Evaluation of PETSc on a Heterogeneous Architecture, the OLCF Summit System: Part 1: Vector Node Performance

Abstract

60439. For information about Argonne and its pioneering science and technology programs, see www.anl.gov.

Toward performance-portable PETSc for GPU-based exascale systems

“…This information guides the decision of subwork units for Response 3. The data presented here is drawn from the more complete analysis in [25],…”

Evaluation of PETSc on a Heterogeneous Architecture, the OLCF Summit System: Part II - Basic Communication Performance

“…The NIC that connects the node to the parallel network is connected to a programmable "local network" that connects it to the CPU memory as well as the GPU memory. This combination of interconnects means that the parallel communication latency and bandwidth (see the first report [8]) are limited by the NIC, the local network, the NVLinks from the CPU to the local network, and the GPU memory but not the CPU memory. However, CUDA-aware MPI calls (send, receive, and waits) must be called by code running on the CPU cores.…”

“…We report on the performance of the Portable, Extensible Toolkit for Scientific Computation (PETSc) [2,3] communication infrastructure using basic Ping-Pong point-to-point communication and regular and irregular nearest-neighbor communication on the IBM/NVIDIA Summit computing system [11] at the Oak Ridge Leadership Computing Facility (OLCF). This report is a continuation of Evaluation of PETSc on a Heterogeneous Architecture, the OLCF Summit System: Part I -Vector Node Performance [8] that introduces the Summit architecture and analyzes simple on-node performance characteristics. This report builds on the previous report's analysis and will not repeat the detailed material.…”