tweska/cern-ssp

CERN Summer Student Project

ROOT GPU Optimization Experiments for High Energy Physics Analysis

Dependencies and Compilers

Package	Version	Notes
ROOT	6.33.01 (5b7b6cc0)	Build from source
Google Test	1.15.2 (b514bdc8)	Build from source
GNU Compiler Collection (gcc/g++)	12.3.0
CUDA Toolkit (includes nvcc)	12.5

Performance Results

The following results are obtained on a machine with an AMD Ryzen 7 5700g processor and NVIDIA GeForce RTX 3060 GPU, by running the ./runBenchmarks.sh script in the root directory.

Note that some runtimes might not add up to the total time reported. This is due to the intermediate times being rounded.

Batched Histogram

	CPU Runtime	GPU Runtime	GPU Percentage	Speedup
Transfer	N/A	543ms	97.8%	N/A
Fill	3383ms	12ms	2.2%	281.9x
Result	N/A	0ms	0.0%	N/A
Total	3383ms	555ms	100.0%	6.1x

DiMuon

	CPU Runtime	GPU Runtime	GPU Percentage	Speedup
Transfer	N/A	163ms	58.3%	N/A
Define + Fill	788ms	116ms	41.7%	6.8x
Result	N/A	0ms	0.0%	N/A
Total	788ms	278ms	100.0%	2.8x

FoldedWMass

	CPU Runtime	GPU Runtime	GPU Percentage	Speedup
Transfer	N/A	0ms	0.0%	N/A
Define + Fill	17792ms	172ms	98.9%	103.4x
Result	N/A	1ms	0.6%	N/A
Total	17792ms	174ms	100.0%	102.3x