COSMICA: a novel parallel GPU code for Cosmic Rays propagation in heliosphere

TL;DR

COSMICA is a GPU-accelerated Monte Carlo simulation code that significantly speeds up cosmic ray propagation modeling in the heliosphere, enabling scalable, high-performance computations.

Contribution

The paper introduces a novel GPU-based Monte Carlo code for cosmic ray propagation that achieves substantial speed-ups and scalability compared to traditional CPU methods.

Findings

40X speed-up from initial GPU porting

1.5X additional speed-up from optimizations

Scalable computations across multiple GPUs

Abstract

The complex structure of interplanetary magnetic fields and their variability, due to solar activity, make it necessary to compute the Cosmic Ray (CR) modulation with numerical simulations. COde for a Speedy Monte Carlo (MC) Involving Cuda Architecture (COSMICA) is a MC code, solving backward-in-time the system of Stochastic Differential Equations (SDE) equivalent to the Parker Transport Equation (PTE). The Graphics Processing Unit (GPU) parallelization of COSMICA code is a game changer in this field because it reduces the computational time of a standard simulation from the order of hundred of minutes to few of them. Furthermore, the code is capable of distributing the computations on clusters of machines with multiple GPUs, opening the way for scaling. In COSMICA we implemented the synchronous broadcasting of memory access for evolving variable samples, the rounding of virtual particle set numbers, to fulfil the GPU blocks, and the exploitation of shared memory to free registers. Furthermore, we compactify the mathematical computations and pass to the lighter momentum formulation of SDE. The first porting of the code on GPU architecture brings it to a speed-up of 40X. The successful optimizations bring 1.5X speed-up.