CUDA error: CUBLAS_STATUS_ALLOC_FAILED when calling cublasCreate(handle)

[cholzer79]

RTX4090
Windows 11

The readme says to install CUDA 11.0 so I installed cuda_11.0.1_451.22_win10.exe

CPU method works, but the GPU method throws this error:
CUDA error: CUBLAS_STATUS_ALLOC_FAILED when calling cublasCreate(handle)``

In case anyone runs into this, install cuda_11.0.3_451.82_win10.exe - that works.
Any later versions like cuda_11.8.0_522.06_windows.exe on the other hand do not work.

So that is a bit annoying.

[jtxnn678]

@cholzer79 Thank you man, had exact problem with CUDA 11.0.1 too. I tried your suggestion installed CUDA 11.0.3 and the error is gone.

However, I noticed that while the error is gone with CUDA 11.0.3 but it runs a lot slower than CUDA 11.0.1 on same video file.

RTX3070
Wndows 11