Matching TF version with cuDNN and CUDA

Hi TensorFlow Community,

I am trying to use GPUs with TensorFlow on a HPC cluster at my university. This means that the cuDNN and CUDA modules I can use are decided for me and it is only the TensorFlow version I can decide.

The available cuDNN and CUDA combos are:
cuDNN 8.4.1 with CUDA 11.7.0
cuDNN 8.8.0 with CUDA 12.0.0

I have tried looking at the tables here , but the versions I have are not listed :frowning:

Do any of you know what Tensorflow version that match one of these cuDNN and CUDA combinations?

I would appreciate any tips or hints or where to look for the answer myself because I need it to work to train a deep learning model for my master’s thesis:))

Hi @I_M, AFAIK You might be able to use tensorflow with mentioned combinations. But in some cases you might not get the expected result and performance.

For that reason we always recommend the community to build the tensorflow with compatible tested build configuration which are mentioned in the official document. Thank You.