←back to thread

Rust CUDA Project

(github.com)
146 points sksxihve | 4 comments | | HN request time: 1.158s | source
1. porphyra ◴[] No.43656375[source]
Very cool to see this project get rebooted. I'm hoping it will have the critical mass needed to actually take off. Writing CUDA kernels in C++ is a pain.

In theory, since the NVVM IR is based on LLVM IR, rust in CUDA should be quite doable. In practice, though, of course it is an extreme amount of work.

replies(1): >>43657023 #
2. pjmlp ◴[] No.43657023[source]
Unless NVIDIA actually embraces this, it will never be better than the C++, alone given the whole IDE integration, graphical debugging and libraries ecosystem.

Unless one is prepared to do lots of yak shaving, and who knows, then NVIDIA will actually pay attention, like it has happened with CUDA support for other ecosystems.

replies(1): >>43663898 #
3. Asraelite ◴[] No.43663898[source]
Yeah, I would prefer to see this effort being done for AMD GPUs. They're not as powerful, but at least everything is open source and developer friendly. They should be rewarded for that.
replies(1): >>43686620 #
4. porphyra ◴[] No.43686620{3}[source]
AMD GPUs actually have fairly powerful hardware. However, despite being open source, calling them "developer friendly" is rather generous. Just look at how TinyCorp struggled with AMD ROCm before resorting to writing a completely sovereign stack.