FAISS-GPU compatibility with Windows and CUDA

[rielmvp]

Hello.

I'm new to using FAISS and have been using the cpu version because of issues with using the gpu version of faiss, however the code that I am running takes too long (around 40 hours), and so I think I need to adapt to the faiss-gpu version and I wanted to inquire if it's possible to do so with my environment.

Here's my environment details:
Windows 10
faiss-gpu 1.9.0 installed through https://anaconda.org/conda-forge/faiss-gpu
CUDA 12.1

I read in the installation markdown in the repo that:
"faiss-gpu, containing both CPU and GPU indices, is available on Linux (x86-64 only) for CUDA 11.4 and 12.1".
However, I thought that since the package is available through conda-forge then it is possible to use faiss-gpu on Windows and that the reason I encountered an error is because of a mismatch with my CUDA version. Could anyone clarify if using faiss-gpu on Windows is possible by any chance, or if I have to pivot and run faiss-gpu with a Linux OS environment? Any other insights would be very helpful.

The error I encountered when I tried to access GPU resources using faiss-gpu is:
15:24:49.602 [error] Disposing session as kernel process died ExitCode: 3221226505, Reason: Faiss assertion 'err__ == cudaSuccess' failed in void faiss::gpu::runL2Norm<T, TVec>(faiss::gpu::Tensor<T, 2, true, long long, faiss::gpu::traits::DefaultPtrTraits> &, bool, faiss::gpu::Tensor<float, 1, true, long long, faiss::gpu::traits::DefaultPtrTraits> &, bool, CUstream_st *) at D:\bld\faiss-split_1734665785306\work\faiss\gpu\impl\L2Norm.cu:258; details: CUDA error 209 no kernel image is available for execution on the device