Discussion Intel current and future Lakes & Rapids thread

Page 925 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DavidC1

Golden Member
Dec 29, 2023
1,147
1,830
96
Granite Rapids definitely has scaling issues, even against Sierra Forest. Few benchmarks where it would be top or be close to the fastest Turin, it gets behind in 2P.

I'm seeing few tests where 2P SRF is maybe 10% faster than 1P but on GNR, it's slower on 2P. Of course those are in workloads that aren't horrible.

Based on SRF results, it scales less than both Turin and the previous Xeon chip too, so some are platform problems.
 

jpiniero

Lifer
Oct 1, 2010
15,326
5,844
136
Who in their right mind wants to go back to 8P systems? The reason why we're getting 192c and more per socket is to stay away from that.

I'm sure there is some demand for it. But it's really weird that the 8 channel models have it and not the 12. Should be the other way around.
 

Joe NYC

Platinum Member
Jun 26, 2021
2,665
3,810
106
Who in their right mind wants to go back to 8P systems? The reason why we're getting 192c and more per socket is to stay away from that.

Especially if you can get 768 threads and 48 DIMMs at commodity pricing.

Intel may have made promises to their legacy Enterprise customers: "You need to stay with Intel, because AMD is not going to deliver 4s-8s servers". Which would explain continuing on this irrational course of perusing 4s-8s servers. while bleeding market share in much bigger 1s and 2s.

Money spent on developing and validating this platform is throwing good money now after bad money spent on initiating this project.

Here is the system mentioned on ServeTheHome:

 

511

Senior member
Jul 12, 2024
902
786
96
Especially if you can get 768 threads and 48 DIMMs at commodity pricing.

Intel may have made promises to their legacy Enterprise customers: "You need to stay with Intel, because AMD is not going to deliver 4s-8s servers". Which would explain continuing on this irrational course of perusing 4s-8s servers. while bleeding market share in much bigger 1s and 2s.

Money spent on developing and validating this platform is throwing good money now after bad money spent on initiating this project.

Here is the system mentioned on ServeTheHome:

FWIW 704 Cores 1408 threads and like 64 Dimms at 8800 MT/s anyway the bleeding will slow down competitively vs EMR/SPR

On a sidenote where are the benchmarks for QAT/IAA/DLB/DLA if someone uses them in silicon it's game over for both Intel and AMDs general purpose cores in those benchmark there is that factor for TCO and they have actual software support for it yet no one is benchmarking these dead silicon from our point of view
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,972
3,641
136
FWIW 704 Cores 1408 threads and like 64 Dimms at 8800 MT/s anyway the bleeding will slow down competitively vs EMR/SPR

On a sidenote where are the benchmarks for QAT/IAA/DLB/DLA if someone uses them in silicon it's game over for both Intel and AMDs general purpose cores in those benchmark there is that factor for TCO and they have actual software support for it yet no one is benchmarking these dead silicon from our point of view
because people dont buy servers based of benchmarks for features no one uses .............
 

Khato

Golden Member
Jul 15, 2001
1,249
321
136
Exactly DC Customers knows more about these than us common folks they can't be fooled into buying inferior products
Indeed. Any of the large customers are performing their own studies on their relevant workloads and could care less about the results from publicly available reviews that we're commenting on. Said reviewers know this, and since the Intel accelerators don't exactly fit into their normal benchmark suite there's little reason for them to include them. I mean, the only way Intel got ServeTheHome to benchmark SPR accelerators was to give them early access months before launch... to benchmark the accelerators only. Though to be fair, STH has at least experimented with accelerator testing on occasion, and does tend to mention their implications on benchmarks where they could have been enabled.

But yeah, large customers aren't going to buy AMD for workloads where Intel accelerators are applicable.