GPU Activity Prediction using Representation Learning

TL;DR

This paper introduces a representation learning method for predicting GPU activity by modeling performance metrics as temporal functions of executed instructions, demonstrating high accuracy on a benchmark.

Contribution

It presents a novel approach using representation learning to predict GPU activity, moving beyond traditional heuristics.

Findings

High prediction accuracy achieved

Effective modeling of instruction flow as activities

Strong predictive power demonstrated on benchmark

Abstract

GPU activity prediction is an important and complex problem. This is due to the high level of contention among thousands of parallel threads. This problem was mostly addressed using heuristics. We propose a representation learning approach to address this problem. We model any performance metric as a temporal function of the executed instructions with the intuition that the flow of instructions can be identified as distinct activities of the code. Our experiments show high accuracy and non-trivial predictive power of representation learning on a benchmark.