Nvidia releases open AI models for quantum computing tasks — ‘Ising’ said to be 2.5x faster and 3x more accurate than existing tools for decoding

Nvidia has released Ising, a family of open-source AI models for quantum processor calibration and real-time error correction decoding. Integrated with Nvidia’s CUDA-Q quantum software platform and the NVQLink QPU-GPU interconnect, which was first introduced last October, the Ising models have been released on GitHub, Hugging Face, and build.nvidia.com.

Nvidia has designed Ising specifically to target two bottlenecks that exist between current quantum hardware and fault-tolerant computing: calibration and decoding. The former is the manual process a QPU so that its qubits behave consistently, while the latter translates redundant measurements from an error-corrected logical qubit into a correction signal, and it only works if it keeps pace with the rate at which new errors appear on the processor.

Ising Calibration is a 35-billion-parameter vision-language model fine-tuned to read experimental measurements from a quantum processing unit (QPU) and infer the adjustments needed to tune it. This reduces calibration time from days to hours when paired with an agent, Nvidia claims.

Article continues below

You may like

The Ising Decoding family, meanwhile, comprises two variants of a 3D convolutional neural network — 0.9 million and 1.8 million parameters, optimized for speed and accuracy, respectively — that perform pre-decoding for surface-code quantum error correction. Nvidia has benchmarked the decoder at 2.5 times faster and three times more accurate than pyMatching, which is the open-source decoder that most quantum research groups use, while requiring ten times less training data.

Sam Stanwyck, director of quantum product at Nvidia, told The Next Platform that today’s best quantum processors produce an error roughly once every thousand operations, and that the logical error rate is directly tied to how quickly decoding runs alongside the hardware. A 2.5 times speedup therefore raises the ceiling on how many gate operations a quantum processor can sustain before its logical qubits break down.

While Ising is open-source, the stack it sits on isn’t. The decoder needs NVQLink’s low-latency interconnect to feed measurement data to a GPU inside the decoding window. The calibration workflows run through CUDA-Q, and the deployment tooling targets Nvidia hardware exclusively.

Nvidia has run with the same pattern with the likes of Nemotron, Cosmos, and GR00T — open the models but keep the surrounding platform proprietary, thereby driving GPU dependencies through the workflow. That way, Nvidia remains deeply integrated with the quantum computing industry despite not building quantum hardware.

Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.

Named adopters include Fermilab, Harvard, the UK National Physical Laboratory, Lawrence Berkeley National Laboratory’s Advanced Quantum Testbed, IQM Quantum Computers, Infleqtion, and IonQ, which is using Ising Calibration directly.

Follow Tom’s Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.