`torch.cuda.is_available()` returns `False`, but `torch.version.cuda` returns `10.2

I am trying to install Pytorch 1.6.0 via Pip package (pip install torch torchvision) on a computing grid in which I don’t have permission to install any software by myself. The torch.cuda.is_available() returns False, but torch.version.cuda returns 10.2.
Here is the output of nvcc --version :

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Wed_Oct_23_19:24:38_PDT_2019
Cuda compilation tools, release 10.2, V10.2.89

, and this is the output of nvidia-smi:

| NVIDIA-SMI 410.78       Driver Version: 410.78       CUDA Version: 10.0     |
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|   0  Tesla P100-PCIE...  Off  | 00000000:3B:00.0 Off |                    0 |
| N/A   23C    P0    26W / 250W |      0MiB / 16280MiB |      0%      Default |
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|  No running processes found                                                 |

How can I fix the problem?

I should add that other versions of CUDA are available on the computing grid, including CUDA 10.0 and CUDA 9.0, but unfortunately according to PyTorch installation none of these CUDA versions are appropriate to install Pytorch via Pip (currently PyTorch with CUDA 9.2, 10.1, and 10.2 can be installed via Pip). Is there any way to solve this problem?

torch.version.cuda indicates which CUDA runtime was shipped in the binaries (or used to build PyTorch from source). It does not tell you, if CUDA can be used as seen by the output of torch.cuda.is_available().

For CUDA10.2.89 you would need an NVIDIA driver >= 440.33 as seen here.

You could update the driver or, if that’s not possible, use the binaries with CUDA9.2, which should work using your driver.

Thank you for the answer.