Ideally, we should execute all/most of the tests with GPU enabled. However, given that GPU-specific code changes are limited, I agree that we can start slowly.
Regards, Arnab.. On Mon, Nov 8, 2021 at 10:57 PM Baunsgaard, Sebastian <baunsga...@tugraz.at.invalid> wrote: > Grat topic. > > It would be nice if we could use some small GPU cloud instances on demand > then maybe we can find some budget if it is not to large a cost. > > Best regards > Sebastian > ________________________________ > From: Janardhan <janard...@apache.org> > Sent: Monday, November 8, 2021 7:30:00 PM > To: dev@systemds.apache.org > Subject: [DISCUSS][SYSTEMDS-3210] Self hosted runner for the GPU > integration test suite > > Hi all, > > TL;DR Integration test for GPU with self hosted runners > > Our integration tests span most of the system including basic checks for > docs. > > One component where we would be needing extensive testing would be GPU, > for this although GPU specific machines are not available on the GitHub > workflow > we with the help self hosted runners [1] could run the testing the > same way current testing is done. > > As a starting point, I have implemented this feature on my fork [2] as > an experimental feature. Planning to also be in touch with INFRA. I am > planning to run this action every saturday and PRs touching the gpu > component > as per the suggestion made by Mark (thank you!). > > Hoping that it helps keep confidence in that part of the code and also > remove the cognitive burden on the gpu devs. :) > > -- > [1] > https://docs.github.com/en/actions/hosting-your-own-runners/about-self-hosted-runners > [2] https://github.com/j143/systemds/commits/main-gpu > > Thank you, > Janardhan >
