Find Inefficiencies and Rapid Model Profiling with CentML DeepView
Outline
In this segment, we introduce the typical deep-learning (DL) workload from a systems perspective, discuss existing tooling that addresses the problem of performance analysis through profiling, and finally provide a hands-on experience of CentML DeepView, an improved interactive visual profiler that helps you identify bottlenecks in your DL workloads.
Preparation
To get the most out of the tutorial, please preferably have the following ready when you attend this tutorial:
- Bring a laptop computer with Visual Studio Code. Install the Remote-SSH plugin, which will be used to launch profiling on a remote workstation.
- Have a remote workstation running Linux with a NVIDIA GPU that you can ssh into. You also need to have Python and CUDA installed.
- (optional) If you wish to install DeepView ahead of time, please consult our documentation. We will also allocate time during the workshop for you to install DeepView.
Slides
Coming soon.
Links
To learn more about DeepView and installing it on your workstation, consult the CentML documentation.