Revisiting AMD EPYC 7773X "Milan-X" Performance With Linux 6.0 + Ubuntu 22.10

[phoronix]

Phoronix: Revisiting AMD EPYC 7773X "Milan-X" Performance With Linux 6.0 + Ubuntu 22.10

Earlier this month I revisited the AMD Ryzen 7 5800X3D Linux performance for looking at the effectiveness of the AMD 3D V-Cache under Linux when now using the very latest Linux kernel along with other new/updated benchmarks of the past several months. While already being very impressed by the performance of AMD EPYC Milan-X since those 3D V-Cache server CPUs launched earlier this year, here is a fresh round of Linux benchmarks looking at the EPYC 7763 vs. 7773X performance when running on a development snapshot of Ubuntu 22.10 paired with the Linux 6.0 development kernel and other newer software packages for a very up-to-date look at the performance potential on the server side.

https://www.phoronix.com/review/ubuntu-2210-milanx

[rhysperry111]

Would be cool to see some general (GPU + CPU) benchmarks at the end of the year for newer hardware comparing all the changes that have happened in both userspace and kernel since the start of the year (I'm looking at you, MGLRU and mesa's current optimisation spree)

[V1tol]

I wonder if this test env had this patch applied.

[Michael]

I wonder if this test env had this patch applied.

No, this testing was done earlier in the month and that patch only came out last week / mainlined Monday.

[S.Pam]

Id love some gaming benchmarks and compare those against 7950x and the latest Intel ones .

[JEBjames]

Michael

Interesting benchmarks.

Really weird results? The oneDNN, PostgreSQL and Apache Cassandra results were much worse on the dual-processor systems?

Also typo? "with the consumer Ryzen 7 6800X3D" should be 5800X3D?

[Michael]

Michael

Interesting benchmarks.

Really weird results? The oneDNN, PostgreSQL and Apache Cassandra results were much worse on the dual-processor systems?

Also typo? "with the consumer Ryzen 7 6800X3D" should be 5800X3D?

Thanks.

Not really, for some workloads depending upon configuration with NUMA configuration can lead to lower performance than 1P.

[JEBjames]

Not really, for some workloads depending upon configuration with NUMA configuration can lead to lower performance than 1P.

I new some workloads were slower...still shocking to see almost half the performance with 2x the CPU power.

Great benchmark. Hopefully, at least for PostgreSQL, someone will dig under the hood. Clearly an opportunity to optimize something there.

[F.Ultra]

I wonder if this test env had this patch applied.

People are putting way too much hope into that patch. The problem it fixes will only appear in very extremely special circumstances.

[carewolf]

People are putting way too much hope into that patch. The problem it fixes will only appear in very extremely special circumstances.

True, but it remains to see if any of those happens in any of our regular benchmarks. That would be interesting to see.