CuPy/Numba/CUDA-Based GPU Backend for TurboSETI Delivers ~7.5–10.6x Speedup in Single-Precision De-Doppler Processing

To the point

Researchers use ground-based radio telescopes to look for technosignatures by detecting narrow, drifting signals, and they sped up TurboSETI with a GPU backend (CuPy, Numba, and custom CUDA kernels) that runs in single precision, achieving about 7.5–10.6× faster than CPU (6.8–9.3× for double precision) as shown by a 105 GB observation dropping from 12 hours to 1 hour 45 minutes on a GTX 1070 Ti, in work by Luigi Cruz, Wael Farah, and Richard Elkins under CC BY 3.0, building on Enriquez’s TurboSETI foundations.