NERSC’s new “Perlmutter” supercomputer was brought online in April 2021. We have identified a set of applications from our user community (eg. https://github.com/marius311/CMBLensing.jl). Of particular interest are applications that make use of multiple GPUs per node. We report the performance of these applications at scale. We hope that our findings will inspire more users to embrace Julia to develop extreme-scale simulation codes.
NERSC’s new “Perlmutter” supercomputer (https://www.nersc.gov/systems/perlmutter/) was brought online in April 2021. The first phase consists of 1500 nodes with 1 AMD “Milan” CPU and 4 NVIDIA A100 GPUs each. In this talk we give an account of our experiences deploying Julia applications on Perlmutter. We manage a central installation of Julia 1.6.x (containing a minimal set of packages -- such as MPI and IJulia -- which require special configuration to run on Perlmutter). We have identified a set of applications from our user community (eg. https://github.com/marius311/CMBLensing.jl). Of particular interest are applications that make use of multiple GPUs per node. We report the performance of these applications at scale. We hope that our findings will inspire more users to embrace Julia to develop extreme-scale simulation codes.