The experimental beta access phase on Leonardo, called LEAP (Leonardo Early Access Program) launched in early April, has ended.
Of the 80 proposals submitted, totaling 85 million hours of computation required on GPU nodes, 13 projects were able to take advantage of Leonardo during the early access.
The selected projects belong to various research areas, from Numerical Analysis and Fluid Dynamics to Theoretical Physics projects and AI and Machine Learning development.
Leap winners were let “in” on the machine in increments of three projects at a time, one group per week. The projects consumed a total of 1 million node-hours, compared to therequired 4.5 million, still a very good result considering that Leonardo “produces” at its best about 30 million node-hours per year.
Approximately 40,000 jobs have been submitted, with average sizes from 128 nodesupwards, with 256, 512 and 1024 node-hours staggers, but also went so far as to experiment with 2048- and 3000-nodes jobs (Leonardo’s Booster module counts 3456 nodes in total), definitely approaching exascale performance.
More information on the open production calls can be found on our resources page.