CachyOS Explores Optimizing Its Kernel With AutoFDO

[mb_q]

Thank you, Peter and the CachyOS team for pushing the boundaries for performance optimizations on the Linux desktop! That's the spirit!

The problem with pushing boundaries is that you may be in an a diminishing returns territory when like exponential effort is needed for linear improvement, and, what's often worse, gains are totally unpredictable. Image benchmarking something when the final assembler depends on the age of thermal paste on the computer that does compilation...

[ms178]

The problem with pushing boundaries is that you may be in an a diminishing returns territory when like exponential effort is needed for linear improvement, and, what's often worse, gains are totally unpredictable. Image benchmarking something when the final assembler depends on the age of thermal paste on the computer that does compilation...

I'd question that line of reasoning as the CachyOS team verifies these kind of changes with a benchmark suite with different workloads. While there might be diminishing returns and more effort to get there which equals to more time and effort on the distro side, it is still the vastly superior (and more efficient) approach if the distribution does this kind of work for a better out-of-the-box experience instead of shifting all of this time and effort to each individual user. And the benchmarks showed a significant improvement in memory-bound y-cruncher and still an 1 - 4 % improvement in everything else, I am glad that CachyOS doesn't shy away from that work as these improvements add up.

From the bigger distros with paid staff and paying customers, I'd expect nothing less to be honest.