melihelibol

6 ms·

1.
by melihelibol 16d ago
It's still linux-only but doesn't require async. You should be able to execute and compose kernels synchronously.
2.
by melihelibol 16d ago
You don't need to use the CUDA (SIMT) programming model if you don't like it. The project includes cutile, which lets you program the GPU using tensors. It feels a lot like programming the GPU using numpy and triton.
3.
by melihelibol 16d ago
Integration with candle is already available! https://github.com/huggingface/candle/pull/3934/
4.
by melihelibol 4mo ago
Yes, the ownership-based safety model and tight Rust integration, plus a pretty fleshed-out host-side tensor API for different workloads (sync, async, replayable CUDA graphs). Though Triton's a Python DSL, so there's no official T
5.
by melihelibol 4mo ago
Burn's great, and we cite it in the paper's related work. Different layer, though: Burn is the high-level DL framework (tensor ops, autodiff, backends), whereas cuTile Rust is the kernel-authoring layer underneath. The closer neig
6.
by melihelibol 4mo ago
cuTile Rust lowers through CUDA Tile IR, which is NVIDIA-specific, so any of those targets (ROCm, Vulkan, Metal, NPUs, OpenCL) would be a new compiler backend. The programming model itself isn't NVIDIA-bound, but the lowering is.
7.
by melihelibol 4mo ago
I think the answer's actually yes. I'll give the short rationale here, but it's a deep topic and I'd be happy to take it further. The hot loop you're describing is the whole reason you'd want to reach for a sys
8.
by melihelibol 4mo ago
Yes! That's exactly the spirit. The readable, single-binary, kernels-included codebase is a big part of what makes Grout fun, and the antirez parallel is accurate. There's a parallel to ThunderKittens too, on the kernel-authoring
9.
by melihelibol 4mo ago
You got it right: This project exposes our tile programming model, whereas cuda-oxide exposes our lower level CUDA-like programming model. Our tile programming model is higher-level: It compiles to what looks like the CUDA-like programming
10.
by melihelibol 4mo ago
Hello, I built cuTile Rust and just posted the paper preprint. Happy to answer questions. TL;DR: Rust gives you fearless concurrency on the CPU, but GPU kernel programming still requires unsafe code. cuTile Rust carries Rust's ownershi
11.
Show HN: cuTile Rust: Safe, data-race-free GPU kernels in Rust(github.com)
106 points by melihelibol 4mo ago|18 comments
12.
Learning Algorithm Beats Ninentdo's Ninja Gaiden Boss by Trial and Error(youtube.com)
2 points by melihelibol 11y ago|0 comments