This might be controversial, but I don't think performance between distributions is really worth considering unless you have a very niche hardware requirement.

Features and community really make the difference.

First of all, I think it's important to state that the 'default' settings that distros ship with and thus you're met with right after a fresh installation try to target a sweet spot in which a lot more besides performance is considered; reliability, security, stability etc. However, depending on your workload and your hardware configuration, it's possible to have it more optimized towards performance through 'tinkering-means'.

However workload, metrics, system and hardware configuration provide so much variability that an exhaustive comparison between distros is just hard to do right. It's possible to find some on Phoronix (and ~~Reddit~~), but testing it yourself on your own hardware is a lot more valuable.

Still, it's possible to draw some basic conclusions based on the available data and common sense:

• Newer versions of the kernel generally have optimizations related to performance, especially for newer hardware. So -for performance sake- it makes sense to pick a distro that always tries to stay as close as possible to the latest kernel release.

• Overhead is in almost all cases detrimental to performance, so more minimal systems seem to score better. A lot of distros offer an ISO that's meant for minimal installations, so those are perhaps worth checking out.

• Compiling yourself (with performance optimizations) or using packages that have been compiled with performance optimizations in mind provide significant improvements that might be worth your time.

Beyond these three it becomes very murky, real quick. I guess (custom) kernel patches/optimizations are worth a mention, but you would have to benchmark it yourself on your own hardware to see if they're even worth the hassle (spoiler alert: some of them should, but it's best to stay objective and without any expectations regarding them).

I'd like to end this with naming some distros that might be worth mentioning in this discussion: Arch, CachyOS, Clear Linux, Garuda, Gentoo, Nobara and PikaOS.

Not sure where to find those kind of benchmarks, but you may want to look at Gentoo. It compiles everything locally, optimising specifically for the hardware.

EDIT: Best I could find for benchmarks: https://www.phoronix.com/review/spring-2020-distros

Do not choose/judge a distribution based on performance benchmarks (this scenario makes them grossly unscientific). More important considerations to measure a distribution are things like development roadmaps, previous history, and philosophy.

Software considerations like Wayland or Xorg or Linux kernel versions (also whether its a zen kernel or not) are also important. Hardware support depends on how up to date the linux-firmware package of the distro is or if you're running a libre kernel or not. These are a lot more useful than a vague definition of "performance" which if we take at face value means that only Gentoo would be on top.

I've just bounced into voidlinux and I'm very impressed by how complete* it is and how snappy. So that's not a very objective measure, but my (old) battery life has jumped from 1.5h on fedora-38 to 2h+ under void (very similar workload). I think that's an indication of how light-on the hardware the OS is - perhaps it's because it's not running systemd. If you're looking to squeeze more out of your hardware, take a look at void!

* the only thing missing from my manifest is mythtv - firefox, libreoffice, emacs syncthing etc and it's all the latest

I think clear Linux is optimised for Intel chips. Not sure if there is an equivalent for amd

DJ Ware adds detailed performance statistics to his reviews.

I don't think it's very useful. Easily changeable default choices should make the most notable differences. The filesystem for example. Another big factor should be the configuration and the version of the kernel and it probably changes more between updates than between distributions.