Data Movement Is All You Need: A Case Study on Optimizing Transformers

TL;DR

This paper identifies data movement as the main bottleneck in training transformers and proposes a global optimization approach that reduces data movement and improves training performance significantly.

Contribution

It introduces a novel recipe for optimizing data movement in transformer training, leading to substantial performance gains over existing frameworks.

Findings

Data movement reduction up to 22.91%

Performance improvement of 1.30x on BERT encoder layer

Performance improvement of 1.19x on full BERT

Abstract

Transformers are one of the most important machine learning workloads today. Training one is a very compute-intensive task, often taking days or weeks, and significant attention has been given to optimizing transformers. Despite this, existing implementations do not efficiently utilize GPUs. We find that data movement is the key bottleneck when training. Due to Amdahl's Law and massive improvements in compute performance, training has now become memory-bound. Further, existing frameworks use suboptimal data layouts. Using these insights, we present a recipe for globally optimizing data movement in transformers. We reduce data movement by up to 22.91% and overall achieve a 1.30x performance improvement over state-of-the-art frameworks when training a BERT encoder layer and 1.19x for the entire BERT. Our approach is applicable more broadly to optimizing deep neural networks, and offers insight into how to tackle emerging performance bottlenecks.