Lessons Learned Developing Performance-Portable QMCPACK

Explore the redesign and reimplementation of QMCPACK, a code for predicting material properties, in this 58-minute webinar from the Exascale Computing Project. Learn about development practices, extensive testing strategies, and approaches for achieving portability and performance on GPUs and CPUs. Gain insights into algorithmic challenges, the ASLA approach, and real-world results that can benefit HPC application developers and facilities. Discover how QMCPACK tackles electron count, GPU parallelization, and scaling issues. Understand the main operations, Mini QMC implementation, and the delayed update technique. Examine version control practices, return on investment considerations, and performance challenges faced during the development process.

Introduction
Outline
What is QMC
Goals
QMCPACK Code
Electron Count
GPU
Parallel Scalability
Main Operations
Mini QMC
Algorithmic Challenges
ASLA Approach
New Approach
Real World Results
Delayed Update
Development Approach
Version Control
Return on Investment
Real World Problems
Performance
Challenges
Conclusion
Questions