Frank Willmore from Boise State University Research Computing presented advanced topics in high-performance computing (HPC) during the 2023 Research Computing Days event. He began by introducing different types of numbers like floating point and discussing number storage and representation in computing.
Willmore showed how rounding errors can occur in floating point numbers due to limited precision and demonstrated this through an exercise of summing floats in C code on the Borah HPC cluster. He also discussed how reductions in MPI parallel code can give varying results depending on process order and MPI implementation.
The presentation also covered the Central Limit Theorem and randomness as sources of error. Willmore showed an MPI code example to generate a normal distribution from random numbers. He emphasized understanding the limits of machine precision and making responsible choices about precision needs.
Full Video
Advanced Topics in High Performance Computing Video Transcript