NGT Cluster Usage
We have very nice GPUs available in the NGT cluster. Would it be possible for the experiments to test and benchmark their code on NGT's hardware? Of course only while the cluster is not already saturated by NGT's own workload.
For ALICE, the following would be very helpful:
- Access to the cluster for only a few non-NGT people.
- Ability to run an O2 benchmark on different GPUs.
- Daily performance monitoring: Test/Benchmark a few selected datasets on different hardware.
- CI-pipeline: A short test on one NVIDIA and one AMD GPU per pull request.
For the CI-pipeline, we would not build, but only run the code on a GPU-enabled runner.
This would help us in particular with the following:
- Decide what hardware ALICE should purchase/rent in the future.
- Facilitate to test if our code runs on different hardware.