I've been keeping an eye on Mojo/Max as it's one of the only things out there that doesn't depend on the Cuda library.
Blockers for me right now, are:
1) still closed source
2) it doesn't support my low-end hardware (just because I don't have a 5090 or whatever doesn't mean I shouldn't be able to do GPU compute things)
I'm guessing if those barriers are ever removed it'll probably be years from now. But hopefully it'll inspire other languages/ecosystems to take on their own Cuda replacement.
1. It'll become open-source with 1.0, as the article mentions.
2. Just requires someone (or you) to write a kernel for your GPU, which is done in Mojo itself. I'd double check the supported GPUs or if someone else has already done it.
bobajeff|2 months ago
Blockers for me right now, are:
1) still closed source
2) it doesn't support my low-end hardware (just because I don't have a 5090 or whatever doesn't mean I shouldn't be able to do GPU compute things)
I'm guessing if those barriers are ever removed it'll probably be years from now. But hopefully it'll inspire other languages/ecosystems to take on their own Cuda replacement.
16bitvoid|2 months ago
2. Just requires someone (or you) to write a kernel for your GPU, which is done in Mojo itself. I'd double check the supported GPUs or if someone else has already done it.
Rochus|2 months ago