Intel Continues To Demonstrate The Importance Of Software Optimizations: Clear Linux + Xeon Max Benchmarks
While the recently released Ubuntu 23.10 is bringing some performance improvements to Intel Xeon Max / Sapphire Rapids, Ubuntu Linux still isn't delivering the best possible out-of-the-box server performance. For that Intel continues to show the importance of software optimizations with the likes of their in-house Clear Linux platform as well as the likes of CentOS Stream having more sensible defaults. Here is a look at the Intel Xeon Max 9480 performance across Ubuntu 22.04 LTS, Ubuntu 23.10, CentOS Stream 9, Fedora Server 39, and Clear Linux 40130.
More Linux distributions are fortunately looking at x86-64 micro-architecture feature levels, function multi-versioning, and other efforts to improve the robustness and performance of modern hardware on Linux. There's also been more upstream optimizations from Intel and other vendors reaching the mainline Linux kernel, GCC, Glibc, and other components over the past year. While Intel hasn't had any big breakthroughs to announce around their in-house Clear Linux distribution recently, it continues to serve as a flagship for Linux x86-64 performance -- and just not for Intel CPUs but AMD platforms too -- with its performance-oriented defaults, various out-of-tree patches tuned by Intel engineers, carrying out PGO / LTO / FMV and other compiler features, etc.
Using the dual Intel Xeon Max 9480 (Sapphire Rapids with HBM2e) processors on the Supermicro Hyper SuperServer SYS-221H-TNR with X13DEM motherboard, I carried out some fresh benchmarks seeing how the Xeon Max out-of-the-box performance compared to a few modern options. The same hardware was used throughout testing with just swapping the out-of-the-box Linux distribution being tested. The operating systems tested for this article included:
- Ubuntu 22.04.3 LTS
- Ubuntu 23.10
- CentOS Stream 9
- Fedora Server 39
- Clear Linux 40130
From there the Xeon Max performance was looked at across a wide variety of different workloads.