Riverlane reports GPU cuts quantum calibration cycle to 2.8 seconds

Riverlane and Nvidia demonstrated a quantum error correction training cycle taking 2.8 seconds on an H100 GPU, compared with approximately three minutes on a single CPU core. The collaboration uses AI to learn decoder settings from experimental data, reducing the need for engineers to adjust individual parameters. Decoders interpret quantum processor signals to identify likely errors. Their settings need recalibration as hardware conditions change. Riverlane's GPU-adapted decoder processed more than four million surface-code syndromes per second, compared with roughly 59,000 on a single CPU core. Riverlane says the speed could enable calibration before individual experiments; the comparison uses a single CPU core.