when will NVIDIA (CUDA) on windows will be supported

[Nyx2022]

hello,i'm interested that i found your lib,but i found that NVIDIA (CUDA) on windows is planned,but i wanna use this lib with NVIDIA (CUDA) supportedas soon as possible,so i wanna know when will it be finished?thank you

[dianlight]

Hi, I'm pleasantly surprised that someone is already interested in this project. I haven't reached the end of the alpha phase yet, and it's not finished yet (I'm writing the release scripts now! :)

In any case, CUDA for Windows should already work, but I haven't had time to test it yet. (I also don't have the hardware! :)
If you test it and give me feedback, I'd be happy to hear it! :)

[dianlight]

In any case, CUDA for Windows should already work, but I haven't had time to test it yet. (I also don't have the hardware! :)
If you test it and give me feedback, I'd be happy to hear it! :)

Sorry. Due to purego limitation now don't work and I can't say when. docs/ROADMAP.md has the details.

[Nyx2022]

Hi, I'm pleasantly surprised that someone is already interested in this project. I haven't reached the end of the alpha phase yet, and it's not finished yet (I'm writing the release scripts now! :)

In any case, CUDA for Windows should already work, but I haven't had time to test it yet. (I also don't have the hardware! :)

your idea and work，the most important of you deserve it

[Nyx2022]

In any case, CUDA for Windows should already work, but I haven't had time to test it yet. (I also don't have the hardware! :)
If you test it and give me feedback, I'd be happy to hear it! :)

Sorry. Due to purego limitation now don't work and I can't say when. docs/ROADMAP.md has the details.

ok,i have a doubt about the difference cuda support with cuda speed suppot,because i found it has supported cuda in windows in docs/ROADMAP.md

[dianlight]

CUDA support is handled directly by llama.cpp, so even on Windows the library is loaded and the GPU is initialized. The problem is that without being able to call the direct CUDA APIs, it is not possible to load the model into the GPU’s RAM, so essentially the support is there but it cannot be used. :(

[Nyx2022]

oh,got it!you are working on this

[Nyx2022]

Hi, I'm pleasantly surprised that someone is already interested in this project. I haven't reached the end of the alpha phase yet, and it's not finished yet (I'm writing the release scripts now! :)

In any case, CUDA for Windows should already work, but I haven't had time to test it yet. (I also don't have the hardware! :)
If you test it and give me feedback, I'd be happy to hear it! :)

hi.i'm testing in windows , but it now return false in isPlatformSupported() in platform_windows.go .so when will it be true.
and i found it now will download llama-b6106-bin-win-cpu-x64.zip not llama-b6106-bin-win-cuda-12.4-x64.zip,so when will it support cuda on windows?