Hardware

IBM Research and AMD open-source Pelagos GPU decoder for qLDPC error correction

IBM Research and AMD released Pelagos (PyPI: pelagos-decoder), an Apache 2.0 licensed, GPU-accelerated decoding framework for quantum error correction, announced at SC26. It runs a Relay-BP-S algorithm in fused Triton kernels and targets AMD ROCm/HIP, NVIDIA CUDA, and Intel SPIR-V GPUs, with up to 97x single-GPU speedup versus one CPU core, according to the report. On 8x AMD Instinct MI355X GPUs it reached 3,845 decodes/s on the One-Gross [[144, 12, 12]] code (6.23x end-to-end scaling), and on ORNL's Lux MI355X system the Two-Gross [[288, 12, 18]] code reached 1,265 decodes/s (7.45x GPU decode scaling). A multi-GPU scaling study (a preprint, not yet peer reviewed) cut eight-GPU host preprocessing on two-gross circuits from 3.21 seconds to 0.73 seconds, and a survey co-authored by AMD, HPE, IBM, IQM, Riverlane, and Xanadu maps hierarchical decoding approaches.

Why it mattersEvaluating qLDPC codes such as bivariate-bicycle memories requires simulating billions of circuit-level error syndromes to resolve logical error rates down to 10^-12, which demands large-scale decoding compute. A vendor-neutral, open-source tool lets researchers run those evaluations on GPUs from several manufacturers.

Read the full story at Quantum Computing Report →

Summarized by AI from the linked article.