Variable Precision Computing for Scientific Applications

Today, we are complacent and wasteful in how we use floating point numbers in our scientific application codes. However, the increasing cost of data movement, coupled with the significant discrepancy in computational throughput between the different levels of precision in HPC hardware, is reshaping how we implement floating point computing and design efficient HPC simulation codes.

This talk will discuss some of our recent efforts in utilizing variable precision computing for scientific applications, to increase performance and make efficient use of existing and emerging HPC platforms.

Location: Cumberland Amphitheatre Date: August 28, 2019 Time: 9:15 am - 9:45 am Daniel Osei-Kuffuor (LLNL)