Presenter: Tommaso Boschi from ETH Zurich
This talk is designed for RSE with an interest in high-performance computing (HPC) and GPU acceleration. No particular skills are required to understand the presentation, but familiarity with C++/CUDA and GPU-parallel algorithms will help fully engage with the technical details.
I will address the computational bottleneck in radiative transfer cosmological simulations, specifically the ray-tracing algorithm for calculating hydrogen column density along ionising photon paths. To overcome this, we have updated pyC2Ray, a modern, massively parallel successor to the widely used C2Ray library. pyC2Ray implements the Accelerated Short-characteristics Octahedral RAy-tracing (ASORA) algorithm, a novel GPU-optimised method for multi-source ray-tracing. The library combines C++/CUDA for performance with a Python interface for usability, enabling customisation without sacrificing efficiency. Compared to C2Ray, the updated version of pyC2Ray now achieves a 800x speed-up, demonstrating the transformative potential of GPU acceleration and specialised software design in cosmological simulations. This work highlights how targeted algorithmic and architectural optimisations can unlock new scales of computational efficiency and physical accuracy.