High-Performance GPU Modeling of Triaxial Galaxy Clusters
Oral Defence Date:
Thursday, March 21, 2013 - 13:00
Professors Dragutin Petkovic, William Hsu and Andisheh Mahdavi (Physics)
To accurately model the dark matter within a galaxy cluster we present a simulation for non-symmetric, triaxially shaped galaxy clusters. To overcome the computational difficulties associated with this morphology, we use GPU computing, and the CUDA framework to parallelize the calculations. We treat each computation block on the GPU as a simulated pixel from a CCD image of a galaxy cluster, and each thread within a block calculates an energy level for the spectra of the gas. By combining every thread from the GPU blocks we get a complete image of a simulated cluster. Using GPUs rather than computing entirely on the CPU we achieve a 100-fold speed increase in generating the spectra of a simulated cluster. Our primary contributions to this project are the creation of the algorithm to solve for the spectra of a non-symmetric galaxy cluster, and parallelizing it to run within the CUDA computing framework.
GPU, galaxy cluster, dark matter, simulation, triaxial