447
CachyOS Is Now the Most Popular Desktop Distro on ProtonDB (boilingsteam.com)
submitted 6 months ago by to c/linux_gaming@lemmy.world
129 comments fedilink hide all child comments
 

It is also first in the Distrowatch rank

https://distrowatch.com/table.php?distribution=cachyos

I distro hopped to it from Bazzite a couple of months ago, and I could not be happier.

If you try the installer, be careful when selecting multiples DE/WM as the conflicts were not listed anywhere for the installation process.

Picking a single environment and then adding the others later was what worked for me.

you are viewing a single comment's thread
view the rest of the comments
[–] 16 points 6 months ago (15 children)

My local LLMs also seem to be slightly faster and for some reason now need less V-Ram.

This is likely due to zram being setup by default

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 15 child comments
    •
  • [–] 14 points 6 months ago (7 children)

    The hell is zram?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 7 child comments
    •
  • [–] 16 points 6 months ago* (2 children)

    Basically it compresses your data in the RAM. Needs a little more work form the CPU but it is still faster than swap. fyi

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 4 points 6 months ago (4 children)

    I had zram setup on my previous OS as well and on cachy the LLM didn't need to use it while on my old OS it did. My guess would be that the driver had a little less overhead.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 3 points 6 months ago (3 children)

    It would still use zram if it's setup, only way it wouldn't is if you gone out of your way to disable it. Combine zram and the bore scheduler and it's going to run better for sure.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 2 points 6 months ago (2 children)

    But if the LLM stays in the V-RAM or even just stays in normal RAM does it still benefit from zram? I thought that only helped when the ram was not enough.

    So I'd say it is more likely the bore scheduler + better drivers.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 4 points 6 months ago (1 child)

    It's all three! Cachy OS sets it up where it applies to both regular ram and vram. Even if just using a swap in ram, it will still be much faster than swapping to disk. Plus to the application, it looks like it has more ram/vram than there is physical ram/vram.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 1 point 6 months ago

    Yes but my previous OS had zram setup as well, unless PopOS was using a worse compression algorithm, zram was not the main player here. Also there is no such thing as zram for vram unless you are talking about hardware compression.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 1 point 5 months ago*

    Zram kills my LLM speeds, big time. I had to disable it or almost disable it (limiting to like 0.5GB) and move swap to the SSD to get my system usable.

    The issue is that LLM frameworks allocate a whole bunch of CPU RAM up-front, statically. So if I'm riiiight at the edge of the limit, that bit of RAM zram eats pushes me over to swapping, which utterly obliterates token gen speed and system responsiveness, and it never recovers once it starts.

    I think CachyOS helps hybrid inference from better default scheduling, though. This is another thing llama.cpp gets very picky about. More optimized default compilation flags, GPU libs and such would also go a long way.

  • permalink
  • fedilink
  • source
  • parent