Correlator Technologies

Correlator technologies John W. Romein, ASTRON Science Goals for a Sub-mm Space Interferometer L orentz Center workshop - 10-14 November 2025 Agenda compare correlator technologies new GPU developments implications for space Science Goals for a Sub-mm Space Interferometer L orentz Center workshop - 10-14 November 2025 Data transport and correlation station processing combine data (real time) calibration, imaging, etc. disk station processing station processing “correlator” WAN this talk's focus Science Goals for a Sub-mm Space Interferometer L orentz Center workshop - 10-14 November 2025 Available correlator technologies •GPU •FPGA •(CPU) •DSP •ASIC Science Goals for a Sub-mm Space Interferometer L orentz Center workshop - 10-14 November 2025 GPU vs FPGA GPU advantages compute efficiency programming effort flexibility rapid technology development FPGA advantages I/O lifetime Science Goals for a Sub-mm Space Interferometer L orentz Center workshop - 10-14 November 2025 GPU tensor cores hardware matrix-multiplication units limited-precision input data ~10x faster than regular GPU cores accelerates training and inference  signal processing→ x B CA = + 32 bitsD16/8/4 bits 16/8/4 bits 32 bits Science Goals for a Sub-mm Space Interferometer L orentz Center workshop - 10-14 November 2025 GPU tensor cores tensor core compute power increases rapidly source: NVIDIA tensor cores Science Goals for a Sub-mm Space Interferometer L orentz Center workshop - 10-14 November 2025 The Tensor-Core Correlator1 GPU correlator library performs (tensor-core) computations highly optimized hides nasty details open source2 rapidly adopted by radio telescopes worldwide 1) J.W. Romein, The Tensor-Core Correlator, A&A 656(A52), Dec 2021 2) https://git.astron.nl/RD/tensor-core-correlator Science Goals for a Sub-mm Space Interferometer L orentz Center workshop - 10-14 November 2025 Performance RTX PRO 6000 Blackwell MaxQ (300W)