MARS: Exploiting Multi-Level Parallelism for DNN Workloads on Adaptive Multi-Accelerator Systems

TL;DR

MARS is a framework that enhances DNN performance on multi-accelerator systems by optimizing accelerator selection and communication-aware sharding, significantly reducing latency.

Contribution

It introduces a novel mapping framework combining computation-aware accelerator selection with communication-aware sharding strategies for DNN workloads.

Findings

Achieves 32.2% average latency reduction on typical DNNs.

Reduces latency by 59.4% on heterogeneous models.

Outperforms existing methods in multi-accelerator DNN mapping.

Abstract

Along with the fast evolution of deep neural networks, the hardware system is also developing rapidly. As a promising solution achieving high scalability and low manufacturing cost, multi-accelerator systems widely exist in data centers, cloud platforms, and SoCs. Thus, a challenging problem arises in multi-accelerator systems: selecting a proper combination of accelerators from available designs and searching for efficient DNN mapping strategies. To this end, we propose MARS, a novel mapping framework that can perform computation-aware accelerator selection, and apply communication-aware sharding strategies to maximize parallelism. Experimental results show that MARS can achieve 32.2% latency reduction on average for typical DNN workloads compared to the baseline, and 59.4% latency reduction on heterogeneous models compared to the corresponding state-of-the-art method.