2

We planning to shift the releases to docker; i.e. the software that we release will be based on docker. We also have an HPC cluster available.

I tried searching the internet but could not find a reference to make the docker build faster by utilising GPUs. If anyone is doing the same or aware of the procedure "How it can be achieved", kindly share the same.

Edit I am not talking about accessing gpu from inside the container, I want to use gpu while running docker build

EDIT I am not sure why the question is marked duplicate? Do we not understand the difference between docker build and docker run? And how do we manage to give power to mark duplicate to people who dont even understand the topic?

Sarvex
  • 763
  • 2
  • 7
  • 19
  • How do you currently build your software on GPU's? – Matt Feb 16 '17 at 09:29
  • Possible duplicate of [Using GPU from a docker container?](http://stackoverflow.com/questions/25185405/using-gpu-from-a-docker-container) – SztupY Feb 16 '17 at 11:05
  • 3
    @Sarvex is your CPU maxed out? I bet not, since creating images is IO intensive, not CPU intensive. If not, why do you think that using GPUs will help? You'll improve speed if you use a RAID array of SSDs. If you need to download images you may see that your *network* is the bottleneck. In this case download the images in advance – Panagiotis Kanavos Feb 16 '17 at 11:17
  • @PanagiotisKanavos It's not about the speed of build. One usecase is that I want to compile a CUDA function which is needed in pytorch when building the docker. – Zekun Zhang Jul 17 '19 at 10:13

0 Answers0