Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

Page 942 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Geddagod

Golden Member
Dec 28, 2021
1,340
1,431
106
Thats NOT what you said. You said (and I quote) " it will only take 1 good Intel gen to sway everything back."
I mean GNR is already competitive with Turin standard iso power, according to AMD. Turin Dense has like a 25% lead.
I fully expect Intel's server position relative to AMD to only get better from here, since they are releasing more competitive products compared to EMR vs Genoa, and GNR ramps.
 

Abwx

Lifer
Apr 2, 2011
11,712
4,630
136
I mean GNR is already competitive with Turin standard iso power, according to AMD. Turin Dense has like a 25% lead.
At same power the regular 128C Zen 5 is almost 20% faster than a 128C Xeon, wich mean that it would consume 35% less at same perf, that s 50% perf/watt advantage at isoperf.

Beside 25% perf lead for Turin D at wich power.?.
At same throughput than the Xeon 128C@322W, and in Phoronix tests, it would be at about 190W since the perf gap is only 20% at stock for their tests suite and with 275W power at stock, that would push the perf/watt advantage at 70%@isoperf.

 

Attachments

  • Screenshot 2025-02-26 at 06-15-56 AMD EPYC 9755 _ 9575F _ 9965 Benchmarks Show Dominating Perf...png
    59.9 KB · Views: 20
  • Screenshot 2025-02-26 at 06-18-22 AMD EPYC 9755 _ 9575F _ 9965 Benchmarks Show Dominating Perf...png
    65.9 KB · Views: 21

511

Golden Member
Jul 12, 2024
1,670
1,489
106
At same power the regular 128C Zen 5 is almost 20% faster than a 128C Xeon, wich mean that it would consume 35% less at same perf, that s 50% perf/watt advantage at isoperf.

Beside 25% perf lead for Turin D at wich power.?.
At same throughput than the Xeon 128C@322W, and in Phoronix tests, it would be at about 190W since the perf gap is only 20% at stock for their tests suite, that s a massive 70% perf/watt advantage at isoperf.

You are messing up the numbers 9755 was consuming 324W and 6980P was 331W also the performance difference was 20% and a 3% power advantage roughly translates to 23% PPW.
 

Abwx

Lifer
Apr 2, 2011
11,712
4,630
136
You are messing up the numbers 9755 was consuming 324W and 6980P was 331W also the performance difference was 20% and a 3% power advantage roughly translates to 23% PPW.

If it perform 20% better at same power it will perform the same at 0.65x the power, that s a given that you can trade much higher efficency for more perfs, the more you increase the perf the more you decrease efficency.

By the same token you ll have to increase the Xeon power by almost 50% to match the stock 128C Zen 5 perfs wise.
 
Reactions: r.p and Markfw

511

Golden Member
Jul 12, 2024
1,670
1,489
106
If it perform 20% better at same power it will perform the same at 0.65x the power, that s a given that you can trade much higher efficency for more perfs, the more you increase the perf the more you decrease efficency.
I see you used maths 🙂
By the same token you ll have to increase the Xeon power by almost 50% to match the stock 128C Zen 5 perfs wise.
These are just assumption without data cause the power performance curve is non linear.

All the benches are not perfectly parallel as well some benches favour Faster fewer cores(64 Core EPYC Winning some are memory bandwidth heavy )
 
Last edited:

Geddagod

Golden Member
Dec 28, 2021
1,340
1,431
106
At same power the regular 128C Zen 5 is almost 20% faster than a 128C Xeon, wich mean that it would consume 35% less at same perf, that s 50% perf/watt advantage at isoperf.

Beside 25% perf lead for Turin D at wich power.?.
At same throughput than the Xeon 128C@322W, and in Phoronix tests, it would be at about 190W since the perf gap is only 20% at stock for their tests suite and with 275W power at stock, that would push the perf/watt advantage at 70%@isoperf.

 

MS_AT

Senior member
Jul 15, 2024
525
1,109
96
Instead of cherry picking, The total performance that ABWX posted and power show Turin is quite a bit faster, and has more cores available. Why do you keep this up ? This is a Zen 5 thread you know.
Does it mean Zen5 cannot be criticized in Zen5 thread at all? Should we have specific thread for this?
Where is that even from so I can see other benchmarks? Totally derailing the thread BTW.
AMD's own ISSCC slides. You will find sources on the previous pages, I think the Japanese page might have the slides or pause the techtechpotato video to read them.
 
Jul 27, 2020
23,091
16,257
146
Where is that even from so I can see other benchmarks? Totally derailing the thread BTW.

Slides are in English:

 

511

Golden Member
Jul 12, 2024
1,670
1,489
106
Server are bought on SpecInt Performance or evaluation by Hyperscaler
Also this is not a single test
Found the score with details COMPILERS AND ALL THE STUFF

6980p
9755
 

Nothingness

Diamond Member
Jul 3, 2013
3,241
2,295
136
That's a single test. You have to look at the aggregate score.
Aggregate scores are evil, see GB6 aggregate score for MP for instance. And Phoronix ones are even worse, mixing single thread and multi thread tests; are you even able to say how they come up with an aggregated score where for some subtests a lower score is better than a higher one?
 

511

Golden Member
Jul 12, 2024
1,670
1,489
106
Aggregate scores are evil, see GB6 aggregate score for MP for instance. And Phoronix ones are even worse, mixing single thread and multi thread tests; are you even able to say how they come up with an aggregated score where for some subtests a lower score is better than a higher one?
This picking server is not easy now like it was for few years just buy AMD for General Purpose or Intel for AI now it's gotten a bit tricky without actually understanding your workload.
Geekbench with SME is one such example apple is flaunting near 4000 GB score with SME swaying the average away.
 

leoneazzurro

Golden Member
Jul 26, 2016
1,095
1,816
136
Aggregate scores are evil, see GB6 aggregate score for MP for instance. And Phoronix ones are even worse, mixing single thread and multi thread tests; are you even able to say how they come up with an aggregated score where for some subtests a lower score is better than a higher one?
It depends, Phoronix has all the subtests score as well, so one can even look at the kind of workload they are most interested about. Better than looking at a single metric, which may be a very relevant metric but not the only one, as well as one should consider i.e. the platform costs, power consumption and so on.
 

Nothingness

Diamond Member
Jul 3, 2013
3,241
2,295
136
It depends, Phoronix has all the subtests score as well, so one can even look at the kind of workload they are most interested about. Better than looking at a single metric, which may be a very relevant metric but not the only one, as well as one should consider i.e. the platform costs, power consumption and so on.
It seems we agree, but you were talking about aggregated score without any further comment in your previous post SPECint rate also is an aggregated score; without looking at individual tests, I find it of little use (especially given how various compilers are cheating on some of the subtests). Basically you were saying to look at Phoronix aggregated score, responding to an aggregated SPECint score, which made me react.

Given the kind of workloads I use, I exactly know what to look at for my needs. And there neither AMD or Intel have the lead (no, I'm not talking about SME) no matter what aggregated scores are.
 

leoneazzurro

Golden Member
Jul 26, 2016
1,095
1,816
136
It seems we agree, but you were talking about aggregated score without any further comment in your previous post SPECint rate also is an aggregated score; without looking at individual tests, I find it of little use (especially given how various compilers are cheating on some of the subtests). Basically you were saying to look at Phoronix aggregated score, responding to an aggregated SPECint score, which made me react.

Given the kind of workloads I use, I exactly know what to look at for my needs. And there neither AMD or Intel have the lead (no, I'm not talking about SME) no matter what aggregated scores are.
I meant more that if you look only at the SPECInt, you see a picture while the total aggregate score paints a different one. And in server space costs and power consumption are critical as well.
So for getting the truth you have to look at the individual scores and see what suits your needs better (and that review is lacking ARM or Atom derivatives) instead of looking at one test (even if it is a test suite, usually many people look only at the SPECInt aggregate).
Then there is the possibility of Turin-X, that would have been interesting...
 
Reactions: Tlh97 and 511

Geddagod

Golden Member
Dec 28, 2021
1,340
1,431
106
I meant more that if you look only at the SPECInt, you see a picture while the total aggregate score paints a different one
Which picture do you think is more accurate? The one posted by AMD themselves at an actual tech conference, or Phoronix?
And Phoronix ones are even worse, mixing single thread and multi thread
EMR ended up looking so much closer to AMD than it should have been due to this
DDR5-8800 is going to be more power hungry than DDR5-6000 so platform power ends up being higher still.
True
Does it mean Zen5 cannot be criticized in Zen5 thread at all? Should we have specific thread for this?
Look at who you are quoting lol
Instead of cherry picking
lol
Why do you keep this up ? This is a Zen 5 thread you know.
The CPU in question is Turin, and the data is from a paper AMD presented about Zen 5.
 

branch_suggestion

Senior member
Aug 4, 2023
610
1,322
96
Which picture do you think is more accurate? The one posted by AMD themselves at an actual tech conference, or Phoronix?

EMR ended up looking so much closer to AMD than it should have been due to this

True

Look at who you are quoting lol

lol

The CPU in question is Turin, and the data is from a paper AMD presented about Zen 5.
Lots of words to say Intel is hopelessly behind in BOM.
GNR with fancy DIMMS is far more cost than standard Turin, both to the customer and Intel.
And it is still slower.
DMR loses ground to Venice, CWF is too late as it gets steamrolled quickly by Venice Dense.
No need to repeat this again.
 
Reactions: Tlh97 and Kepler_L2

leoneazzurro

Golden Member
Jul 26, 2016
1,095
1,816
136
Which picture do you think is more accurate? The one posted by AMD themselves at an actual tech conference, or Phoronix?
To be precise, I think the SPECInt or the Phoronix aggregate score are useful but neither of them has any decisive value, per se. When they differ, especially with non-negligible difference, then better to check a multiple real-world test suite, for having a better picture of the situation, and for this only Phoronix did something.
 

Abwx

Lifer
Apr 2, 2011
11,712
4,630
136

How is it that Z5 does here 1260 while pairing two 128C CPUs get you 1360 per CPU for a total of 2720, that would be overscaling, so isnt this slide erroneous with a typo.?

Edit : if two 128C do 1360/CPU in a 2P configuration i would imagine that a single 128C would do more than 1360 but certainly not 1260.
 

Attachments

  • photo013_o.jpg
    57.4 KB · Views: 12
Last edited:

StefanR5R

Elite Member
Dec 10, 2016
6,320
9,717
136
Slides are in English:

Sadly, it's not the complete slide deck. E.g. the end notes are missing. I just looked for them because...
How is it that Z5 does here 1260 while pairing two 128C CPUs get you 1360 per CPU for a total of 2720, that would be overscaling, so isnt this slide erroneous with a typo.?

Edit : if two 128C do 1360/CPU in a 2P configuration i would imagine that a single 128C would do more than 1360 but certainly not 1260.
...maybe the end notes [9] and [11] could clue us in. Or maybe not, because such endnotes are never going into every last detail.
It could have been something as simple and arbitrary as the 2P system having better cooling than the 1P system. Besides, there is always some chip-to-chip variance.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |