Speaker
Michael Steven Farrington
(Harvard University (US))
Description
In this talk, we demonstrate a GPU-friendly extension to the FHist.jl. Within the analysis computing ecosystem, moving to GPU has been a steady trend, however, since any event we select almost always goes to a histogram, it would be a huge bottleneck if histogramming requires moving data GPU -> CPU.
We briefly discuss the design and implementation of GPU-backed histogram, and then show some benchmarks to inform the audience what kind of performance tradeoff is being made when you use GPU for histogramming.
Authors
Jerry 🦑 Ling
(Harvard University (US))
Michael Steven Farrington
(Harvard University (US))
