ihpcss-gpu-perf

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

ihpcss-gpu-perf

★ 9 Open GitHub ↗

Material for the International HPC Summer School course on GPU Performance Analysis

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# International HPC Summer School - GPU Performance Analysis

## Prerequisites

Ready-to-use profiling data files have been prepared ahead of time and are available in the `profiles` folder.
To visualize and analyze them *locally on your laptop*, a local installation of NVIDIA Nsight Compute and NVIDIA Nsight Systems is required.
Having an NVIDIA GPU is _not_ required.

You can either install the tools separately ([nsight compute](https://developer.nvidia.com/tools-overview/nsight-compute/get-started), [nsight systems](https://developer.nvidia.com/nsight-systems/get-started), might require a free NVIDIA developer account), or install the [CUDA toolkit](https://developer.nvidia.com/cuda-downloads) which bundles both tools.
In any case, please follow the instructions for your OS.

## Course Content

All course material is collected and available at [https://github.com/SebastianKuckuk/ihpcss-gpu-perf](https://github.com/SebastianKuckuk/ihpcss-gpu-perf) (this repository).

The course material is organized as a [Jupyter notebook](doc/ihpcss-2026-gpu-analysis.ipynb).
You can view it on github, in VS Code and other IDEs, or set up a JupyterHub instance to run on a compute node of your local cluster (support for that may vary).

## Instructions

Clone the repository to *your notebook* to visualize the profiles locally.
```bash
git clone https://github.com/SebastianKuckuk/ihpcss-gpu-perf
```

The next steps are only required in case you want to make your own measurements on a remote machine, e.g. on Bridges.

First, clone the repository on the target machine.
Then load required modules, e.g.
```bash
module load nvhpc
```
and start an interactive job, e.g.
```bash
interact -p GPU-shared --gres=gpu:v100-32:1 --time 2:00:00
```

You can copy files to your local machine via the usual means (scp, etc.).
On Bridges, you can also use the [on-demand web interface](https://ondemand.bridges2.psc.edu/).

ihpcss-gpu-perf

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

ihpcss-gpu-perf

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README