Although a compatible GPU is required I have to mention CUDA, or similar parallel language for at least coding each brain process into separate CPU cores:
In both the spatial network model I’m experimenting with and a cortical sheet HTM models are based upon are hexagonally located CCU’s that need to be horizontally interconnected into a single network. The question is then (regardless of being written for parallel or not) what is the most efficient way to start with perhaps the simplest thing possible and pass waves by making each 6 bit CCU output the Not/Complement of its input? Optionally it’s then possible to cancel out signals that arrive at the same time, timestep. Either way the wave pattern matches the (signal geometry) concordant pair numbers from live rats during 2 frame place avoidance, though I can’t yet in neuroscientific detail explain exactly why.
The way a GPU forces the problem to be sorted out into information pipelines always seemed to me like the ideal starting point, where from there it’s coded into the closest language to CUDA, using loops instead. After that anything that gets written would be easy to recode for parallel.
I thought it would be helpful to describe the programming dilemma I’m having. It’s for me a very first step in setting up a hexagonal array. What the horizontal connections between neighboring CCU’s look like in parallel code might help set a standard for all of us working with them.