RAFI -- A Ray/Work Forwarding Infrastructure for Data Parallel Multi-Node/Multi-GPU Computing
cs.DCcs.GR2026-05v1license
Abstract
We present RaFI, a CUDA and MPI based software framework that simplifies the task of building GPU-enabled data-parallel software where rays or similar work items need to migrate between different GPUs. RaFI provides a simple interface for CUDA kernels to forward such work items to other GPUs, while under the hood managing all the CUDA and MPI related work required to make this happen. We describe RaFI's motivation and implementation, and show its potential in several example applications.
Cite
@article{arxiv.2605.30294,
title = {RAFI -- A Ray/Work Forwarding Infrastructure for Data Parallel Multi-Node/Multi-GPU Computing},
author = {Ingo Wald and Serkan Demirci and Alper Sahistan and Stefan Zellmann and Andrea Paris and Patrick Moran and Milan Jaros and Tatiana von Landesberger and Ugur Gudukbay and Valerio Pascucci},
journal= {arXiv preprint arXiv:2605.30294},
year = {2026}
}