Contributor
Kavishka Attanayake

Acts GPU R&D - Optimization of GPU tracking pipeline


Mentors
Beomki Yeo, Charles Leggett
Organization
CERN-HSF
Technologies
c++, cuda
Topics
science, hpc, GPU Programming
Traccc is a demonstrator for GPU tracking algorithms. This project focuses on optimizing GPU utilization of traccc using CUDA Multi Process Service (MPS) and improving compute times by introducing caching allocators, moreover if time permits, using CUDA-MIG to improve GPU utilization as well. Upon completion a report comparing the benchmarking results between the CUDA implementation using MPS and multi-process CPU implementation will be submitted.