This sample is WIP. In that it is not very well tested yet. The intended behavior is that for a multi-gpu context, this will simulate particles on one GPU, and render them on the other.
In general, because the particles are small and plentiful,
you should run with -w=1000 -h=1000 otherwise you may not
see them`
Given that the simulation and rendering are independent operations, this should provide some amount of benefit.
You can make tweaks to:
particle_data_shared.h:TOTAL_PARTICLES
particle_velocity_update.comp:PARTICLE_SPLIT
in order to change the proportion of compute vs raster work.