AMD EPYC Genoa/Genoa-X & Bergamo vs. Intel Xeon Sapphire Rapids On Ubuntu 23.10

[phoronix]

Phoronix: AMD EPYC Genoa/Genoa-X & Bergamo vs. Intel Xeon Sapphire Rapids On Ubuntu 23.10

While Ubuntu 23.10 isn't a long-term support (LTS) release and thus won't see too much exposure in the enterprise space, it's worthwhile today looking at the AMD EPYC and Intel Xeon server performance. It's interesting for a look ahead being just a few months until Ubuntu 24.04 LTS and with Ubuntu 23.10 incorporating GCC 13 that will also be the default compiler of Ubunu 24.04 among other close software package versions, the kernel not too far off from what will be in this next LTS release, and with Ubuntu 23.10's Linux 6.5 kernel bringing some nice performance optimizations. So with that said I recently wrapped up some fresh benchmarks looking at the current generation Intel Xeon Scalable Sapphire Rapids performance against AMD EPYC 9004 Bergamo, Genoa, and Genoa-X processors.

https://www.phoronix.com/review/ubuntu-2310-epyc-xeon

[geerge]

So the Xeon 8490H 2P's main competitor is the 9754 1P, which on average was ~10% slower for less than half the power consumption. The 8490H list price is $17k each, the 9754 list price is $12k, not to mention rack density, power density, platform costs. Factors behind in multiple ways, it's kind of impressive.

Sierra Forest needs to be god damned magic just for them to see parity in the distance.

edit: Would power determinism mode work for 1P? You've concluded that power determinism mode helped by ~10%, and the 9754 1P is ~10% behind the 8490H 2P...

[Michael]

So the Xeon 8490H 2P's main competitor is the 9754 1P, which on average was ~10% slower for less than half the power consumption. The 8490H list price is $17k each, the 9754 list price is $12k, not to mention rack density, power density, platform costs. Factors behind in multiple ways, it's kind of impressive.

Sierra Forest needs to be god damned magic just for them to see parity in the distance.

edit: Would power determinism mode work for 1P? You've concluded that power determinism mode helped by ~10%, and the 9754 1P is ~10% behind the 8490H 2P...

Power determinism works fine in 1P, just that there is only so much time in a day. Each of these server runs per CPU takes about 1 day. Been working on this data since last month and have only so much time to benchmark everything, so saved power determinism for 2P

[onlyLinuxLuvUBack]

Phoronix: AMD EPYC Genoa/Genoa-X & Bergamo vs. Intel Xeon Sapphire Rapids On Ubuntu 23.10

While Ubuntu 23.10 isn't a long-term support (LTS) release and thus won't see too much exposure in the enterprise space, it's worthwhile today looking at the AMD EPYC and Intel Xeon server performance. It's interesting for a look ahead being just a few months until Ubuntu 24.04 LTS and with Ubuntu 23.10 incorporating GCC 13 that will also be the default compiler of Ubunu 24.04 among other close software package versions, the kernel not too far off from what will be in this next LTS release, and with Ubuntu 23.10's Linux 6.5 kernel bringing some nice performance optimizations. So with that said I recently wrapped up some fresh benchmarks looking at the current generation Intel Xeon Scalable Sapphire Rapids performance against AMD EPYC 9004 Bergamo, Genoa, and Genoa-X processors.

https://www.phoronix.com/review/ubuntu-2310-epyc-xeon

bro this is the first dance battle ubuntu 24.04 vs ubunu ?
where do i get this ubunu ?

[darkbasic]

EPYC 9684X 2P in power determinism mode is over 48% faster than the Xeon Platinum 8490H 2P which consumes more. Impressive.