Interconnect Bandwidth Heterogeneity on AMD MI250x and Infinity Fabric

TL;DR

This paper investigates how heterogeneity in interconnect bandwidth among AMD MI250x GPUs affects data transfer performance and visibility through programming APIs, providing insights for optimizing multi-GPU system usage.

Contribution

It characterizes interconnect heterogeneity on AMD MI250x GPUs and offers practical insights for developers working with such heterogeneous multi-GPU systems.

Findings

Interconnect bandwidth varies significantly among GPUs.

API visibility of heterogeneity impacts performance tuning.

Insights aid in optimizing multi-GPU workloads.

Abstract

Demand for low-latency and high-bandwidth data transfer between GPUs has driven the development of multi-GPU nodes. Physical constraints on the manufacture and integration of such systems has yielded heterogeneous intra-node interconnects, where not all devices are connected equally. The next generation of supercomputing platforms are expected to feature AMD CPUs and GPUs. This work characterizes the extent to which interconnect heterogeneity is visible through GPU programming APIs on a system with four AMD MI250x GPUs, and provides several insights for users of such systems.