Figure 2.

Comparison of execution times for each section of the Cpufit and Gpufit programs. The horizontal bars represent the amount of time spent in each section of the algorithm while processing a dataset consisting of $N=5.0\times {10}^6$ 2D Gaussian fits. Four additional sections are required for Gpufit: GPU memory allocation, copying data to the GPU, copying results from the GPU, and GPU memory de-allocation. All sections of the fit algorithm required less time when executed on the GPU, with the largest differences corresponding to the computational bottlenecks of Cpufit. For details of the simulated data and fit parameters, see the Supplementary Information.