Could not load library libcudnn_ops_infer.so.8 about whisper-standalone-win

[Purfview]

Lol you know what I do have NVIDIA GTX 1050 on windows host machine but I wasn't aware :)

2GB VRAM? Should be ~enough for medium model.

Could not load library libcudnn_ops_infer.so.8

GPU execution requires cuBLAS and cuDNN libs for CUDA v11.x:
https://developer.nvidia.com/cuda-toolkit
https://developer.nvidia.com/cudnn

Or alternatively you can download libs from there: https://github.com/Purfview/whisper-standalone-win/releases/tag/libs

[Purfview]

I've no idea how it reports that you have CUDA when you don't.

What ./whisper-faster --checkcuda shows?

You can force CPU with --device=cpu.

[arekgotfryd]

Thanks for the tip with --device=cpu!
That solves it.

[Purfview]

What ./whisper-faster --checkcuda shows?

[arekgotfryd]

CUDA device: 1

[Purfview]

And what GPU you have in your system?

CUDA device: 1

Yeah, reports that there is one CUDA device.

[arekgotfryd]

Lol you know what I do have NVIDIA GTX 1050 on windows host machine but I wasn't aware :)
Sorry for that!

[NaiveInvestigator]

Lol you know what I do have NVIDIA GTX 1050 on windows host machine but I wasn't aware :)

2GB VRAM? Should be ~enough for medium model.

Could not load library libcudnn_ops_infer.so.8

GPU execution requires cuBLAS and cuDNN libs for CUDA v11.x: developer.nvidia.com/cuda-toolkit developer.nvidia.com/cudnn

Or alternatively you can download libs from there: libs (release)

TY this helped a lot! Could we mention this in the README for newcomers?

[Purfview]

TY this helped a lot! Could we mention this in the README for newcomers?

This is literally written at the Releases.

[NaiveInvestigator]

It is but for most ppl when they see the right side and tend to click on whisper-faster and not on the Releases link which leads them to the executables.

[Purfview]

You see same info whatever you click there.

[Largo]

I'm on fedora 38. My research found that seemingly cuda and cuDNN are separate installs? I don't have the lib in /usr/local/cuda-12.4/lib64/.

So I did the following:

sudo find / -type f -name libcudnn_ops_infer.so.8

found it in /opt/resolve/libs/libcudnn_ops_infer.so.8

export LD_LIBRARY_PATH="/opt/resolve/libs:$LD_LIBRARY_PATH"

And then it works!