Three different INTEL based HPC systems are used to benchmark an application of the LifeV library for running simulations of patient-specific cardiovascular hemodynamics. The targeted INTEL architectures rely on the Hashwell-Broadwell family of processors. Running times and scalability measures are collected with two real-size experiments. A third small-size test case is used to profile the code, exposing the effect of compiler vectorization, MPI efficiency and memory footprint. Profiling showed an unexpected low degree of floating point functional units usage, and a low percentage of effective vectorization. Extensive code redesign is likely necessary to best exploit the architectural features available in INTEL Knight Landing processors.
Benchmarking a Hemodynamics Application on Intel Based HPC Systems
Fedele M.;Veneziani A.
2018-01-01
Abstract
Three different INTEL based HPC systems are used to benchmark an application of the LifeV library for running simulations of patient-specific cardiovascular hemodynamics. The targeted INTEL architectures rely on the Hashwell-Broadwell family of processors. Running times and scalability measures are collected with two real-size experiments. A third small-size test case is used to profile the code, exposing the effect of compiler vectorization, MPI efficiency and memory footprint. Profiling showed an unexpected low degree of floating point functional units usage, and a low percentage of effective vectorization. Extensive code redesign is likely necessary to best exploit the architectural features available in INTEL Knight Landing processors.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
