Fat vs. thin threading approach on GPUs: application to stochastic simulation of chemical reactions

19 October 2010
<p>We explore two different threading approaches on a graphics processing<br /> unit (GPU) exploiting two different characteristics of the current GPU<br /> architecture. The fat thread approach tries to minimise data access time<br /> by relying on shared memory and registers potentially sacrificing<br /> parallelism. The thin thread approach maximises parallelism and tries to<br /> hide access latencies. We apply these two approaches to the parallel<br /> stochastic simulation of chemical reaction systems using the stochastic<br /> simulation algorithm (SSA) by Gillespie. In these cases, the proposed<br /> thin thread approach shows comparable performance while eliminating the<br /> limitation of the reaction system's size.</p><p>Link to paper:&nbsp;</p> <p><a target="_blank" href="http://people.maths.ox.ac.uk/erban/papers/paperCUDA.pdf">http://people.maths.ox.ac.uk/erban/papers/paperCUDA.pdf</a></p>
  • Junior Applied Mathematics Seminar