A meeting on detailed strategies to benchmark MaX flagship codes took place last October at CINECA.
Benchmarking is in fact an extremely important task for a twofold reason. On the one hand, it provides an immediate feedback to the developers on the impact of their work on the codes’ performance. This information is far from trivial since in general there is a substantial difference between the numerous platforms where the scientific codes are executed and the one where the development actually takes place.
On the other hand, the benchmarking activity also gives precious information on the impact of the changes of hardware, operating system, and library eco-system on the code performances. Moreover it may provide indications on the effectiveness of the various algorithmic procedures implemented in exploiting the increasing computational power provided by the technological evolution.
The one day-long discussion, which involved many developers from the Yambo, QE and AiiDA projects, discussed in depth the pivotal aspects concerning code performance that must be analyzed and monitored over time, and the instruments to be used for this benchmarking activity. A detailed benchmarking activity is scheduled for the next months.