OMEN Scaling on a variety of HPC Platforms
We have access to a variety of sytems and performed scaling experiments on these machines. These include a local cluster at Purdue "Steele" consistes of dual quad-core Intel chips which are simply interconnected with Gigabit Ethernet. This results in leveling off of scaling at around 4k cores. The Indiana University / Purdue system Big Red is based on an IBM JS21 system connected with a fast network. In terms of raw performance that system performs best. Purdue also hosts a SiCortex with around 3k cores. These cores run significantly slower than than standard CPUs which can immediately seen with the direct comparison.
The scaling experiment at 4k cores does not involve the outermost voltage loop. As typical device curves contain 50-100 bias points for a whole device characteristic, we can estimate that OMEN will run just fine on a machine of 200,000+ cores.