ICL Research Profile

Overview

The emergence of deep learning as a leading computational workload on large-scale cloud infrastructure installations has led to a plethora of releases of heavily specialized hardware accelerators that can tackle these types of problems much more efficiently. These new platforms offer new floating-point representation formats with reduced mantissa precision and modified exponent range at significantly higher throughput rates, which makes them more attractive in terms of improved performance and energy consumption. To leverage these unprecedented advances in computational power for numerical linear algebra solvers, the Mixed-Precision Numerical Computing effort endeavors to seize this new opportunity and deliver unprecedented levels of performance while still maintaining accuracy and stability on par with the classic IEEE 754 formats (e.g., single precision or double precision).