How optimized is the container in regards to CPU performance?

I am interested in deep reinforcement learning using the GPU to train the model and the CPU to generate new examples for the network.

To that end I am wondering what degree of optimization I can expect for different CPUs or if I should assume no optimization and care for tweaking the image myself. My main goal is to avoid the overhead of looking into optimizations are already taken into account.

A more specific question that I have right now is: Which flavor of BLAS/LAPACK is installed?

We do no tuning for different CPUs, and we don’t explicitly install BLAS/LAPACK.

For more details on what is in the containers see http://docs.nvidia.com/deeplearning/dgx/support-matrix/index.html