GPU Photon Transport Simulation StudiesMary Murphy Undergraduate, UW-MadisonDmitry Chirkin IceCube at UW-MadisonTareq AbuZayyad IceCube at UW-River Falls
ObjectivesComparison of i3mcml and ppc-gpuCheck for muon and flasher simulation output agreementCompare performance with time trialsTest simulation parameters in ppc-gpuVary light yield, ice properties, muon track generator, sensitivity parameters, etc. to observe and analyze response
Comparison of i3mcml and ppc-gpuTime trials run with same input f2k muons, same GPU deviceIn both simulations, subsequent versions reduced run timeLatest version of ppc-gpu is 123 times faster than CPU versionMuon Simulation Run Time Comparison
SimulationI3mcml v0.9.6.2I3mcml v0.9.8.2Ppc-gpu v8Ppc-gpu v31Real Time (s)53.523.1535.118.0CPU Process Time (s)53.22.906.422.20CPU System Time (s)0.290.390.830.69GPU Time (s)--19.8627.8014.51Number of Hits41,39641,20241,40340,556
Comparison of i3mcml and ppc-gpuFlasher Simulation Run Time ComparisonTime trials run flashing 1.e8 photons on dom 20, string 20Similar time trials run on different locations around detector to eliminate any variations due possibly to which dom is flashing
SimulationI3mcml v0.9.6.2I3mcml v0.9.8.2Ppc-gpu v8Ppc-gpu v31Real Time (s)12.585.119.184.46CPU Process Time (s)12.230.560.610.30CPU System Time (s)0.210.340.180.24GPU Time (s)--4.218.613.97Number of Hits150,778150,164152,294150,060
Comparison of i3mcml and ppc-gpuDemonstrates strong agreement!DOM Occupancy plot for flasher run down string 83
Simulation parameters in ppc-gpuComparison of Assembly and GPU versions of PPC
Simulation parameters in ppc-gpuCascade light yield parameterizationnew M. Kowalskis vs. older Ch. Wiebusch cascade parametrizations: up to 20% difference; same bare muon parametrization
Simulation parameters in ppc-gpuAHA ice propertiesbulk ice propertiesAngular Sensitivity
Simulation parameters in ppc-gpuAngular SensitivityAHA ice propertiesSimulation parameters in ppc-gpuAngular Sensitivitybulk ice properties
Simulation parameters in ppc-gpuAngular SensitivityAHA ice propertiesbulk ice properties
Simulation parameters in ppc-gpuMuon Track Generator
COG distribution for MMC / CORSIKA muons after Simple Majority Triggering Multiplicity 8 applied
Simulation parameters in ppc-gpuMuon Track Generator
Simulation parameters in ppc-gpuOM Sensitivity
The EndBack-up slides follow.
Output comparison of ppc-gpu and i3mcml
MKOW (red) vs. old default (black) COG distribution
DOM Occupancy with input CORSIKA muon bundles, in bulk and aha model ice
COG distribution for various OM sensitivity levels
COG distribution for various OM sensitivity levels, in bulk ice
Top Related