Joint INFN-UNIMI-UNIMIB Pheno Seminars

VegasFlow and PDFFlow: offloading Monte Carlo simulations to hardware accelerators

by Dr Juan M. Cruz Martínez (University of Milan)



Graphical Processing Units (GPUs) are surpassing CPUs as target devices for many "embarrassingly parallel" calculations.
While this trend is true for many private industries, in the field of High Energy Physics they are still underused and their potential unrealised.

In this talk I present the VegasFlow and PDFFlow libraries for the offloading of complicated MonteCarlo-based simulations to hardware accelerators such as GPUs. These libraries, based on a "tensorization" of the calculation, aim to easy the task of the developers when trying to  transition to new hardware where past expertise on high performance calculations on CPU could even be a hindrance.