Quantum Machines runs NVIDIA's CUDA-Q QEC decoders in real time on its OPX1000 controller
Short answer: Quantum Machines has connected its OPX1000 quantum control system to NVIDIA's CUDA-Q QEC decoding software, so error-correction data can be decoded while a quantum circuit is still running rather than afterwards. The company published its first results in a release dated Chicago, 16 November.
Quantum computers make errors constantly, and a classical computer has to read the stream of error checks, called syndromes, and work out corrections fast enough to keep up. QM built a bridge to NVIDIA's CUDA-Q Realtime API, and its controller sends syndromes over NVIDIA NVQLink to a decoding server and gets corrections back within the same shot.
In the test, a single CUDA-Q program ran a surface-code memory experiment on the OPX1000, with 10,000 shots for each of nine set-ups: code distances 7, 11 and 15 at 25, 50 and 75 rounds. The syndromes were pre-sampled at a 0.1% physical error rate and decoded with PyMatching on an NVIDIA GH200 server. QM reports a steady 2.4 microsecond two-way link between controller and decoder, and decoding of 0.19 to 0.22 microseconds per round at distance 7 and 0.56 to 0.67 microseconds at distance 11, below the 1 microsecond mark it uses for keeping pace with a fast error-correction cycle. At distance 15 it took 1.2 to 1.5 microseconds per round, which QM says is where more optimisation matters.
This first version routes data through the host CPU, with direct GPU dispatch to follow, and supports static memory circuits only. Quantum Computing Report notes that early access for OPX1000 customers opens in December 2026. "Error correction only works if decoding can keep pace with the quantum hardware," said QM co-founder and chief technology officer Yonatan Cohen.
The useful part for labs is flexibility: users can switch between decoders, such as PyMatching or NVIDIA's fusion decoder, with a configuration change. The open question is how the numbers hold up with live qubits and at larger code distances.
