D4.5 Final report on code profiling and bottleneck identification

This final report presents a general summary of the profiling and benchmarking activities in WP4. Thanks to this task we have at first identified bottlenecks and inefficiencies in the flagship codes; and we continue to provide the developers a check on their performance progress. Our screening is based on a defined set of scientific test cases on which we benchmark the codes. This collection contains the main types of calculation that can be executed by MaX codes in supercomputing centres. The tested systems have sizes that span several orders of magnitude. We can thus assess the efficiency of codes in exploiting as much computational power as possible, either in the case of very large systems (many thousands of atoms); or performing High Throughput Computations (HTC) on small and medium-size systems, to quickly explore wide configuration spaces. The bottlenecks identified by the benchmarks were in many cases unexpected or overlooked. They were revealed by our tests thanks to their very large size, or because -with the intent to mimic the resources accessible in exascale machines- we executed them with an (at the moment) unusually large amount of computing resources.