Faster Scientific Image Analysis with NVIDIA cuPhoton | NVIDIA Technical Blog

This post introduces NVIDIA cuPhoton, an open-source CUDA-X toolkit providing GPU-accelerated modules (xDataReader, xRep, xPois, xFit, xScan) for scientific image analysis. It details a workflow for loading FITS data directly onto the GPU using xDataReader, aligning images with xRep, PSF matching and subtraction with xPois, dipole fitting with xFit, and candidate review with xScan. The post highlights performance gains of up to 14,900x for image loading and 14,550x for signal processing compared to an x86 CPU baseline on representative workloads. It also demonstrates scalability across multi-GPU, multi-node systems like NVIDIA Grace Blackwell and Vera Rubin, and provides a walkthrough of a synthetic astronomical image processing pipeline.