Intel Linux Kernel Optimizations Show Huge Benefit For High Core Count Servers

Written by Michael Larabel in Software on 29 March 2023 at 01:00 PM EDT. Page 1 of 5. 19 Comments.

Earlier this month I wrote about Intel engineers working on more big optimizations to the Linux kernel with a focus on enhancing the kernel's performance at high core counts. The numbers shared then were very promising and since then I've had more time looking at the performance impact of Intel's stellar software optimization work and its impact on real-world workloads. Here is a look at how Intel's pending kernel optimization patches are a huge deal for today's high core count servers.

Intel Linux kernel optimizations
A teaser for the article ahead... With Intel CPU core counts going up, so are the Intel efforts to optimize the Linux kernel for greater scalability.

See the article earlier this month for what set of this latest benchmarking storm: Intel Continues With More Big-Time Optimizations To The Linux Kernel. Following that, I set off to run many different benchmarks for seeing how these optimization patches become extremely important with today's high core count servers.

This article is looking at Intel's scalability improvements to the Linux kernel using their Sapphire Rapids reference server with two Xeon Platinum 8490H processors.

For concisely looking at the cumulative impact of their pending work, I used Intel's latest Clear Linux build that has those aforementioned patches and their other in-progress tuning/concurrency work. I compared the performance of Intel's Clear Linux optimizations against that of Ubuntu 23.04 in its current development state. But I didn't just do a 2-way comparison but rather tested each distro/kernel across varying thread counts to show how Intel's kernel optimization effort is really paying off for high core count servers.

Intel Linux kernel optimizations

This round of testing was carried out on both operating systems using Intel's Eagle Stream server with two flagship Intel Xeon Platinum 8490H "Sapphire Rapids" processors. From there both operating systems / kernels were tested at 240 threads (the default capacity; 120 cores + SMT for the two 8490H processors), 120 threads (disabling SMT/HT), 60 threads, 30 threads, 15 threads, 8 threads, and 4 threads for looking at the scaling performance across a wide variety of workloads. At 120 threads and lower it was all physical cores being tested with Hyper Threading disabled.

Clear Linux, Ubuntu SPR Scaling

The results of this Intel kernel tuning effort really speak for themselves... Not only does it mean a huge win for Intel Xeon Scalable servers with dozens of cores or hitting 200+ threads in a dual socket server, but these optimizations apply for AMD hardware too... Next week I will have up similar benchmarks with AMD EPYC Genoa where the results are even more compelling at 384 threads. With that said, let's get right to the results.

Related Articles