Long data centre aisle of server racks with blue lighting, no people, logos or readable text
AI-generated image

Research

IBM and AMD release Pelagos, an open-source GPU tool for testing quantum error-correction decoders

Published · Last updated

Short answer: IBM and AMD have released Pelagos, an open-source framework that uses GPUs to evaluate quantum error-correction decoders at large scale, IBM announced at SC26 this week.

Error-correcting codes protect quantum information by repeatedly checking for signs of errors. Those checks produce data called syndromes, and a decoder works out the most likely correction. Proving that a decoder works well can mean simulating billions of sampled error patterns, which takes a lot of classical computing. Pelagos, named after the Greek word for "open sea", is meant to make those campaigns faster and easier to run.

The tool is built with the Triton programming framework and supports both AMD and NVIDIA GPUs, IBM says. Earlier studies showed up to 97 times the throughput of a single CPU core. New research released alongside it shows the GPU decoding stage scaling to roughly 7.5 times the speed of a single GPU across eight AMD Instinct MI300X and MI355X accelerators. The framework has also been run on Lux, an AMD-powered system at Oak Ridge National Laboratory.

IBM and AMD also published a survey paper on hierarchical decoding, which argues that fault-tolerant quantum computing will need advances in the classical systems that handle decoding, orchestration and data movement, not just in quantum hardware. Pelagos complements the quantum error-correction co-design platform that ORNL, IBM and AMD announced earlier this month.

The code is on GitHub, so other groups can check the performance claims for themselves. That makes it a practical release for decoder researchers rather than a hardware milestone.

← All news