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

Page 943 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

511

Golden Member
Jul 12, 2024
1,670
1,489
106
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.
compiler flags and OS and other stuff makes the result variable as well.
 
Reactions: Thibsie

Geddagod

Golden Member
Dec 28, 2021
1,340
1,431
106
Lots of words to say Intel is hopelessly behind in BOM.
The term BOM was never once mentioned in that comment, because no one was talking about it up till now lol.
GNR with fancy DIMMS is far more cost than standard Turin, both to the customer and Intel.
And it is still slower.
Who ever said GNR was the better product?
DMR loses ground to Venice,
Why do you think that would happen?
CWF is too late as it gets steamrolled quickly by Venice Dense.
This prob will end up being true. We will see though.
No need to repeat this again.
You aren't him, bro.
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.?
Could be, honestly, idk, but doesn't change much of my take away at least.
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.
The Zen 5C configuration also had worse than linear scaling.
Sadly, it's not the complete slide deck. E.g. the end notes are missing. I just looked for them because...
Should be up on IEEE in like 2 weeks
 

branch_suggestion

Senior member
Aug 4, 2023
610
1,322
96
Why do you think that would happen?
Because IDC. Big cores are doomed until Unified Core.
Fancy DIMMs are not a requisite you can or can not use them.
True, perf difference is whatever either way.
Yes but not generational gap.
If Intel had the platform cost advantage it would be fine.
They do not.
DMR venice CWF is not out yet so you should reserve judgement by than also DMR is not a refreshed arch like RWC it is going to have 3 architectural jump with APX and stuff.
DMR is same paradigm as GNR, which is a strictly worse paradigm than Venice.
Better hope 18A clocks high.
see you in 2026.
None of the special features will save it.
 

Geddagod

Golden Member
Dec 28, 2021
1,340
1,431
106
Because IDC. Big cores are doomed until Unified Core.
And yet GNR is within striking range of Turin classic.
Diamond Rapids should have P-cores with similar IPC vs Venice too, something which GNR did not enjoy.
If Intel had the platform cost advantage it would be fine.
They do not.
You are seeing ghosts dude. Not a single person here is saying GNR is better than Turin lol.
DMR is same paradigm as GNR, which is a strictly worse paradigm than Venice.
Wdym
None of the special features will save it.
Igh dude lol.
 

Abwx

Lifer
Apr 2, 2011
11,712
4,630
136
The Zen 5C configuration also had worse than linear scaling.

We re talking of the regular 128C, one CPU is supposed to do 1260 while two CPUs manage to do miraculously 1360 per CPU.

Beside if the 6980P was that close in Integer it wouldnt be that lagging in 7Zip, wich according to AT is quite representative of servers perfs in real world.

7Zip score are at the top of the page :


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.
Two CPUS in a box are unlikely to be better cooled than a single one in a box half the volume, say two 128C in a 4U rack and a single 128C in a 2U.

The volume is halved in the second case but the rack total surface is almost the same, so the second case has lower thermal resistance overall.
 
Last edited:

Geddagod

Golden Member
Dec 28, 2021
1,340
1,431
106
We re talking of the regular 128C, one CPU is supposed to do 1260 while two CPUs manage to do miraculously 1360 per CPU.
I'm saying the Zen 5C variant also exhibits that same trait. Each CPU is supposed to do 1510 while two CPUs manage to miraculously do 1550 per CPU.
It doesn't seem to be a typo, but due to the litany of potential reasons the other people have posted in this thread.
Beside if the 6980P was that close in Integer it wouldnt be that lagging in 7Zip, wich according to AT is quite representative of servers perfs in real world.

7Zip score are at the top of the page :
Pretty sure Zen 5 is ahead of GNR in a similar category of spec2017 of as 7zip.
Besides, the gap is like 20% there. That's not exactly a generational lead (other than maybe Zen 3).
 
Reactions: Tlh97 and 511

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
20,947
3,376
126
You know my only RANT about AMD PC's...

If your

MOUNTING SYSTEM IS SO BAD THAT MOST AFTERMARKET HEAT SINKS REQUIRE REMOVAL, WHY OH WHY DO YOU NOT CHANGE IT?

EVERY Ryzen cpu system i have built which i put an aftermarket heat sink required me to remove the existing one completely, and put on the custom mount.

Why even have one to begin with then, why not just use pushpins like Intel on OEM heat sinks and not have us remove the non working retention every new build?

OH and That ThreadRipper + EPYC requires a special TORQ tool to secure CPU otherwise u can and probably will run into memory errors.
 

Thunder 57

Diamond Member
Aug 19, 2007
3,402
5,624
136
You know my only RANT about AMD PC's...

If your

MOUNTING SYSTEM IS SO BAD THAT MOST AFTERMARKET HEAT SINKS REQUIRE REMOVAL, WHY OH WHY DO YOU NOT CHANGE IT?

EVERY Ryzen cpu system i have built which i put an aftermarket heat sink required me to remove the existing one completely, and put on the custom mount.

Why even have one to begin with then, why not just use pushpins like Intel on OEM heat sinks and not have us remove the non working retention every new build?

OH and That ThreadRipper + EPYC requires a special TORQ tool to secure CPU otherwise u can and probably will run into memory errors.

That's quite a rant of things I never heard of. I mean I think TR has that going on but I wouldn't know. Did you have a family member or friend ask you to work on their Ryzen and it gave you trouble recently?
 
Reactions: yottabit

HurleyBird

Platinum Member
Apr 22, 2003
2,785
1,506
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?

GB multi-core scores suck because Primate Labs decided to make them suck.

Phoronix scores are awesome because when you take an enormous number of samples the central limit theorem kicks in and makes them awesome.
 

Schmide

Diamond Member
Mar 7, 2002
5,635
832
126
You know my only RANT about AMD PC's...

If your

MOUNTING SYSTEM IS SO BAD THAT MOST AFTERMARKET HEAT SINKS REQUIRE REMOVAL, WHY OH WHY DO YOU NOT CHANGE IT?

EVERY Ryzen cpu system i have built which i put an aftermarket heat sink required me to remove the existing one completely, and put on the custom mount.

Why even have one to begin with then, why not just use pushpins like Intel on OEM heat sinks and not have us remove the non working retention every new build?

OH and That ThreadRipper + EPYC requires a special TORQ tool to secure CPU otherwise u can and probably will run into memory errors.

Most still use the back plane.

LianLi HydroShift LCD I just installed takes the cake though. You remove the plastic clips, keeping the back plane, put in curved plastic clips with the screws from the original mounting bracket. Mount the block on the curved plastic clips in separate holes. (aka adapter) Then cover up the mess you just made with a plastic cover.

But surprisingly the temps are really really good and it's white with side tubes on the radiator and has an LCD.
 
Last edited:
Reactions: Elfear

Schmide

Diamond Member
Mar 7, 2002
5,635
832
126
(replyed trying to edit a misspelling)

Edit: The intel mounting for the HydroShift isn't a beauty queen either. LianLi provides a back plate with extending posts with spacers that the block screws into. At least both use the same plastic cover to hide the mess.
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
22,473
12,328
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?
In defense of Phoronix, they do a better job with their server/workstation benchmarks than they do their desktop benchmarks. The selections seem more relevant (at least to me), and most (but not all) are aggressively multithreaded.
 

Nothingness

Diamond Member
Jul 3, 2013
3,242
2,295
136
In defense of Phoronix, they do a better job with their server/workstation benchmarks than they do their desktop benchmarks. The selections seem more relevant (at least to me), and most (but not all) are aggressively multithreaded.
It's still hard to say in their selection what benefits from MT, what benefits from unrealistic parallelization, what is purely ST. OTOH having a large selection helps people looking at the workloads that matter to them; and that's why I was pleading against aggregated score.

Something similar applies to GB6, but at least there you clearly know what is supposedly ST or MT. And they don't mix results measured in seconds with results measured with scores. Phoronix aggregated score remains a mystery to me (I quickly searched for that information and couldn't find it).
 
Reactions: yuri69 and MS_AT

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
26,765
15,791
136
It's still hard to say in their selection what benefits from MT, what benefits from unrealistic parallelization, what is purely ST. OTOH having a large selection helps people looking at the workloads that matter to them; and that's why I was pleading against aggregated score.

Something similar applies to GB6, but at least there you clearly know what is supposedly ST or MT. And they don't mix results measured in seconds with results measured with scores. Phoronix aggregated score remains a mystery to me (I quickly searched for that information and couldn't find it).
Problems with GB6 are that is more for desktops, not for servers. Also, Phoronics does not count much for avx-512, it does little benchmarks for that, but in servers it is important.
 

Abwx

Lifer
Apr 2, 2011
11,712
4,630
136
It's still hard to say in their selection what benefits from MT, what benefits from unrealistic parallelization, what is purely ST.
Since the average difference is 20% and that ST differences are quite lower one can
deduct that average MT difference is higher than 20% to get the ST + MT geomean at 20%.
 

MS_AT

Senior member
Jul 15, 2024
525
1,109
96
Phoronics does not count much for avx-512, it does little benchmarks for that, but in servers it is important.
Well, they do definitely more tests with avx512 enabled software than they do compile tests, not to mention they often have specific avx512 comparison articles, while they completely ignore throughput related nature of compilation workloads for CI/CD servers which is usually the role of 64 core+ servers.
 
Reactions: lightmanek

Nothingness

Diamond Member
Jul 3, 2013
3,242
2,295
136
Also, Phoronics does not count much for avx-512, it does little benchmarks for that, but in servers it is important.
They made 2 reviews with AVX-512 for Turin:


Too bad we don't have results for competing Intel chips in the same runs, but I guess these can be extracted from their DB.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
26,765
15,791
136

yuri69

Senior member
Jul 16, 2013
623
1,083
136
Phoronics does not count much for avx-512, it does little benchmarks for that, but in servers it is important.
I've just flagged this post as trolling.

Let's do a breakdown of the benchmarks included in https://www.phoronix.com/review/amd-epyc-9965-9755-benchmarks/14.

They got 11 pages of benchmarks. I've *quickly* glanced over them:
2) OpenSSL uses AVX512
3) John the Ripper uses AVX512; RockDB aka Databases most probably uses AVX512
4) HPC nearly every single one uses AVX512
5) see 4)
6) OpenFOAM uses AVX512
7) OSPRay uses AVX512
8) Embree uses AVX512 heavily, OpenVKL uses AVX512
9) NumPy uses AVX512, 7zip uses AVX512
10) uvg266 uses AVX512, SVT-AV1 uses AVX512
11) libavif most probably uses AVX512; GraphicsMagick most proabably uses AVX512
12) OpenVINO uses AVX512 heavily, there are binaries of Tensorflow wich uses AVX512 - dunno if Phoronix uses that one
13) xmrig uses AVX512; Stockfish uses AVX512
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
20,947
3,376
126
That's quite a rant of things I never heard of. I mean I think TR has that going on but I wouldn't know. Did you have a family member or friend ask you to work on their Ryzen and it gave you trouble recently?

Yeah a cousin was building the 5000 series threadripper and noticed memory issues.
I thought it was ram voltage, and went though almost 3 hours of debugging before i decided to hit Google.
Realized if the CPU is not Torq'd properly, the contact can be uneven, sometimes causing memory errors.

I then asked did you use the special AMD tool?
He was like what AMD tool?

Then i showed him this picture:


Because i remember when i built my EPYC a while back, i had to wait for that tool to come in to finish the build.

He then looked at me and went... AMD has a special tool?
I then went YES... u want HEDT? HEDT requires special tools.
I brought my tool over, properly torqued everything down, and poof.. all the errors disappeared.

Which is why i started that rant... i just finished building 3 AMD systems this week.
A 9800X3d. 9700X and a 9600X. And i was annoyed as hell having to remove those plastic clips in EVERY board, and saving them incase i ever needed to do a RMA, because i can bet you Gigabyte / ASUS will straight up deny the RMA if you do not reinstall those plastic stupid clips.

Oh and incase i get accused of Anti AMD...
Pictures for proof:



Yes i love the X870 board... its a awesome motherboard.
The 9000 series Ryzen's are also SOLID.
Sad to say Intel really lost this and the gap is only getting bigger each generation.
 

Attachments

  • 20250227_114245.jpg
    468.6 KB · Views: 3
Last edited:

Rigg

Senior member
May 6, 2020
610
1,445
136
I've built a bunch of AM4/AM5 systems using many different coolers. The AMD mounting system is totally fine. It provides compatibility for all of their different stock coolers and the back plate is easily adapted to aftermarket mounting hardware. I don't even remember the last time I installed an AM4 compatible aftermarket cooler that didn't at least use the stock back plate. The back plate not being permanently attached could be slightly annoying on AM4 in some circumstances, but they fixed that on AM5. The 2 mounting tabs and 4 screws take about 30 seconds to remove, put in plastic bag, and toss in the the motherboard box for safe keeping. Attaching aftermarket mounting adapters to the back plate generally doesn't take much longer. Older AM3 heatsinks that use the clip-on brackets will also be able to use the AM4/5 stock plastic tabs.There are also some aftermarket coolers that use the stock plastic tabs. I had a a little 90MM be quiet! tower cooler that did. No offense @aigomorla , but this seems like an extremely lazy complaint. The plastic tabs offer additional cooler compatibility so I don't see how this is a bad thing.

The plastic push in system that intel uses for it's stock coolers is absolutely terrible in comparison. I've seen those stupid things pop out of the hole and stop making contact with the IHS on several occasions when servicing systems. Properly torquing a server CPU into a massive LGA socket isn't unique to AMD. Hell, even LGA 1700 can have memory issues if you don't use a contact frame.
 
Last edited:

lightmanek

Senior member
Feb 19, 2017
508
1,244
136
I don't understand where the issue is, AMD ships every Threadripper CPU with a dedicated tool, and if you somehow throw it away before installing CPU, 3 seconds search on internet turns out correct torque spec. which is 1.5Nm. Use any torque wrench you want and set it to 1.5Nm.

I've build 4 TR setups myself and took my own CPU about 10 times out of the socket to change cooler or for maintenance. If you follow instructions telling you in which order to fasten screws, you will not have any problems.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
26,765
15,791
136
I don't understand where the issue is, AMD ships every Threadripper CPU with a dedicated tool, and if you somehow throw it away before installing CPU, 3 seconds search on internet turns out correct torque spec. which is 1.5Nm. Use any torque wrench you want and set it to 1.5Nm.

I've build 4 TR setups myself and took my own CPU about 10 times out of the socket to change cooler or for maintenance. If you follow instructions telling you in which order to fasten screws, you will not have any problems.
I don't know how that converts, but I saw 13-15 inch pounds. If that the same, I apologize.
 
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/    |