AMD EPYC 9374F Linux Benchmarks - Genoa's 32-Core High Frequency CPU

Written by Michael Larabel in Processors on 15 November 2022. Page 14 of 14. 21 Comments

Across the span of around 200 benchmarks carried out, above is a look at the recorded CPU power consumption over the entire span of day of benchmarking for each CPU. The EPYC 9374F had an average power consumption of 195 Watts and a recorded peak of 275 Watts. When in the power determinism mode was where the EPYC 9374F had a 220 Watt average and 327 Watt peak for this mix of single and multi-threaded workloads carried out. Surprisingly the EPYC 9374F had lower overall power consumption on average than the EPYC 75F3 predecessor in the 1P configuration. In the 1P configuration was also lower power consumption on average than the Xeon Platinum 8362 and 8380 that it was usually destroying with better performance.

When taking the geometric mean of all the benchmarks that successfully ran on all tested processors, this is how things shake out. Keep in mind there were around 200 benchmarks carried out with a mix of single and multi-threaded workloads with some tests that scale exceptionally well and others that struggle to effectively make use of 384 threads afforded by the EPYC 9654. Meanwhile some workloads that make effective use of AVX-512 find their performance up to multiple times better with Genoa. In any event this overall look at performance shows the shear strength of EPYC Genoa processors with their Zen 4 architectural improvements most notably led by AVX-512, twelve channels of DDR5 system memory, elevated TDP, and other enhancements leading to huge generational uplift.

From this geo mean, the EPYC 9374F in a single socket configuration nearly matched the Xeon Platinum 8380 2P... The EPYC 9374F was running at 1.5x the speed of the EPYC 75F3, in large part thanks to many AVX-512 workloads relevant in the server and HPC space. Particularly if you are frequently running many AVX-512 heavy workloads, that alone should be reason enough to warrant an upgrade to AMD EPYC 9004 series hardware. In a 2P configuration, the EPYC 9374F was still 1.5x the speed of the EPYC 75F3 2P Milan configuration -- or 1.51x the speed of the Xeon Platinum 8380 2P setup with this large variety of benchmarks carried out.

For those with frequency sensitive workloads more than the ability to scale up to 192 cores / 384 threads with the flagship EPYC 9654 in a two socket configuration, the EPYC 9374F is a very interesting part for the 32-core sweet spot and expected to sell for less than $5k. Current platform costs with pricey DDR5 system memory eat away some of the excitement, but overall the EPYC 9004 series performance continues to prove astounding. Especially if leveraging AVX-512 in your production workloads, the incredible performance and power efficiency should easily prove a worthwhile investment and make for a compelling TCO.

Like with the EPYC 9554 and 9654 testing, switching over to the power determinism mode from the default performance determinism allowed for an extra boost of performance. The EPYC 9374F in the power determinism mode saw its performance increase by just under 4% overall but some workloads benefited more significantly from this alternate mode.

Those wishing to go through all of the benchmarks in full from this Ubuntu 22.10 + Linux 6.0 comparison across the wide span of Intel and AMD server processors can find the data via this result page for further analyzing all of the individual workloads and the complementary performance-per-dollar and performance-per-Watt metrics. Thanks to AMD for providing the Genoa review samples and stay tuned for further benchmarks looking at the performance with varying number of memory channels, compiler benchmarks, and other tuning tests.

If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal tips are also graciously accepted. Thanks for your support.


Related Articles
About The Author
Author picture

Michael Larabel is the principal author of Phoronix.com and founded the site in 2004 with a focus on enriching the Linux hardware experience. Michael has written more than 20,000 articles covering the state of Linux hardware support, Linux performance, graphics drivers, and other topics. Michael is also the lead developer of the Phoronix Test Suite, Phoromatic, and OpenBenchmarking.org automated benchmarking software. He can be followed via TwitterLinkedIn,> or contacted via MichaelLarabel.com.