CUDA version for DeepSpeech 0.6.1

I would like to update to DeepSpeech 0.6.1, but I am unable to find the information what is the correct CUDA version for currently used Tensorflow 1.15.0. The Google website does not mention this configuration: https://www.tensorflow.org/install/source#linux

Can anyone confirm if it is CUDA 10.0 with CUDNN 7.4 like TF 2.0 or previously used 1.14.1 or rather CUDA 10.1 with CUDNN 7.6 like TF 2.1?

1.15 is the version used on the current master. The correct version for 0.6.1 is 1.14.

So what is the correct CUDA version for 1.15.0?

TF 1.15 with CUDA 10.1 work fine and fast for me

Default TensorFlow Docker CI is still 10.0: https://github.com/tensorflow/tensorflow/blob/r1.15/tensorflow/tools/ci_build/Dockerfile.gpu#L1

I don’t get why you dismiss @dabinat’s remark, he’s right, 0.6.1 is r1.14.

1 Like

Right, thanks guys. My apologies @dabinat, that’s not my best day…

1 Like