Supervisor: Benjamín Hernández (Oak Ridge National Laboratory)
Abstract: We present performance analysis on OpenPOWER architecture of an algorithm to generate transversal views of atomistic models. The algorithm was implemented with data parallel primitives in NVIDIA Thrust for architecture portability. We report performance results on IBM Power9 CPUs (OpenMP, Intel Threading Blocks) and NVIDIA Volta GPUs (single and multi GPU). We also evaluate CUDA unified memory performance, exposed by NVIDIA RAPIDS Memory Manager library (RMM).
ACM-SRC Semi-Finalist: yes
Poster Summary: PDF
Back to Poster Archive Listing