Arch Linux CachyOS Benchmarks Of x86-64-v3 & x86-64-v4 Repositories
The Arch Linux based CachyOS Linux distribution aims to be a "blazingly fast and customizable Linux distribution" that is aggressive with its performance optimizations. CachyOS takes to leveraging compiler optimizations like Link-Time Optimizations (LTO), the BORE scheduler, and also offering package archives compiled for x86-64-v3 and x86-64-v4 in allowing the distribution's packages to be catered toward newer Intel and AMD processors. In this article is a comparison of CachyOS packages from their main archive, the x86-64-v3 optimized packages, and then the x86-64-v4 wares that can be beneficial for modern Intel Xeon and AMD EPYC / AMD Ryzen systems.
The x86-64-v3 micro-architecture feature level as a reminder is for Intel Haswell or AMD Excavator and newer processors. The x86-64-v3 level mandates AVX/AVX2 support, BMI1/BMI2, FMA, MOVBE, and various other newer instructions. Being able to assume Advanced Vector Extensions (AVX) support in particular can be a big win for the performance in sensitive applications.
It's x86-64-v4 that is currently the highest micro-architecture feature level and basically adds in AVX-512 requirements atop x86-64-v3. So with x86-64-v4 it's only compatible with Intel Skylake and newer server processors, Intel Core Rocket Lake CPUs with AVX-512 or Ice Lake / Tigerlake, and then all AMD Zen 4 processors.
Most Linux distributions are eyeing x86-64-v3 as a baseline for their future Linux distribution releases or at least going for x86-64-v2 and then augmenting their package selection with a x86-64-v3 package archive for those running on AVX-supported hardware. In the case of CachyOS they already offer x86-64-v3 and x86-64-v4repositories. Via /etc/pacman.conf you can configure your system for using a particular repository and there are also various CachyOS helper scripts available too.
For those curious about the performance impact, I recently ran some benchmarks on CachyOS using the HP Z6 G5 A workstation with AMD Ryzen Threadripper 7995WX processor. Being Zen 4, AVX-512 is supported and in turn works for testing both x86-64-v3 and x86-64-v4 repositories.