A central spike queue could mean that only those spikes scheduled for arrival get processed, rather than having to check for spike arrival at each synapse. Perhaps this could save a memory access or so for each synapse that doesn't have an arriving spike. In principle, it should save on the number of memory accesses needed to roll a queue forward for every synapse. It may also improve coalescing of some memory accesses. This is a non-trivial project; good for someone who is interested in going into the details of GPU performance and optimization.
A central spike queue could mean that only those spikes scheduled for arrival get processed, rather than having to check for spike arrival at each synapse. Perhaps this could save a memory access or so for each synapse that doesn't have an arriving spike. In principle, it should save on the number of memory accesses needed to roll a queue forward for every synapse. It may also improve coalescing of some memory accesses. This is a non-trivial project; good for someone who is interested in going into the details of GPU performance and optimization.