Advertisement

Researchers from CCFE, EPCC and the Universities of York and Lund have made substantial recent optimizations for the well-known plasma turbulence code, GS2. This included a total rewrite of the routines that calculate the response matrices required by the code's implicit algorithm, which has significantly accelerated GS2’s initialization, typically by a factor over 10. Taken together these optimizations have vastly reduced wall-time, as illustrated by the impressive factor 20 gain in speed that was achieved for a benchmark calculation running on 8,192 cores. The optimized code achieves scaling efficiencies close to 50% at 4,096 cores and 30% at 8,192 cores for a typical calculation, compared to efficiencies of 4% and 2% respectively prior to these optimizations.

Advertisement
Advertisement