Computing for Change
SC Theater
SCinet
Students@SC
Technology Challenge
:
Enabling High-Level Parallel Abstractions to DCA++ Using HPX and GPUDirect
SessionSC Theater
Presenter
Event Type
Computing for Change
SC Theater
SCinet
Students@SC
Technology Challenge
Registration Categories
TP
EX
EXH
TimeThursday, 21 November 201912:30pm - 1pm
LocationBooth 981
DescriptionThis project focuses on enabling high-level parallel abstraction to DCA++ using HPX and GPUDirect. DCA++ is a high-performance research software framework to solve quantum many-body problems with cutting edge quantum cluster algorithms. We added light-weight user threading provided by HPX (High Performance ParallelX) to DCA++ extending the current application to asynchronous task-based execution model. We designed a benchmark on comparing GPUDirect communication technique, a high bandwidth low latency switched fabric interconnect method, and traditional communication technique and obtained 17x speedup using GPUdirect on the benchmark application on Summit supercomputer.
Back To Top Button