Performance Tuning Process

With the growing significance of computing, parallel computing devices, such as graphics processing units (GPUs) and neural network processing units (NPUs) have become pivotal in the field of artificial intelligence and various other sectors. Computing efficiency, or referred to as computing performance, has attracted more and more attention.

This document introduces the concepts of performance, performance tools, and methods to tune performance of training models on Ascend devices. For details about the performance tuning process, see Figure 1.