jblache opened a new pull request, #454: URL: https://github.com/apache/mesos/pull/454
This PR contributes support for static MIG (multi-instance GPU) configurations in the Mesos containerizer, whereby nvidia GPUs can be divided into isolated instances to better use GPU resources. With these changes, Mesos will be able to pick up the MIG configuration and properly report available MIG GPUs instead of reporting the underlying, physical GPUs. I have made an attempt at playing nice with the `--resources` and `--nvidia_gpu_devices` agent options, however more scrutiny/work will definitely be beneficial in this area (we don't use these options in our setup). In addition to this PR, an update to the NVML library is required. These changes were built and tested with NVML taken from the 460.73.01 driver package. This code has been running in production on A100 GPUs for several months now and is being contributed back in the hope it will be useful and beneficial to the community. Much more can be done with MIG, and this only scratches the surface. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
