Skip to content

CI jobs updated

Daniel Campora Perez requested to merge allen_tdr_no_ut_cut into allen_tdr

This MR reworks the CI to use resources in a more balanced way. It also enables CUDA Debug build again.

  • The GPU location is fetched from the runner description. Every runner description contains the GPU slot in its name. Ie. n4060101:0 is a runner for the GPU 0 in node n4060101. n4060101:1 is a runner for GPU 1, and so on.
  • The build of Allen is done in the shared nodes with label cvmfs. CVMFS contains the required CUDA 10.2 and gcc 8.3.0 releases.
  • The build of Allen is done with ninja now, which is faster and uses less memory.
  • Only node n4062201:0 has the label x862630v4, which means the CPU throughput and CPU physics efficiency runs can only run on this node. This node does not use its GPUs for the build system. This is done to avoid interferences between the GPU and CPU throughput runs. In the future, a node without GPUs can be used.

Closes #121 (closed)

Edited by Daniel Campora Perez

Merge request reports