Most active commenters
  • viraptor(3)

←back to thread

OpenGL 3.1 on Asahi Linux

(asahilinux.org)
512 points simjue | 11 comments | 06 Jun 23 13:46 UTC | HN request time: 0.001s | source | bottom
Show context
nightski ◴[06 Jun 23 14:20 UTC] No.36213208[source]▶
This is great work and I commend it. But in other threads people are acting like Asahi Linux hardware support is 100% complete. My fear is that if I were to go this route and purchase the hardware I'd be seeing fraction of the performance and capability I would in Mac OS. To be honest this blog post seems like the project has a long ways to go, not that it is nearly completion.

I just can't justify buying hardware from a company that is so hostile to developers and hackers as nice as it may be.

replies(9): >>36213287 #>>36213309 #>>36213359 #>>36213764 #>>36213841 #>>36214046 #>>36214150 #>>36214656 #>>36221582 #
1. viraptor ◴[06 Jun 23 14:26 UTC] No.36213309[source]▶
> I'd be seeing fraction of the performance and capability

You'd temporarily lose some hardware support (documented) while it's being worked on. But I'm not sure why you expect losing performance? This is running native code. Same binary will run the same on both systems (+/- the llvm version differences).

replies(3): >>36213391 #>>36213406 #>>36213726 #
2. nightski ◴[06 Jun 23 14:31 UTC] No.36213391[source]▶
Specifically GPU drivers, which can dramatically impact performance. Especially if I am attempting to run any kind of ML workload from Linux. I'm assuming it's basically a non-starter at this point and one is forced to use Mac OS.
replies(1): >>36213492 #
3. zamadatix ◴[06 Jun 23 14:32 UTC] No.36213406[source]▶
E.g. the performance you can get out of the GPU at the moment is a subset of what you can get out of the hardware. Or as another more generic example, until this latest release CPU boost states weren't enabled due to lack of proper cpuidle driver which resulted in regressed single thread performance.

There is nothing inherent about running Linux that will require it be slower, in some cases it will/is even faster, but the lack of everything being fully supported does actually impact performance right now. It has been getting better with time.

replies(3): >>36213752 #>>36213839 #>>36217581 #
4. viraptor ◴[06 Jun 23 14:37 UTC] No.36213492[source]▶
Yes. I put that in the capability rather than performance basket though. As in, you can't access the compute shaders yet, rather than: you can and they're slower.
5. Timon3 ◴[06 Jun 23 14:52 UTC] No.36213726[source]▶
Does the CPU run at similar frequencies between Mac OS and Linux (since they're writing their own drivers this isn't guaranteed)? Is the scheduling done similarly? Are there any special hardware modes you have to activate with e.g. binary blobs?

There are a bunch of factors that could affect performance even under the same OS (try underclocking your CPU or play around with schedulers). Given the mostly non-existent documentation from Apple I'd strongly suspect that average-case performance will stay worse on the Linux side for a long time.

replies(1): >>36217530 #
6. rowanG077 ◴[06 Jun 23 14:54 UTC] No.36213752[source]▶
Even with the regressed performance it beat osx in a ton of workloads.
7. kytazo ◴[06 Jun 23 14:58 UTC] No.36213839[source]▶
CPU wise at least its been on par if not better from its first days. At least this is what the various benchmarks at the time showed.

Makes you wonder about how the rest of the system components will compare when they're finished.

8. hedora ◴[06 Jun 23 19:02 UTC] No.36217530[source]▶
The currently top-rated top-level HN comment goes into those details. This release significantly improves CPU power management, to the point where it should be similar to MacOS.

Some of this stuff is handled by binary blobs that get installed/upgraded by MacOS, and are running by the time Linux boots.

With the previous release, power per watt and absolute performance were already better than high-end x86 laptops, so if your question is "is this faster and more power efficient than my other Linux laptop?", the answer is probably yes.

If you're asking if it will beat MacOS's perf/watt in all scenarios, the answer will be no for a long time. However, it is probably already beating MacOS in many practical scenarios.

replies(1): >>36221747 #
9. hedora ◴[06 Jun 23 19:05 UTC] No.36217581[source]▶
Has anyone tried Steam under Linux? It's quite bad under MacOS (2016 casual games stutter, and there is no 32 bit support).

I know rosetta doesn't exist under Linux, but I don't see any options to run steam / proton under rosetta either.

replies(1): >>36219336 #
10. viraptor ◴[06 Jun 23 21:27 UTC] No.36219336{3}[source]▶
> Has anyone tried Steam under Linux?

Through FEX, yes https://vt.social/@lina/110068264684987710

> I know rosetta doesn't exist under Linux

It does these days! https://developer.apple.com/documentation/virtualization/run...

11. MBCook ◴[07 Jun 23 01:46 UTC] No.36221747{3}[source]▶
Right. Everything I’ve seen implies they’re both using the hardware the same way (no “limp mode” or such).

Linux tends to be faster because the kernel is just far better optimized than Apple’s appears to be. Massive sums of money (and person hours) have been spent speeding up Linux networking, file systems, scheduling, etc. and it was all sent back to mainline.

Apple’s kernel team can’t possibly compete based on resources alone. They do their best but MS, Google, Amazon, Redhat, and so many others are constantly improving Linux to squeeze out every last drop of performance.

So in many ways Linux is better optimized. I remember one of the developers posting a few weeks ago about just how much faster code compilation was under Linux, because the file system layer is so much better. It was like 6x or something. Sort of an accidental ideal benchmark for stressing that.

But you won’t be disappointed.