A key point to stress from Alessandra's talk was that:
"ATLAS (and CMS) under a significant pressure to make use of GPU because large DoE HPC use GPUs."
Actions:
Alessandra Forti and Andy McNoab to come up with a naming convention for GPU tags. Daniel showed some slides that indicated the performance of different jobs on different GPUS was extremely variable. Enabling users to specify their preferred GPU would be useful.
Alastair to arrange another meeting on GPU is around 3 - 4 months. Lots of things discussed were very interesting but for the moment it was left to interested sites to pursue deployment of GPUs by themselves. We can re-evaluate this at the next meeting.