The compatibility between TensorFlow and CUDA versions depends on the specific version of TensorFlow you are using.
Here is a general compatibility table for TensorFlow and CUDA:
| TensorFlow version | CUDA version | cuDNN version |
|---|---|---|
| 2.7.x | 11.4 | 8.2 |
| 2.6.x | 11.2 – 11.4 | 8.1 |
| 2.5.x | 11.0 – 11.2 | 8.0 |
| 2.4.x | 11.0 | 8.0 |
| 2.3.x | 10.1 – 11.0 | 7.6 |
| 2.2.x | 10.1 – 11.0 | 7.6 |
| 2.1.x | 10.0 | 7.4 |
| 2.0.x | 10.0 | 7.4 |
Note that these are general guidelines, and it’s always best to check the tensorflow documentation for the specific version you’re using to ensure compatibility with your CUDA and cuDNN versions.
You should also ensure your GPU driver version is compatible with your chosen CUDA version.
That’s it.
