Phased Array Scaling Benchmark

What this benchmark is

A 15.5 GHz phased array, taken from a single patch element to a 32×32 array of 1,024 elements, run in Nullspace EM and in a legacy commercial solver at every step in the series. Cases through 16×16 ran on the same 64 GB engineering laptop for both tools. The customer who ran the comparison assessed the results as equivalent in accuracy across a number of standard engineering measures.

Nullspace EM completed every case. The legacy tool completed cases through 8×8, then exhausted memory at 16×16 and 32×32 and did not finish on the laptop or on a 256 GB workstation. At every size where both tools finished, Nullspace EM was between 3× and 23× faster and used roughly an order of magnitude less memory. The 1,024-element case finished in 5.7 hours at 170 GB. The specific sizes are an illustration, not a limit: what the benchmark documents is the scaling behavior, where memory grows at a rate that keeps large problems within reach of hardware an engineering team already owns.

Inside the benchmark

• The full results table: runtime and peak memory for both tools at every array size, one element through 1,024

• Test methodology: geometry, lattice spacing, feed configuration, mesh rules, and the solver settings used for each tool

• Hardware specifications for both platforms, and the conditions under which each case failed

• Caveats and limitations: tested conditions, what was not tested, and how to request a reproducible test package