4 ms·

Asahi linux has most things working, except the accelerated GPU. Just in the last week or so a triangle was rendered, so progress is has been made. Sadly whil

by sliken 4y ago

Asahi linux has most things working, except the accelerated GPU. Just in the last week or so a triangle was rendered, so progress is has been made. Sadly while apple has 128 bit wide memory on the low end (66-100GB/sec on the mac mini and mba), 256 bit wide (200GB/sec), 512 bits wide (400GB/sec), and 1024 bits wide (800GB/sec). Nothing wider than 128 bits looks to be coming to standard laptops or desktops in the non-apple world. I don't really count HEDT chips like the threadripper pro, since they are very expensive and very limited, and burn many 100s of watts.

criddell 4y ago
Asahi Linux is, for me, the most interesting thing going on in Linux these days.
- sliken 4y ago
  Agreed. I'd likely have a studio today if the GPU port worked, but unless it comes out RSN I'd likely wait for the M2 based studio to come out. The chip has other magic inside as well. I'm hoping Linux continues to implement support for the MatMul instruction (not just vector multiplies), 16 trillion op neural engine, various encode/decode video accelerators, etc. I've heard vague references to compressed swap to help make the most of limited ram (m1 limit was 16GB).
- dmix 4y ago
  What package manager/distro will Asahi Linux be using/basing off of?
  - viraptor 4y ago
    It's based on arch, but there are already people who ported the necessary changes to nix and something else I can't recall. The idea though is that once things get upstreamed, any distro will work.
    - dmix 4y ago
      Perfect, looking forward to when it's stable.
fomine3 4y ago
Massive memory bandwidth is for GPU, so not matter unless you utilize iGPU.
- sliken 4y ago
  I disagree, Anandtech tested the M1 max (409GB/sec bandwidth), and the CPUs managed 224GB/sec with the faster cores, and a total of 243GB/sec with fast and slow cores. Sure that's quite a bit less than 409GB, but there's other major parts of the chip like the GPU, video codec acceleration, and neural engine. Seems pretty unique and useful to me, not everything fits in cache.
  - fomine3 4y ago
    That's true that CPU core consumes massive available bandwidth on AT's test, but it doesn't mean that such bandwidth is used for real CPU usage. I don't oppose memory bandwidth is meaningful for total SoC including iGPU/ML/Video parts.

by sliken 4y ago

criddell 4y ago

Asahi Linux is, for me, the most interesting thing going on in Linux these days.

sliken 4y ago
Agreed. I'd likely have a studio today if the GPU port worked, but unless it comes out RSN I'd likely wait for the M2 based studio to come out. The chip has other magic inside as well. I'm hoping Linux continues to implement support for the MatMul instruction (not just vector multiplies), 16 trillion op neural engine, various encode/decode video accelerators, etc. I've heard vague references to compressed swap to help make the most of limited ram (m1 limit was 16GB).
dmix 4y ago
What package manager/distro will Asahi Linux be using/basing off of?
- viraptor 4y ago
  It's based on arch, but there are already people who ported the necessary changes to nix and something else I can't recall. The idea though is that once things get upstreamed, any distro will work.
  - dmix 4y ago
    Perfect, looking forward to when it's stable.

fomine3 4y ago

Massive memory bandwidth is for GPU, so not matter unless you utilize iGPU.

sliken 4y ago
I disagree, Anandtech tested the M1 max (409GB/sec bandwidth), and the CPUs managed 224GB/sec with the faster cores, and a total of 243GB/sec with fast and slow cores. Sure that's quite a bit less than 409GB, but there's other major parts of the chip like the GPU, video codec acceleration, and neural engine. Seems pretty unique and useful to me, not everything fits in cache.
- fomine3 4y ago
  That's true that CPU core consumes massive available bandwidth on AT's test, but it doesn't mean that such bandwidth is used for real CPU usage. I don't oppose memory bandwidth is meaningful for total SoC including iGPU/ML/Video parts.