We have been working on Deep learning based software development in python. We have in our possession one GPU based deep learning machine: PARAM SHAVAK.
This machine is having two NVIDIA P5000 GPU cards. So far we have been able to successfully use only one GPU card for processing. But we are UNABLE to used both GPU cards at same time through our code.
To more specific, we are using Python language with a popular Deep learning based Computer Vision library: DLIB(http://dlib.net/)
We have been using the library to extract any human face features from video frames. Before extracting the face needs to be detected in (video) frame. So far the application uses one 1 GPU at a time. This leaves other(2nd) GPU unused hence whole system is left underutilized.
The DLIB library which we are using is using underlying CUDA implementation for GPU utilization.
To summarize, can anyone help with some mechanism though which a python code can automatically distribute the load to the second GPU if required. Can CUDA provide any such feature for load balancing.