Why does torch.cuda.is_available() return false even though I have it installed?

I have an NVidia GeForce GTX 1650 Ti Graphics card. I have already checked the compatibility of my graphics card with CUDA 12.1.

I ran the nvidia-smi command. Here are the results :

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 531.14                 Driver Version: 531.14       CUDA Version: 12.1     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                      TCC/WDDM | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf            Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce GTX 1650 Ti    WDDM | 00000000:01:00.0 Off |                  N/A |
| N/A   37C    P8                4W /  N/A|      0MiB /  4096MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+

I just installed Pytorch using :

conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch-nightly -c nvidia

But, for some reason, torch.cuda.is_available() returns false.

Here are some more details :

• Results of nvcc --version :

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Mon_Apr__3_17:36:15_Pacific_Daylight_Time_2023
Cuda compilation tools, release 12.1, V12.1.105
Build cuda_12.1.r12.1/compiler.32688072_0

• When I run python -m torch.utils.collect_env, I get:

PyTorch version: 2.1.0.dev20230608
Is debug build: False
CUDA used to build PyTorch: Could not collect
ROCM used to build PyTorch: N/A

[...]

Is CUDA available: False
CUDA runtime version: 12.1.105
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: GPU 0: NVIDIA GeForce GTX 1650 Ti
Nvidia driver version: 531.14

[...]

[conda] pytorch                   2.1.0.dev20230608    py3.10_cpu_0    pytorch-nightly
[conda] pytorch-cuda              12.1                 hde6ce7c_5    pytorch-nightly
[conda] pytorch-mutex             1.0                         cpu    pytorch-nightly
[conda] torchaudio                2.1.0.dev20230608       py310_cpu    pytorch-nightly
[conda] torchvision               0.16.0.dev20230608       py310_cpu    pytorch-nightly

Sometimes when working on projects with many dependencies one package might override the pytorch installation with a CPU only build which will wreck your environment.

My recommendation would be to isolate the problem by creating a totally new environment to see if the pytorch installation is the problem or if it's something deeper.

In a new directory from either PowerShell or cmd run:

python -m venv venv
.\venv\Scripts\activate  # Windows
# ./venv/bin/activate  # Unix

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

python -c "import torch; print(torch.cuda.is_available())"

You can double check that you are using the python from your venv by running:

which python  # Windows
# where python  # Unix

If the output still says false then you have confirmed that it is not pytorch's fault.

Next make sure your drivers are up to date: https://www.nvidia.com/download/index.aspx

And you have at least CUDA 11 installed: https://developer.nvidia.com/cuda-downloads

It might be worth rolling back your CUDA installation from 12 to 11 to see if that will work.

Finally, you can take solace in the fact that you are not the only one facing this problem recently. It may be the fault of recent NVIDIA software.