Performance comparison between Peregrine and Eagle
November 09, 2018
NREL’s new Eagle supercomputer has a peak performance of 8 PetaFLOPS, or about 3.5 times that of Peregrine. As has been noted, peak performance doesn’t always translate to actual gains that applications can attain. To get a better idea of the real-world gains NREL’s HPC community is likely to see, the Eagle benchmark team designed a test suite that covered varying job sizes, types, and resource demands. We’re happy to report actual productivity improvements in excess of 200%, with the Eagle cluster able to run 2.7x as many high-throughput (1-node) VASP jobs, 2.3x as many mid-size (~20-node) Nalu fluid dynamics runs, and 3.6x as many large-scale (~200-node) geometric multigrid problems as could be achieved on Peregrine.