CentOS Stream & Clear Linux Achieve Greater Performance On 4th Gen Xeon Scalable Sapphire Rapids, EPYC Genoa

Written by Michael Larabel in Operating Systems on 3 February 2023 at 08:48 AM EST. Page 3 of 6. 9 Comments.
oneDNN benchmark with settings of Harness: IP Shapes 1D, Data Type: bf16bf16bf16, Engine: CPU. Clear Linux: Xeon 8490H 2P was the fastest.
oneDNN benchmark with settings of Harness: IP Shapes 3D, Data Type: bf16bf16bf16, Engine: CPU. Clear Linux: Xeon 8490H 2P was the fastest.

In some of the workloads tested, Intel's Clear Linux was delivering staggering performance advantages over the other tested Linux distributions. Clear Linux carries a number of AVX-512 optimized libraries that are used by default on capable hardware among other tuning already carried out by Intel engineers for Sapphire Rapids.

oneDNN benchmark with settings of Harness: IP Shapes 3D, Data Type: bf16bf16bf16, Engine: CPU. Clear Linux: Xeon 8490H 2P was the fastest.

The great Clear Linux leads with oneDNN did equate to increased CPU power consumption.

oneDNN benchmark with settings of Harness: Recurrent Neural Network Training, Data Type: bf16bf16bf16, Engine: CPU. Clear Linux: Xeon 8490H 2P was the fastest.
oneDNN benchmark with settings of Harness: Recurrent Neural Network Inference, Data Type: bf16bf16bf16, Engine: CPU. Clear Linux: Xeon 8490H 2P was the fastest.

The Clear Linux numbers though show the potential for not only Intel 4th Gen Xeon Scalable but also AMD 4th Gen EPYC when really maximizing the OS tuning and optimizations around AVX-512 and looking to exploit the full potential of modern x86_64 processors.

Cpuminer-Opt benchmark with settings of Algorithm: x25x. CentOS Stream 9: EPYC 9654 2P was the fastest.

In other benchmarks like Cpuminer-opt with some crypto algorithms the Clear Linux performance on Sapphire Rapids also shot out ahead of Ubuntu and CentOS Stream.

miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. Clear Linux: EPYC 9654 2P was the fastest.
miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. Clear Linux: EPYC 9654 2P was the fastest.

Clear Linux also had a solid showing when it came to the miniBUDE HPC benchmark.

miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. Clear Linux: EPYC 9654 2P was the fastest.
miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. Clear Linux: EPYC 9654 2P was the fastest.

In the case of miniBUDE the Clear Linux gains were while enjoying roughly the same power consumption as with the other Linux distributions, so it was a win for power efficiency too.


Related Articles