Substitute to forced Nvidia graphics card

[Bellisario]

From the videos I saw, I started loving ALiEn, but the problem is that I can't install it because I don't have an Nvidia graphics card. Someone know if there is a way to run it equally or there are other solutions that I can consider?

[chrxh]

This would require converting the CUDA code into OpenCL. I think there are tools for this. If someone knows more about it, that would be a great!

[Ccode-lang]

Hi, I was browsing with this issue in mind and found this: https://github.com/ROCm-Developer-Tools/HIPIFY

I don't know how well this might integrate with the current code at the moment but we shouldn't have to rewrite all the cuda code.

[chrxh]

HIP sounds promising. That could be an option for the future.

[Foxlum]

HIP sounds promising. That could be an option for the future.

This is very much possible now that windows has HIP support (The runtime is universal with RDNA2, and 3 on windows), and is actually usable. (as in buildable and running, but it doesn't have the required libraries for stuff like pytorch or similar, but since this is raw CUDA code starting with HIPIFY is a good start into a HIP port that is universal for both)

[CrispyPin]

https://github.com/vosen/ZLUDA is probably also worth trying, it lets you run CUDA applications on AMD graphics.