Table 1. Compute systems used and their configuration. The machine “xrmlite” is a Linux workstation at Northwestern University. The cluster “bebop” is at the Laboratory Computing Resource Center (LCRC) at Argonne National Laboratory (with four Northwestern University nodes included), while the cluster “theta” is at the Argonne Leadership Computing Facility (ALCF) [69,70]. With both “bebop” and “theta,” we used only a fraction of the large number of available nodes for the strong scaling studies described in Sec. 6.2. We note that “xrmlite” has two Quadro P5000 GPUs, each connected to the CPU via PCIe3.0.

Compute System	Processor	Cores/node	Memory/node	Interconnect
xrmlite	2x Xeon E5-2620 v4	16	512 GB	N/A
bebop/LCRC	Xeon Phi 7250	68	192 GB	Omni-Path Fabric
theta/ALCF	Xeon Phi 7230	64	192 GB	Aries Dragonfly