1.) Can we train multiple models in the same GPU (Eg: tesla V100). what are the ways to do it?
2.) How to set up an AI lab with a server which has 2 Tesla V100 16 or 32 GB how many students can use it simultaneously to train different models with DIGITS Interface.
3.) How many people can use DGX station and DGX server 1 for training different multiple models at simultaneously in a LAB setup?
4.) How can use GPU virtualization for Deep learning training?
You can train multiple models in the same GPU at the same time as long as the GPU memory is still available. However, the training speed will be slow.
DIGITS can discover both GPUs and display them as available GPUs for training. Users can assign which GPU (or GPUs) to use for training. No special configuration is needed.
It depends on the complexity of your model and runtime requirements on GPUs.
We recently received a bug report on DIGITS w/ tensorflow backend in vGPU environment. Bascially, you can run DIGITS w/ Caffe backend in vGPU without problem, but it requires a small patch to have DIGITS w/ tensorflow work in vGPU.
We suggest you take a look at NVIDIA’s Deep Learning Institute and see if it meets your special requirements on AI lab setup.
You can set yourself For Deep Learning for this you can visit Nearest training Institute for AI Lab Setup.