Abstract

Heterogeneous systems provide new opportunities to increase the performance of parallel applications and data intensive on clusters with CPU and GPU architectures. Currently, applications that utilize GPU devices run their device-executable code on local devices in their respective hosting-nodes. This talk describes a package for running OpenMP, C++ and unmodified OpenCL HPC applications on clusters with many GPU devices. This Many GPUs Package (MGP) includes an implementation of the OpenCL specifications and extensions of the OpenMP API that allow applications on one hosting-node to transparently utilize cluster-wide (CPUs and/or GPUs) devices. MGP provides means for reducing the complexity of programming and running parallel applications on clusters, including scheduling based on task dependencies and buffer management. The talk presents MGP and the performance of its internals.

This is joint work with Tal Ben-Nun, Ely Levy and Amnon Shiloh.