Tensorflow2.15.0 version compatibility issue

TensorFlow General Discussion
1

Hi there:
I am trying to deploy tensorflow on my hpc, I can get the version of nvidia drive which is < NVIDIA-SMI 525.105.17 Driver Version: 525.105.17 CUDA Version: 12.0 > . And sametimes I’ve tried tensorflow 2.9.0 / 2.12/ 2.13.1 / 2.14.0 vailable = tf.test.is_gpu_available() is_cuda_gpu_available = tf.test.is_gpu_available(cuda_only=True) is_cuda_gpu_min_3 = tf.test.is_gpu_available(True, (3,0)) All were “False”.

When I tried pip install tensorflow[and-cuda]==2.15.0, The answer turned to the “True”, but I still keep recieve the error E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered

I don’t know if the program crashes because of this, but my program does crash the system when calling tensorflow.

If my assumptions are correct I have multiple versions of cuda on my computer right now, which path should I export the correct path to cudnn, etc. from, although I’m not sure it’s necessary.

Any answer would be helpful, cheers!

3

This is caused by the same global objects being linked into both tensorflow_cc, and tensorflow_framework. I reported this against 2.14, but apparently they released 2.15 anyway with the same problem. We need a bazel expert to fix this.

See my analysis in this thread: cuDNN, cuFFT, and cuBLAS Errors · Issue #62075 · tensorflow/tensorflow · GitHub

1 Like
5

the tensorflow versions you indicate don’t support cuda 12.0 . Try >=11.8 < 12.0. Also Tf Versions > 2.10 no longer work on windows [ FOR GPU] . Need to use WSL.