Improving GPU Simulations of Spiking Neural P Systems

In this work we present further extensions and improvements of a Spiking Neural P system (for short, SNP systems) simulator on graphics processing units (for short, GPUs). Using previous results on representing SNP system computations using linear algebra, we analyze and implement a compu- tation simulation algorithm on the GPU. A two-level parallelism is introduced for the computation simulations. We also present a set of benchmark SNP sys- tems to stress test the simulation and show the increased performance obtained using GPUs over conventional CPUs. For a 16 neuron benchmark SNP system with 65536 nondeterministic rule selection choices, we report a 2.31 speedup of the GPU-based simulations over CPU-based simulations.