Using 50k GPUs in one HTCondor cloud-based pool

December 6, 2019

Read about how at SC19 UCSD got over 50k GPUs from the three largest Cloud providers, and integrated them in a single a HTCondor pool to run production science jobs. At peak, this cloud-based dynamic HTCondor cluster provided more than 90% of the performance of the #1 TOP100 HPC supercomputer.