Well-deserved! Dongarra created ATLAS (automatically-tuned linear algebra software), which made fast linear algebra (BLAS and LAPACK) available to everyone, rather than just those willing to pay the big bucks for chip vendors' proprietary libraries. Basically, ATLAS tries out a bunch of different strategies for each routine at compile/install time, and picks the fastest, adapting to chip pipelines, cache sizes, etc. I'm not sure how it fits in to the modern world of GPUs and TPUs, but it made a lot of scientific software run faster.
Tools like atlas are often used on CPUs. GPUs and TPUs have their own fast linear algebra implementations.