AMD Zen 4 EPYC CPU Benchmarked Showing a 17% Single Thread Performance Increase from Zen 3

[Uskompuf]

The next-generation flagship AMD GENOA EPYC CPU has recently appeared on Geekbench 5 in a dual-socket configuration for a total of 192 cores and 384 threads. The processors were installed in an unknown Suma 65GA24 motherboard running at 3.51 GHz and paired with 768 GB of DDR5 memory. This setup achieved a single-core score of 1460 and multi-core result of 96535 which places the processor approximately 17% ahead of an equivalently clocked 128 core EPYC 7763 in single-threaded performance. The Geekbench listing also includes an OPN code of 100-000000997-01 which most likely corresponds to the flagship AMD EPYC 9664 with a max TDP of 400 W according to existing leaks.



View at TechPowerUp Main Site | Source

 

[prtskg]

17% ahead in single thread or multiple?

 

[usiname]

17% ahead in single thread or multiple?

17% single, 28% multi

 

[Dyatlov A]

17% single, 28% multi

17% single at same clock? And can do 10%+ higher clocks?

 

[Jimmy_]

damn AMD is heading ahead in comparison to intel in severs with full speed.

 

[usiname]

192 cores.... that's insane i thought its will have 96 cores as max.

It is double socket system, the max is 96 cores

17% single at same clock? And can do 10%+ higher clocks?

This is geekbench, I don't trust nor to the reported clocks, nor to the scores

 

[Jimmy_]

It is double socket system, the max is 96 cores


This is geekbench, I don't trust nor to the reported clocks, nor to the scores

yeah i read that later my bad

Geekbench is quite sus with results though

 

[Crackong]

damn AMD is heading ahead in comparison to intel in severs with full speed.

If we talk about server CPUs, Intel is still struggling to go onto year 2021.

 

[Jimmy_]

If we talk about server CPUs, Intel is still struggling to go onto year 2021.

View attachment 259110

Yeah true - intel has around ~500 bugs for sapphire rapids
they ll delay and are delaying even more ( DG2 and server CPUs) - I don't remember if intel has followed any of the road maps these days!

 

[Daven]

If we talk about server CPUs, Intel is still struggling to go onto year 2021.

View attachment 259110

Its insane that Cooper Lake which is just a rehash of cascade lake is still the top server CPU from Intel. Why is everything coming so late from that company?

 

[ADB1979]

It is double socket system, the max is 96 cores

Genoa is up to 96 cores per CPU... A 50% increase from previous gen because it uses (max) 12 CPU dies instead of 8

 

[usiname]

Genoa is up to 96 cores per CPU... A 50% increase from previous gen because it uses (max) 12 CPU dies instead of 8

And why you quote my coment? I said exactly the same thing, 96 cores max...

 

[ADB1979]

It is double socket system, the max is 96 cores


This is geekbench, I don't trust nor to the reported clocks, nor to the scores

And why you quote my coment? I said exactly the same thing, 96 cores max...

96 cores PER CPU.

 

[Punkenjoy]

it look unclear to me if this gain is with boosting or with clock fixed at 3.5 GHz.

I assume that it's with boosting. That would mean we could expect similar gain for Ryzen CPU. If it's at static clock and Ryzen 7000 is clock way higher than Zen 3, this will mean huge gain.

We will know more soon. but still that is a nice gain.

 

[Lew Zealand]

it look unclear to me if this gain is with boosting or with clock fixed at 3.5 GHz.

I assume that it's with boosting. That would mean we could expect similar gain for Ryzen CPU. If it's at static clock and Ryzen 7000 is clock way higher than Zen 3, this will mean huge gain.

We will know more soon. but still that is a nice gain.

Geekbench does not report any boosting status. All it detects is the default speed as reported to the OS so this could just as easily be a fully boosted or a locked to default speed part. My guess is boosted.

 

[ADB1979]

it look unclear to me if this gain is with boosting or with clock fixed at 3.5 GHz.

I assume that it's with boosting. That would mean we could expect similar gain for Ryzen CPU. If it's at static clock and Ryzen 7000 is clock way higher than Zen 3, this will mean huge gain.

We will know more soon. but still that is a nice gain.

The Zen 4 clocks are a lot higher than Zen 3, a fair chunk of the total performance gains are via clock speeds, and yes, that does include "boosts", but also, all-core loads are a lot higher, as are clocks for 2-4 core loads, so multithreaded tasks and lightly threaded tasks will perform better pro-rata than single threaded as the relative clock increases are higher.

 

[HenrySomeone]

Geekbench does not report any boosting status. All it detects is the default speed as reported to the OS so this could just as easily be a fully boosted or a locked to default speed part. My guess is boosted.

Yup, it truly looks as though single thread increase in the realm of 15% is all AMD will muster with zen4... That won't be nearly enough to catch up even with Alder Lake, much less Raptor Lake and it will be hopelessly inferior to Meteor Lake, against which it will most likely compete later in its cycle. Still, they should remain competitive in the server segment for a while due to the really high core count, desktop though...not so much:

 

[Lew Zealand]

Yup, it truly looks as though single thread increase in the realm of 15% is all AMD will muster with zen4... That won't be nearly enough to catch up even with Alder Lake, much less Raptor Lake and it will be hopelessly inferior to Meteor Lake, against which it will most likely compete later in its cycle. Still, they should remain competitive in the server segment for a while due to the really high core count, desktop though...not so much:

This is an Epyc CPU. So the point is to compete with Intel Server CPUs, which hasn't had a new one since what, 2020? Earlier? And won't until some undefined time "next year". Alder Lake and all those other consumer CPUs are irrelevant in the server space. Playing fanboy games and imagining a fanciful future about consumer CPUs based on a single benchmark of a server CPU is laughable.

 

[ADB1979]

Yup, it truly looks as though single thread increase in the realm of 15% is all AMD will muster with zen4... That won't be nearly enough to catch up even with Alder Lake, much less Raptor Lake and it will be hopelessly inferior to Meteor Lake, against which it will most likely compete later in its cycle. Still, they should remain competitive in the server segment for a while due to the really high core count, desktop though...not so much:

R23 is a great gauge for scaling cores, but no one who uses the underlying technology that powers R23 will ever use it in a real world situation on a single core. How about a "real world" comparison, rather than a benchmark that has nothing to do with reality.!

Yes, the Alder Lake CPU's are great (competition at last), and yes both Raptor Lake and AMD 7000 will be out shortly, but AMD 8000 is only 1-year away (depending on a million factors).!

Be grateful that we have competition in all arenas, and new technology all happening at once, it's glorious

 

[HenrySomeone]

This is an Epyc CPU. So the point is to compete with Intel Server CPUs, which hasn't had a new one since what, 2020? Earlier? And won't until some undefined time "next year". Alder Lake and all those other consumer CPUs are irrelevant in the server space. Playing fanboy games and imagining a fanciful future about consumer CPUs based on a single benchmark of a server CPU is laughable.

Single thread increases in the Epyc line gen over gen has so far quite closely followed the ones in the desktop space and AMD themselves have stated 15% in their slides (which their ardent fanboys claim was just them underselling themselves, lol ), so it's far from basing anything on a single benchmark, but rather their own claims that are seeming more and more like reality with every passing week.

R23 is a great gauge for scaling cores, but no one who uses the underlying technology that powers R23 will ever use it in a real world situation on a single core. How about a "real world" comparison, rather than a benchmark that has nothing to do with reality.!

Yes, the Alder Lake CPU's are great (competition at last), and yes both Raptor Lake and AMD 7000 will be out shortly, but AMD 8000 is only 1-year away (depending on a million factors).!

Ahhh, whatever doesn't show AyyyyMD in the best of light, is invalid, right? Be sure to mention that to Wizzard (and most of the other reviewers), so he(they) can adapt their methods accordingly! Oh, and you are nothing if not optimistic if you believe zen5 is only a year away... (8000 series might actually be, but that will be more akin to 2000 over 1000, i.e. a polished turd )

 

[Panther_Seraphin]

This uplift is really nice and by the looks of its its a NON 3d cache with those numbers reported by Geekbench.

Now if the number uplift basically means this keeps up with Alderlake design per core in a single thread but has 40% more cores per socket that is MASSIVE in the datacenter environment in terms of density capability and min/maxing U space utilisation.

Now how the HBM intergration on Sapphire Rapids will impact performance will be interesting to see as 3D cache on the previous Gen Milan vs Non 3d cached models could provide a 10% uptick in server/workstation workloads in situations where large cache would be beneficial (think database crunching/medical modelling etc)

Intel I think will take the Single/Low threaded crown even in the datacenter so things like FinTech etc will be throwing all the money at Intel regardless.

Im just glad we have 2 VERY competative options in the market pushing innovation/improvements with each release vs the Stagnation Of Sandy to Coffee Lake when Bulldozer/Piledriver/Excavator was the only competition.

 

[Minus Infinity]

I find it impossible to believe top spec 96 core Genoa will be any less than 50% faster in MT given architectural, cache, and clock speed improvements. Desktop Zen 4 is going to be > 35% faster in MT with the same core counts as Zen 3. If cRaptor Lake 13900 can do 60% better in some MT tests compared to 12900 with 50% more cores, no reason Epyc won't either. 28% would be average even in the same core configs as Milan.

 

[ThrashZone]

Hi,
Not sure how one cools 384 threads

 

[InVasMani]

With cool white shoe laces!

 

[usiname]

96 cores PER CPU.

Ah, now I understand, you have problem with the understanding of information.

For the clueless people, I didn't mean 96 cores max for the double socket system (48+48), I mean 96 per cpu.