Testing Six Different Linux Distributions On The Intel Core i9 13900K "Raptor Lake"

[phoronix]

Phoronix: Testing Six Different Linux Distributions On The Intel Core i9 13900K "Raptor Lake"

For those wondering about the out-of-the-box performance of different modern Linux distributions when running the new Intel Raptor Lake processors, here are six different distributions running on the current flagship Core i9 13900K processor. Tested this round was CentOS Stream 9, Clear Linux, Debian Bookworm (Testing), EndeavourOS, Fedora Workstation 37, and Ubuntu 22.10.

https://www.phoronix.com/review/linux-2022-raptorlake

[otoomet]

Would be interesting to know what are the factors behind the different performance of the distros. I can see Clear Linux and Bookworm use the performance governor, while others use powersave. Are there other power-related tuning differences? I'd love to see similar tests run in settings where maximum power is fixed to something reasonable, and see if it makes a difference.

[tildearrow]

No openSUSE?

What's the exact model of your memory modules and graphics card?

[rene]

No T2 Linux? https://t2sde.org disappointing :-/

[Kjell]

No NixOS?

[Linuxxx]

It should be noted that most Linux distros these days tend to set the THP (Transparent Huge Pages) to "always", whereas Ubuntu defaults to the sane upstream default of "madvise", which means that hugepages are only provided if explicitly requested by an user-space application.

While always enabling hugepages can definitely help with delivering greater performance (mostly in benchmarks), it can also cause huge latency-spikes introduced by the khugepaged daemon.

Therefore, even RedHat suggests not enabling them by default if aiming for a low-latency Linux system (which includes gaming, among general desktop smoothness).

[Danny3]

I wish it was tested Debian Unstable instead of Testing.
Debian with KDE Plasma insteaf the default Gnome would also be very nice to see.
And I with the Debian developers would take more optimizations from Clear Linux and other distros that excel in some test.

Anyway, thanks for the benchmarks!

[leoheck]

No TempleOS?

[stormcrow]

It should be noted that most Linux distros these days tend to set the THP (Transparent Huge Pages) to "always", whereas Ubuntu defaults to the sane upstream default of "madvise", which means that hugepages are only provided if explicitly requested by an user-space application.

While always enabling hugepages can definitely help with delivering greater performance (mostly in benchmarks), it can also cause huge latency-spikes introduced by the khugepaged daemon.

Therefore, even RedHat suggests not enabling them by default if aiming for a low-latency Linux system (which includes gaming, among general desktop smoothness).

Tuning for benchmarks is "the new black"... the current fashion. Doesn't matter how applicable the changes are to practical day to day usability so long as synthetic benchmarks look good to credulous blog readers.

Let's all lock the governor to "performance"! Nevermind it kills battery life on laptops! Force THP to "always" because it let's us look good on irrelevant performance numbers while possibly sacrificing usability and sacrificing memory allocation integrity.

Beware the configuration du jour of the 'consensus'. Read the documentation, and if you need to the source code itself, instead. In fact, if you're hosting a database system you don't want THB enabled as one of many examples.

[schmidtbag]

No Rebecca Black OS?