Parallel Solutions for Voxel-Based Simulations of Reaction-Diffusion Systems

Table 4

Evaluation of the execution times considering the three different CUDA devices available on the second cluster nodes for the 4096-compartment system. The number of blocks we consider is multiple of the number of SMs available on these devices. Speedup values are computed considering the TimePerStep values.

Device	Blocks threads	TransferTime (ms)	TimePerStep (ms)	Speedup
CPU	—	88.9	29.9	—
GTX-580	32-32	177.0	4.9	6.1
	64-32	205.0	2.4	12.3
K20	13-128	198.6	2.9	10.2
	26-128	182.7	1.5	20.5
	52-128	189.9	0.8	37.8
	78-128	208.6	0.9	32.8
GTX-Titan	14-128	137.4	2.2	13.5
	28-128	136.1	1.1	27.6
	56-128	139.2	0.5	57.4
	84-128	146.5	0.6	48.1