that statement on the first PHX2 test. And yes, I am hoping for some proper analysis by Phoronix, STH, etc. (too bad, this is nothing that could be expected anymore from AT).

 

[adroc_thurston]

I am basing that statement on the first PHX2 test

There are power contributors not v/f curve.
Instrument the thing properly.

 

[Abwx]

I am basing that statement on the first PHX2 test. And yes, I am hoping for some proper analysis by Phoronix, STH, etc. (too bad, this is nothing that could be expected anymore from AT).

Those measurement are dubbious, if we look at the voltages at 3GHz the c core has almost 20% higher voltage than the regular core, that would imply more than 40% higher power at same frequency, and i dont think that this can be compensated by lower capacitance.

Beside i ll point that densification is not forcibly synomymous of lower capacitance, it can be quite the contrary since the capacitance between 2 conductors increase proportionately the the closeness., this behaviour include the distance and resulting capacitance between the transistors s terminals since they are at different electrical potentials.

 

[Shmee]

Any additional updates on Storm Peak? I am very curious for a new HEDT platform, especially one that excels in gaming as well as having the extra IO.

 

[jpiniero]

Any additional updates on Storm Peak? I am very curious for a new HEDT platform, especially one that excels in gaming as well as having the extra IO.

Not much other than "Soon"

 

[Racan]

Impressive energy efficiency from the 7800X3D in Cyberpunk 2077 Phantom Liberty.

Cyberpunk 2077 Phantom Liberty im Tech-Test: 60 Prozessoren im größten PCGH-Benchmark-Vergleich!

Das Warten hat ein Ende: Hier finden Sie 60 CPUs, getestet in Cyberpunk 2077 Phantom Liberty. Welche Prozessoren liegen vorne?

www.pcgameshardware.de

Cyberpunk 2077 Phantom Liberty: Leistungsaufnahme zwischen 38 und 223 Watt - 60 CPUs überprüft

Parallel zum größten CPU-Benchmark-Test von PCGH gibt es für Cyberpunk 2077 Phantom Liberty eine große Analyse der Leistungsaufnahme von Prozessoren.

www.pcgameshardware.de

 

[Markfw]

I still think that both should use at least 6000 memory.

 

[Mopetar]

Impressive energy efficiency from the 7800X3D in Cyberpunk 2077 Phantom Liberty.

Cyberpunk 2077 Phantom Liberty im Tech-Test: 60 Prozessoren im größten PCGH-Benchmark-Vergleich!

Das Warten hat ein Ende: Hier finden Sie 60 CPUs, getestet in Cyberpunk 2077 Phantom Liberty. Welche Prozessoren liegen vorne?

www.pcgameshardware.de

Cyberpunk 2077 Phantom Liberty: Leistungsaufnahme zwischen 38 und 223 Watt - 60 CPUs überprüft

Parallel zum größten CPU-Benchmark-Test von PCGH gibt es für Cyberpunk 2077 Phantom Liberty eine große Analyse der Leistungsaufnahme von Prozessoren.

www.pcgameshardware.de

View attachment 86029

View attachment 86030

You should really have included the 7700X in the power chart. It's not so much that the 7800X3D is amazing as much as Intel CPUs are just bad from a power efficiency perspective in this title.

The Zen 3D parts have always had great showings in these type of charts, in part because they can't clock as high which cuts down in the power draw by a fair bit. However the 7700X also has low power use (only 88W) compared to the Intel CPUs.

 

[Markfw]

I just want to post this here, since its basically about Zen 4. Its about a contest we have going on using primegrid. The more recent CPUs of any mfg with recent FMA/AVX/AVX-512 benefit, so Xeons mostly all have an equal hand. Many of these people use cloud resources. I am pretty sure Nick (3rd place) has 138 computers, most of which appear to be VMs from the cloud, and many are 12 core VMs from a Intel® Xeon® Gold 6242 Processor. First place is an assortment of older Xeons, 70 servers with 140 chips with 14 or more cores each.

I am in second place. I started with only 5 7950x's , 3 9554 Genoas and a 9654 Genoa. I later added ONE 7950x, and fired up a dual 7v12 (128 cores) and a dual 7763 (ES chips, 128 cores). so 14 CPUs in 12 boxes, but I am very close to first place, and have had more than s3rd and 4th place put together most of the time.

What I am trying to point out is the power of Genoa. While wildly different in core count, they are close to number one overall. This contest has people from all over the world. in my own set of computers, one 9554 produces more points than 128 cores of Rome, and more than 128 corfes of Milan. Not sure, but it might be for more than all 256 cores of those.

This is why I have invested in the 4 Genoa chips.
Details as of a few minutes ago day 8.5 of 10 days total.

1​ Pavel Atnashev Ural Federal University 69 960 791.29​ 749​ 2​ markfw TeAm AnandTech 61 071 609.87​ 902​ 3​ Nick BOINC@AUSTRALIA 32 980 727.83​ 603​ 4​ tng Antarctic Crunchers 30 543 476.05​ 670​

A little math. First place has a minimum of 1960 cores, but a few of those have 18 or 22 cores each, so rounding up to 2000 is not much of a stretch.

Edit: Correction by skillz
"His Xeons range from 10 core 2660 V2 to 12 core 2695 V2 CPUs. 70 servers with dual CPUs is 140 CPUs.
If they were all 12 core parts that's a total of 1,680 Cores."

I think you get the gist though. 384 with avx-512 with 256 more with none. The slowest Genoa times were 23 hours. The 7V12's were 48 hours, and the 7763's were 36 hours.

Mine is 640 cores. Yes, they are much newer than the 7 year old Xeons, but I have less than 1/3rd of the cores, and all of the Xeons have avx-512, as well as the Zen 4's. (one of the biggest pluses for this competition), but I still have 88% of his score, and all the 256 of the Rome/Milan and the 7950x were not added until about day 5.

Edit: there is also a 3495x Intel in this competition. Genoa blows the doors off of it also.