Title: Opticks GPU optical photon simulation via NVIDIA OptiX and a mental model for effective application of GPUs
Speaker: Dr. Simon Blyth (National Taiwan University)
Time: 10:00, Thursday 5th December 2019
Location: B326 Main Building
Simulations of optical photon transport are essential for the design, optimization and analysis of diverse optical systems from medical imaging scanners to neutrino detectors such as JUNO and dark matter search experiments.
However the computational and memory costs of simulating large numbers of photons using traditional sequential techniques cause processing and memory bottlenecks that limit our understanding. This optical photon simulation problem is particularly severe for the simulation of the JUNO response to cosmic muons, due to the large diameter of the scintillator and high photon yields.
Opticks is an open source project that applies state-of-the-art GPU ray tracing techniques via NVIDIA OptiX and integrates this with Geant4 simulations. This results in drastic speedups of optical photon simulation, exceeding a factor of 1000 compared to single threaded Geant4, with a single NVIDIA Turing GPU. This can solve the optical photon simulation problem for JUNO and other experiments limited by optical photons.
In addition to presenting Opticks the seminar will also attempt to convey the very different way in which you must think about your data processing to take best advantage of the abundant parallel teraflops provided by GPUs, covering parallel data structures and tools such as NumPy and CUDA Thrust.
About the speaker:
Dr. Simon Blyth received his D.Phil degree from Oxford University working on the DELPHI experiment at CERN on radiative muon decays. He continued at LEP on the L3 experiment as a Research Associate with Carnegie Mellon based at CERN, making the best L3 measurement of the forward-backward asymmetry of b quarks as well as managing the L3 luminosity detector and analysis. In 2002 he joined the Belle group at the National Taiwan University measuring six color suppressed B meson decay branching fractions. Subsequently he transitioned to the Daya Bay experiment, with National United University in Taiwan working on database management and software infrastructure development vital to Daya Bay. Returning to National Taiwan University in 2013 as he set himself the goal of solving the optical photon simulation problem faced by PMT based experiments using an innovative GPU based approach.