GPU support

[fkorotkov]

Description

It will be nice to support option to have GPUs attached to containers.

container:
  image: nvidia/cuda:10.0-runtime-ubuntu18.04
  cpu: 4
  memory: 12G
  gpu: 2
  gpu_type: nvidia-tesla-v100 # do we need the choice?

Context

See Twitter conversation here: https://twitter.com/acberceanu/status/1151884558725459968

Anything Else

Free limits (if any?) and compute credits pricing are TBD since GPUs are EXPENSIVE 😅

Should be there an option to choose between different GPU like nvidia-tesla-k80, nvidia-tesla-p100, nvidia-tesla-p4, nvidia-tesla-v100, or nvidia-tesla-t4? Does it make sense for such workloads? 🤔

[fkorotkov]

Prices for K80 and P4 are the most reasonable that can allow free tier for OSS. 🤔

[fkorotkov]

On the bright side adding support to Linux Community Cluster and GKE integration is pretty simple.

[ax3l]

Should be there an option to choose between different GPU [...]

yes, would certainly be very useful to test on various architectures :+1:

[jedbrown]

Note that GPU's do not need to be exclusive, particularly on a free tier. Many CI jobs will have build overhead at a minimum, during which time the GPU could be busy with other things.

[berceanu]

@fkorotkov what's the status on this?

[fkorotkov]

There is no bandwidth right now because of coming-soon issues and some other requests from core users. Don't have an exact ETA for GPUs unfortunately. Most likely closer to the end of Q3.