In these experiments, we vary the number of the executables (referred to as compute unit (CU)) on a varying number of cores of Titan. We meeasure strong and weak scalability as defined in Ref. Each executable is Gromacs, consistent with this use case.(https://docs.google.com/document/d/1a8i38Z_aROQgylRNtbsePGH6UovRJgg0WW4gbk5kW4A/edit#heading=h.8tk04bz0vj23)
We execute two experiments with:
- Strong Scaling: 512 tasks executed using between 128 and 512 cores
- Weak Scaling: Betwen 128 and 512 tasks executed on between 16 and 1024 cores, so as to keep ratio of number of tasks to cores constant (=1).
In both experiments we record the time to completion (TTC) of the execution as a measure of strong and weak scalability.
