Why Linux is picking up at a rapid pace for AI usage? | Hacker News Reader

Why Linux is picking up at a rapid pace for AI usage?

5 pointskeshwanianup7 comments

GNU/Linux has a best potential like an OS to be a layer beneath the AI usage (for example agentic), lightweight, good compatibility, for CLI-only use etc. AI on side, I prefer GNU/Linux as a EDC desktop too, especially with KDE if you want to getting along after Windows

AI + nix makes it so easy to make reproducible personalized GUI software that generally doesn't just up and brick on you.

Well, I'm a dev so I'm already biased towards linux and I've been living in the terminal for a while (AI helps a lot with sysadmin/infra stuff too! Having it crunch away at my own microvm framework rn), but I'm literally just vibe coding my GUI stuff rn and it's great

You say 'picking up' like anyone would ever use Windows to run hermes or even an AI model.

Why not? There are plenty running on Windows.

windows is not the ideal os for developers, prefer using one close to your cloud environments, don't shrug off the additional complexity of wsl, it's not the same

windows cannot be made as safe either, simple things like running agents as their own user rather than your user

What? Do people use WSL to pretend they are on Linux? Is that a thing??

Noticed Ollama has a Windows version, ok, I just wonder why, but that is for desktop, but WSL? Does anyone really use that? And why?

Nothing I program is OS-specific, and all the software I use is cross-platform. So, I'm not sure what "not ideal for developers" means in this context. I'm curious what workflows you're using where the OS makes a difference. One area I can think of is GUI programs where you're targeting OS-specific APIs, for the best performance.

Regarding ML/AI or other computational programming: My thought is: Does this have an nvidia GPU with an up-to-date driver so my CUDA kernels work and What SIMD width is the CPU? Can I run 512-bit SIMD like AVX-512 for 16 x f32 computations at once? The OS doesn't matter here as long as it provides threads, a scheduler, and an allocator, which all the major ones do. And that my toolchain can compile for it.

Reply on news.ycombinator.com