Computational Modeling and Simulations
Nikolay Simakov 

gapsheader.png
In the computational biophysics Poisson and Poisson-Boltzmann equations (PE and PBE) are widely used as relatively fast way to estimate electrostatic contribution to free energy. Often many applications require PE to be solved for a large number of system configurations. This can be a highly computationally demanding task. Modern graphical processing units (GPU) have enjoyed rapid progress over the last decade providing up to several TFLOP (single precision) of computational power. Therefore, it appears attractive to utilize GPU to solve PE and PBE.


I have implemented GPU and CPU version of Poisson equation solver. The initial speed-up over CPU was about hundred times. However, the successful optimization CPU version with emphasis on use SIMD instructions results in 4 times boost in performance of CPU version. This and the update of the CPU to more modern CPU drop the GPU’s outperformance to more modest 7 times (see Figure 1). Still this is a worth a consideration choice given that multiple GPUs can be installed into single system. The performance of such system will be comparable to that of a small cluster.


GAPS_Performancef

Figure 1 - Benchmark of CPU and GPU solver for calculation of single ion potential in α-hemolysin; next to columns is shown the improvement of GPU solver over CPU version run on 4 cores; the calculations were done with grid scale of 4 grid/Å and with grid size of 5143. Calculations were done on a single CPU desktop with Intel Core i7 930 processor and NVIDIA GTX 480 GPU.


This work was done in Dr. Maria Kurnikova research group. For further details see Dr. Maria Kurnikova research group web-site and the draft of our article here.

Nikolay Simakov