Completed
Distributing computation
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
Facilitating Electronic Structure Computations on GPU-based Exascale Platforms
Automatically move to the next video in the Classroom when playback concludes
- 1 HPC Best Practices Webinar Series
- 2 Algorithms and performance portability for electroni structure
- 3 Speeding up electronic structure calculations to ena
- 4 Running MD on exascale platforms
- 5 Main numerical kernels for electronic structure calculations
- 6 Developing alternative solvers based on polynomial matrices
- 7 Implementation divided into two libraries
- 8 Using OpenMP for GPU offloading
- 9 General implementation strategy
- 10 Computer Science challenges
- 11 BML: supported (shared memory) matrix formats
- 12 BML: Supporting multiple data types in a C code
- 13 BML: Fortran interface is important for targeted application codes
- 14 BML: Unit test/Continuous integration
- 15 Offloading to GPU
- 16 Offloading strategy
- 17 GPU offloading with OpenMP
- 18 Challenges in interfacing with optimized vendor libra
- 19 Using a synthetic Hamiltonian matrix for Performanc Benchmarking
- 20 rocSPARSE performance on Crusher @ OLCF
- 21 Chebyshev expansions for modest matrix sizes (metals)
- 22 Chebyshev expansion of Density Matrix
- 23 Exploiting GPU concurrency in calculating Chebysh terms
- 24 Distributing computation
- 25 Balancing computational cost and accuracy with matrix thresholding
- 26 A non-intrusive implementation
- 27 What about wavefunction-based solver? (Planewaves...)
- 28 Numerical Discretization of DFT problem
- 29 Parallel scaling/performance on Summit
- 30 Lesson learned: Efficiently using GPUs requires a lo work!
- 31 Acknowledgments
