Statistics – Computation
Scientific paper
Jan 2010
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2010aas...21540125j&link_type=abstract
American Astronomical Society, AAS Meeting #215, #401.25; Bulletin of the American Astronomical Society, Vol. 42, p.222
Statistics
Computation
Scientific paper
The use of graphics processing units (GPUs) for computationally intensive tasks has gained a significant amount of traction in the field of high performance computing. For highly parallelizable applications, GPUs provide a factor of 20-100 increase in speed over conventional CPUs which could dramatically improve the performance of the LSST data management and simulation systems. Here, we present the results of porting a source catalog simulation code (galfast) to NVIDIA GPUs that resulted in typical speedups of 25-100x over the prior CPU-based code.
galfast is a generator of realistic mock catalogs of astronomical sources (e.g., stars in the Galaxy), originally written to interpret SDSS observations. Within LSST, galfast stellar catalogs are used as inputs to image simulations (ImSim), as well by several several Science Collaborations to estimate LSST yields of various astrophysical objects. It is available as a web service at http://hybrid.mwscience.net.
The inner routines of galfast were modified to run on NVIDIA GPUs ( 3000 lines of CUDA C code, four weeks of porting time). The effort resulted in typical speedups of order 100x (single Tesla S1070 GPU vs. a single contemporary CPU core). The time to generate simulated catalogs of Milky Way stars decreased from days to under an hour, becoming entirely bound by disk data storage speed. In an extreme example, a catalog computation that used to take 1hr was reduced to 0.37 seconds. While algorithmic changes are partially responsible for the speedup, we conservatively estimate that at least a factor of 25 speedup came from the use of GPUs.
The success of this effort vividly demonstrates the ability of GPUs to bring significant speedup to LSST modeling and science. Work is ongoing to bring GPU-based acceleration to LSST ImSim code.
Cosic Kresimir
Ivezic Zeljko
Juric Mario
Vinkovic' Dejan
No associations
LandOfFree
Accelerating LSST Source Catalog Simulations with Graphics Processing Units does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Accelerating LSST Source Catalog Simulations with Graphics Processing Units, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Accelerating LSST Source Catalog Simulations with Graphics Processing Units will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-964776