Experiments and Data Presentation in High Performance Computing

Explore a 36-minute lecture on conducting experiments and presenting data in high performance computing (HPC). Learn from Georg Hager, a renowned expert in the field, as he covers essential topics including key questions to ask, measurement techniques, statistical variation, and best practices for data presentation. Gain insights into raw and derived metrics, general guidelines for experiments, and effective graph creation. Examine the pros and cons of curve fitting, and understand the differences between runtime and performance scaling. This lecture, part of a seminar on "Efficient Multi- and Manycore Programming" at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), references important work by Hoefler and Belli on scientific benchmarking in parallel computing systems. Enhance your skills in HPC experimentation and data presentation to improve your research and analysis capabilities.

Intro
Questions to ask
The most important question
Some useful general hints
What and how to measure
Raw and derived metrics
Statistical variation (II)
General guidelines
Graphs: the good, the bad, and the ugly
Curve fitting
Runtime or performance scaling?
Runtime or performance?