TABLE I.

Average (amortized) run time per simulation time step for each method. The fastest method (cubewalkers) for the two hardware configurations is bolded. Note that cubewalkers on consumer hardware outperforms parallel adaptations of other tools running on specialty high-performance computing hardware.

Software	Hardware	time $(\mu \mathrm{s})$
cubewalkers	consumer	0.11
cubewalkers	specialty	0.067
cana (serial)	consumer	40
cana (parallel)	specialty	1.1
booleannet (serial)	consumer	1300
booleannet (parallel)	specialty	37