THOR: GPU-accelerated synthetic observations
A portable, massively parallel framework for radiative transfer and synthetic observations of galaxy formation simulations.
THOR is a GPU-accelerated framework for synthetic observations and post-processing of galaxy formation simulations. It forward-models stellar and gas emission and absorption, including Monte Carlo radiative transfer (MCRT) for resonant emission lines, directly from complex astrophysical environments.
Written in C++20 and SYCL, THOR runs across CPUs and GPUs while scaling to distributed-memory systems with MPI. The same framework supports idealized geometries, uniform grids, and the native particle or mesh structures used by cosmological simulation codes. This combination makes it possible to move from small verification problems to full simulation volumes without changing the underlying analysis workflow.
Performance
Code-paper benchmark of the Lyman-alpha “Neufeld solution” (T=10⁴ K, τ₀=10⁶) across resonant-line MCRT codes. The July 2025 comparison used Clang 18, an Intel 8360Y CPU with 36 cores at 2.40 GHz, and an NVIDIA A100 40 GB GPU.
Lyman-alpha radiative transfer weak scaling to approximately 200 billion volume elements, with memory distributed across as many as 15,000 CPU cores and 200 GPUs.
Code paper
The architecture, verification suite, science applications, and performance benchmarks are presented in Byrohl & Nelson (2025), THOR: a GPU-accelerated and MPI-parallel radiative transfer code. See the THOR papers page for related applications and work based on its predecessor codes, voroILTIS and ILTIS.