Back to Volume
Paper: A GPU-Based Visualization Method for Computing Dark Matter Annihilation Signal
Volume: 475, Astronomical Data Analysis Software and Systems XXII
Page: 73
Authors: Yang, L.; Szalay, A.
Abstract: We present a novel GPU-based visualization method for computing the dark matter annihilation signal for cosmological dark matter simulations. The technique increased the speed of rendering by more than 1,000 times. In a previous study, using a code running on regular CPUs, each particle's contribution was explicitly calculated pixel by pixel over a HEALPIX map, then remapped onto a Molleweide projection. Using Via Lactea II simulation (∼ 400M particles), it takes over 7 hours for a single thread CPU (∼3 GHz) to complete an all-sky map with NSIDE=512 resolution. Our novel method is based on a separate stereographic projection for each hemisphere, and a hardware accelerated rendering pipeline on a GPU (OpenGL). We project the particles instead of the celestial sphere to the tangent plane with a skewed flux profile appropriate for the STR projection. OpenGL's Point Sprite feature and shader language allow us to render those eccentric circular flux profiles at the rate of more than 10M particles per second. The new method can process a single snapshot of the Via Lactea II data in less than 1 minute with a single NVIDIA GTX 480 GPU, including I/O, with effective rendering time less than 24 seconds. Using an approximate normalization for the flux, accurate to 2.5% in total flux, the rendering can be done in less than 13 seconds. The stereographic images corresponding to the two hemispheres are then warped to an all-sky image in the Molleweide projection, and are in good agreement with the result from the regular CPU code, at similar resolution.
Back to Volume