Linux 6.6 Looks To Be Very Lucrative For AMD Server Performance

Written by Michael Larabel in Software on 19 September 2023 at 02:10 PM EDT. Page 3 of 4. 21 Comments.

PostgreSQL benchmark with settings of Scaling Factor: 100, Clients: 1000, Mode: Read Write. EPYC 9654: Linux 6.6-rc1 was the fastest.

PostgreSQL benchmark with settings of Scaling Factor: 100, Clients: 1000, Mode: Read Write, Average Latency. EPYC 9654: Linux 6.6-rc1 was the fastest.

While Genoa-X is seeing improvements with Linux 6.6, it's not nearly as significant as the Genoa and Bergamo speed-ups over Linux 6.5... Perhaps leading credence to the gains in part are due in part to the workqueue changes in Linux 6.6 around multiple L3 caches?

TensorFlow benchmark with settings of Device: CPU, Batch Size: 16, Model: ResNet-50. Xeon Platinum 8490H: Linux 6.5.1 was the fastest.

TensorFlow is one of the non-database workloads showing very nice gains on Linux 6.6 for AMD EPYC. With TensorFlow both Genoa and Genoa-X posted very positive gains but here the Genoa-X results were unchanged.

TensorFlow benchmark with settings of Device: CPU, Batch Size: 32, Model: ResNet-50. Xeon Platinum 8490H: Linux 6.5.1 was the fastest.

The Intel Xeon Platinum 8490H results were also unchanged. Sapphire Rapids leads here thanks to Advanced Matrix Extensions (AMX).

TensorFlow benchmark with settings of Device: CPU, Batch Size: 64, Model: ResNet-50. Xeon Platinum 8490H: Linux 6.5.1 was the fastest.

EPYC 9684X results being unchanged is quite peculiar since in these instances the Linux 6.6 gains now put the EPYC 9754 and 9654 ahead of the Genoa-X part where traditionally this server CPU with 3D V-Cache held an advantage.

21 Comments - Next Page

Page: 1 2 3 4 Next Page