Application speedups over A100

Nvidia highlighted four HPC applications when presenting potential speedups for their new Hopper architecture at GTC 2022: Climate modeling, genomics, lattice quantum chromodynamics, and 8K-point three-dimensional FFT.  The FFT in particular was shown to perform about 30x faster over A100 on a multi-GPU H100 system where the GPUs were connected via an NVLink network. It will be interesting to see how our own applications perform on such systems.