Accelerating space-instrument data processing: benefits of on-board GPUs for scientific missions

Abstract

Future space missions generate data volumes that exceed downlink limits, making efficient on-board processing essential. Radiation-tested embedded GPUs offer high parallel performance and low power consumption, well suited to space systems. We evaluate three GPU-accelerated instrument pipelines on an NVIDIA Orin NX: the OBPMark #1.1 wide-field optical calibration and correction pipeline, the Euclid NIR preprocessing chain for cosmic-ray removal and calibration, and the THESEUS XGIS coded-mask imaging and transient-detection algorithm. The calibration pipeline saturates around 665–670 Mpix/s and the NIR chain peaks at 799 Mpix/s at full-focal-plane scale, both within a 6.5W module budget, while XGIS sky-image reconstruction completes in 0.562ms, roughly 1480× faster than an optimized CPU implementation. These results show that a single radiation-tested embedded GPU can deliver reliable, high-throughput, low-latency processing for next-generation scientific missions.