Towards a Single-Host Many-GPU System
Authors:
- 1 September 2018
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 140-147
Abstract
As computation-intensive tasks such as deep learning and big data analysis take advantage of GPU based accelerators, the interconnection links may become a bottleneck. In this paper, we investigate the upcoming performance bottleneck of multi-accelerator systems, as the number of accelerators equipped with single host grows. We instrumented the host PCIe fabric to measure the data transfer and compared it with the measurements from the software tool. It shows how the data transfer (P2P) helps to avoid the bottleneck on the interconnection links, but multi-GPU performance does not scale up as expected due to the control messages. We quantify the impact of host control messages with suggestions to remedy scalability bottlenecks. We also implement the proposed strategy on Lulesh to validate the concept. The result shows our strategy can save 59.86% time cost of the kernel and 13.32% PCIe H2D payload.Keywords
This publication has 11 references indexed in Scilit:
- A PCIe Congestion-Aware Performance Model for Densely Populated Accelerator Servers2017
- TrueNorth Ecosystem for Brain-Inspired Computing: Scalable Systems, Software, and Applications2016
- ImageNet Large Scale Visual Recognition ChallengeInternational Journal of Computer Vision, 2015
- Caffe2014
- Effective sampling-driven performance tools for GPU-accelerated supercomputers2013
- Kernel Weaver: Automatically Fusing Database Primitives for Efficient GPU Computation2012
- Design of Hardware-Based Communication Performance Measurement Tool2012
- Parallel Performance Measurement of Heterogeneous Parallel Systems with GPUs2011
- Quantifying NUMA and contention effects in multi-GPU systems2011
- The Scalable Heterogeneous Computing (SHOC) benchmark suite2010