[UBUNTU 16.04] Tensorflow-gpu 1.11 with Compute capability 3.0 (with cuda 9 and cudnn 7.3)

The minimum required Cuda capability is 3.5: Tensorflow-gpu 1.11

Ignoring visible gpu device (device: 0, name: GeForce GTX 770, pci bus id: 0000:01:00.0, compute capability: 3.0) with Cuda compute capability 3.0. The minimum required Cuda capability is 3.5

The minimum required Cuda capability is 3.5(tensorflow gpu 1.11)

Solution:

The tensorflow require GPU card with Cuda compute capability of 3.0 or higher when building from source and 3.5 or higher when using pre-built binaries (3.4 Cuda capability when downloading tensorflow using pip).

Follow the instructions given on Tensorflow website. I have used Cuda 9 and cudnn 7.3 and installed using the details on the tensorflow website.

During configuration, I have mentioned python 3.5, Cuda 9, cudnn 7.3, I was getting an error regarding nvcc 2.2, so I used 1.3 as I won’t be using multiple GPUs, I had only one GPU. It asked for compute compatibility versions to compile tensorflow using those capabilities. I had mentioned only 3.0 compute capability.

After compiling, I have packaged it using the mentioned steps.

The final tensorflow package is:

Tensorflow-gpu 1.11 (cuda 9 and cudnn 7.3)

Download the tensorflow-gpu package and install it using:

$ pip install <path/filename.whl>

 

Related Links:

Leave a Reply

Your email address will not be published.