Tensorflow versions and their compatible CUDA versions

If you’ve used NimbleBox, you know you can upgrade or downgrade your environment to different CUDA versions (according to the Tensorflow version you’d like to use).

When you reset your environment from the dashboard, Tensorflow 2.2 is automatically installed (irrespective of the CUDA version selected). Please refer the below table to understand which Tensorflow version to install.

The following are the Tensorflow versions that are compatible with their respective CUDA versions:

GPU:

Tensorflow Version CUDA Version Python Version
tensorflow-2.4.0 CUDA11 3.6-3.8
tensorflow-2.3.0 CUDA10 3.5-3.8
tensorflow-2.2.0 CUDA10 3.5-3.8
tensorflow-2.1.0 CUDA10 2.7,3.5-3.7
tensorflow-2.0.0 CUDA10 2.7,3.3-3.7
tensorflow_gpu-1.15.0 CUDA10 2.7,3.3-3.7
tensorflow_gpu-1.14.0 CUDA10 2.7,3.3-3.7
tensorflow_gpu-1.13.1 CUDA10 2.7,3.3-3.7
tensorflow_gpu-1.12.0 CUDA9 2.7,3.3-3.6
tensorflow_gpu-1.11.0 CUDA9 2.7,3.3-3.6
tensorflow_gpu-1.10.0 CUDA9 2.7,3.3-3.6
tensorflow_gpu-1.9.0 CUDA9 2.7,3.3-3.6
tensorflow_gpu-1.8.0 CUDA9 2.7,3.3-3.6
tensorflow_gpu-1.7.0 CUDA9 2.7,3.3-3.6
tensorflow_gpu-1.6.0 CUDA9 2.7,3.3-3.6
tensorflow_gpu-1.5.0 CUDA9 2.7,3.3-3.6

NimbleBox instances do not support CUDA8.

You can change your CUDA version in the Project Settings by clicking the gear icon on the top-right corner of the project card on your NimbleBox Dashboard.