Teams must perform a set of calculations while remaining inside a power limit. Applications this year included POP, Mr. Bayes, Pretty Fast Analysis (PFA), and Gadget.
I very much enjoy seeing what the teams come up with. This year one of the problems includes a code that benefits greatly from GPU acceleration but is not required. As teams are limited by the amount of power it was great to see that some teams decided that GPU was not the way to go. For a balanced system to get the greatest amount of work done a few decided to forgo GPU hardware all together.
I was very impressed with how deep down some teams had gone, and to what lengths they went to get the most out of their power budget.
- Genetic algorithms for optimizing parameters
- Oil immersed cooling
- Over/Under Clocking
- Scheduling of node power states (shutdown GPU nodes to turn on CPU nodes)