The Hopper GPU delivered as much as 4.5 times the performance of its predecessor, writes Samuel K. Moore.

Here is the link: https://spectrum.ieee.org/mlperf-inferencing