Approximate High-Performance Computing: A Fast and Energy-Efficient Computing Paradigm in the Post-Moore Era

Harshitha Gopalakrishnan Menon | 20-ERD-043

Executive Summary

We are developing algorithms and a computational framework to enable high-performance computing programs to use approximate computing techniques to achieve significant performance and energy savings. The results of our investigation could inform future use of the most feasible approximate computing techniques for software workloads at national laboratories and the development of hardware requirements to support those workloads.

Publications, Presentations, and Patents

Parasyris, Konstantinos, Ignacio Laguna, Harshitha Menon, Markus Schordan, Daniel Osei-Kuffuor, Giorgis Georgakoudis, Michael O. Lam, and Tristan Vanderbruggen. "HPC-MixPBench: An HPC Benchmark Suite for Mixed-Precision Analysis." In 2020 IEEE International Symposium on Workload Characterization (IISWC), pp. 25-36. IEEE, 2020.

Parasyris, Konstantinos, Giorgis Georgakoudis, Harshitha Menon, James Diffenderfer, Ignacio Laguna, Daniel Osei-Kuffuor, Markus Schordan. “Analyzing The Effect of Approximations on OpenMP HPC Applications.” In SC20: International Conference for High Performance Computing, Networking, Storage and Analysis. IEEE, 2020.

Diffenderfer, James, Daniel Osei-Kuffuor, and Harshitha Menon. "QDOT: Quantized Dot Product Kernel for Approximate High-Performance Computing." arXiv preprint arXiv:2105.00115 (2021).

Menon, Harshitha. “Approximate High Performance Computing: A fast and energy efficient computing paradigm in the post-moore era.” Presented at LANS seminar at Argonne National Laboratory. May 2020.

Parasyris, Konstantinos, Giorgis Georgakoudis, Harshitha Menon. “HPAC: A framework for High Performance Approximate Computing.” Software released at https://github.com/koparasy/HPAC

Parasyris, Konstantinos. "HPC-MixPBench: An HPC Benchmark Suite for Mixed-Precision Analysis." Presented at 2020 IEEE International Symposium on Workload Characterization (IISWC). IEEE, Oct 2020.

Laguna, Ignacio, Harshitha Menon, James Diffenderfer, Giorgis Georgakoudis, Daniel Osei-Kuffuor, Konstantinos Parasyris, Markus Schordan. “Challenges and Opportunities to Co-design Approximate Computing Scientific Applications.” Position paper at ASCR Workshop on Reimagining Codesign. Feb 2021.

Diffenderfer, James. "QDOT: Quantized Dot Product Kernel for Approximate High-Performance Computing." Presented at xSDK ECP meeting. July 2021.

Parasyris, Konstantinos. " HPAC: Evaluating Approximate Computing Techniques on HPC OpenMP Applications." Presented at Work in Progress seminar at Lawrence Livermore National Laboratory. Aug 2021.