AMD EPYC 8534P / EPYC 8534PN Benchmarks - Siena Delivers Incredible Value & Energy Efficiency For Linux Servers

[phoronix]

Phoronix: AMD EPYC 8534P / EPYC 8534PN Benchmarks - Siena Delivers Incredible Value & Energy Efficiency For Linux Servers

Back in September AMD rounded out their Zen 4 server product line-up with the EPYC 8004 "Siena" processors that are optimized for delivering excellent energy efficiency with leading performance-per-Watt and maximizing value both for initial server costs and ultimately the TCO. These single-socket server chips are quite interesting for a range of workloads form the edge to networking and more. In today's article are benchmarks of the top-end AMD EPYC 8534P and EPYC 8534PN 64-core server processors and showing how they can take on Intel Xeon Platinum "Sapphire Rapids" in raw performance and blow the competition out of the water when it comes to the incredible performance-per-Watt and value.

https://www.phoronix.com/review/amd-epyc-8534p-benchmarks

[JEBjames]

Michael

Great benchmarks

Typo page 2

"where it cane to the latency" should be "came"

extra white space page 6
"power determinism , there was" (extra space before the comma).

Thanks for crunching these benchmarks. The power efficiency charts...wow it's a shockingly difference. Not even close.

[Scramblejams]

Great article, thanks! Did you happen to capture any idle power consumption numbers? I’m always interested in those.

[Mark Rose]

I see you benchmarking Cassandra. Have you considered benchmarking https://www.scylladb.com/ ? It's a drop-in replacement, but scales much better, and a lot of large Cassandra users are switching to it.

[Michael]

Great article, thanks! Did you happen to capture any idle power consumption numbers? I’m always interested in those.

The minimum readings are in effect idle but no long term idle power readings, can try to collect osme next time.

[Michael]

I see you benchmarking Cassandra. Have you considered benchmarking https://www.scylladb.com/ ? It's a drop-in replacement, but scales much better, and a lot of large Cassandra users are switching to it.

Only faintly heard of it, so many databases these days... Will check it out and add a test profile for it if it pans out well. Thanks!

[schmidtbag]

Impressively low power draw for 64 cores.

[stormcrow]

Meanwhile another bug discovered in AMD Zen 3 (maybe others gens?) CPUs. The write up is geared towards programmers who can read code. TL;DR: The bug is in the CPU rather than the C, Rust, or Python code. Python just doesn't trigger it because the interpreter isn't using the faulty instruction. Bug has been submitted to AMD for analysis and correction or workarounds. It'll require a microcode fix or programmers to avoid the faulty instruction (not likely on that front - AMD is going to have to issue a patch). I just thought I'd link to it here because it can affect anyone as it's a memory operation, not just programmers running something esoteric.

Rust std fs slower than Python!? No, it's hardware!

[trifud]

How does it compare to ARM in performance-per-Watt?

[Anux]

I have the feeling that with every new benchmark Intels efficiency gets worse.

Meanwhile another bug discovered in AMD Zen 3 ...

Nice read THX.

How does it compare to ARM in performance-per-Watt?

Pretty similar I guess, but some benchmarks would be nice. I don't think Michael has any high core count ARMs, he mostly tested AWS-systems but without power consumption obviously. https://www.phoronix.com/review/graviton3-amd-intel