Would Nvidia DIGITS Container image automatically detect the number of GPUs on the target machine and use them? For example, for one server I may have 2 GPUs and for another server with 4 GPUs, but the Nvidia DIGITS container image is the same. For a server with 4 GPUs, and I would like to use only 2 GPUs, how could I control the number of GPUs to be used? Thanks!
An nvidia docker container is normally able to use all the GPUs in the machine. You can restrict this using NV_GPU environment variable, when you start the container. Read about GPU Isolation here:
[url]https://github.com/NVIDIA/nvidia-docker/wiki/nvidia-docker[/url]
Thanks!
One more question:
If I use DIGITS on DGX-1 with 8 V100, and select using all of these 8 GPUs, but my model is not large to have to do model parallelism on more than 1 GPU, will CUDA/DIGITS still automatically do model parallelism on all 8 GPUs, or it will automatically do data parallelism on 8 GPUs? Thanks!
We created a new “Deep Learning Training and Inference” section in Devtalk to improve the experience for deep learning and accelerated computing, and HPC users:
https://devtalk.nvidia.com/default/board/301/deep-learning-training-and-inference-/
We are moving active deep learning threads to the new section.
URLs for topics will not change with the re-categorization. So your bookmarks and links will continue to work as earlier.
-Siddharth