Dynamic Space-Time Scheduling for GPU Inference

TL;DR

This paper introduces a dynamic space-time scheduling approach for GPU inference that significantly improves GPU utilization, throughput, and latency predictability by leveraging both temporal and spatial multiplexing techniques.

Contribution

It presents novel techniques for dynamic space-time multiplexing in GPU inference, demonstrating substantial improvements over traditional methods.

Findings

Up to 5x potential for GPU utilization improvement

3.23x increase in floating-point throughput with the prototype

7.73x increase over time-only multiplexing

Abstract

Serving deep neural networks in latency critical interactive settings often requires GPU acceleration. However, the small batch sizes typical in online inference results in poor GPU utilization, a potential performance gap which GPU resource sharing can address. In this paper, we explore several techniques to leverage both temporal and spatial multiplexing to improve GPU utilization for deep learning inference workloads. We evaluate the performance trade-offs of each approach with respect to resource-efficiency, latency predictability, and isolation when compared with conventional batched inference. Our experimental analysis suggests up to a 5x potential for improved utilization through the exploration of more advanced spatial and temporal multiplexing strategies. Our preliminary prototype of a dynamic space-time scheduler demonstrates a 3.23x floating-point throughput increase over space-only multiplexing and a 7.73x increase over time-only multiplexing for convolutions, while also providing better isolation and latency predictability.