Nvidia has launched Ising, a family of open-source artificial intelligence models designed to tackle two of the central engineering obstacles in quantum computing: calibrating fragile qubits and correcting the errors they produce.
Qubits, the basic units of quantum information, are highly sensitive to environmental interference and require continuous tuning to remain usable, while error correction is essential to compensate for the noise that accumulates during calculations.
Ising addresses both problems through two distinct capabilities.
Ising Calibration is a vision-language model that interprets processor measurements to automate continuous calibration, reducing tuning time from days to hours by replacing manual adjustment with an AI-driven feedback loop.
Ising Decoding comprises two variants of a 3D convolutional neural network, one optimised for speed and one for accuracy, designed to identify and correct errors in quantum calculations faster than existing tools.
Nvidia said the decoding models deliver up to 2.5 times faster performance and 3 times higher accuracy than pyMatching, the open-source library that currently serves as the standard benchmark for quantum error correction decoding.
"AI is essential to making quantum computing practical," said Jensen Huang, Nvidia's founder and chief executive.
The models are already in use at a range of leading research institutions, including Fermi National Accelerator Laboratory, Harvard's John A. Paulson School of Engineering and Applied Sciences, Lawrence Berkeley National Laboratory's Advanced Quantum Testbed, Academia Sinica and the UK National Physical Laboratory, as well as quantum hardware company IQM Quantum Computers.
Nvidia is providing a cookbook of workflows, training data and NIM microservices alongside the models, allowing developers to fine-tune Ising to their specific quantum hardware while keeping proprietary data local.
Ising integrates with Nvidia's existing quantum computing tools, including CUDA-Q, its programming framework for hybrid quantum-classical systems, and NVQLink, the interconnect designed to link quantum processing units with Nvidia graphics processing units.
Related reading
- Samsung and AMD sign memory partnership deal to power next generation of AI chips
- Meta and Arm to co-develop data centre chips built for AI workloads
- Google: Retailers must clean product data for artificial intelligence
The models are available through Nvidia's open model portfolio on GitHub, Hugging Face and build.nvidia.com.
The release reflects Nvidia's broader push to position its hardware and software as essential infrastructure for quantum computing, a field that remains largely experimental but has attracted significant investment from technology companies, national laboratories and governments as researchers work toward systems capable of outperforming classical computers on commercially relevant problems.
The recap
- NVIDIA released Ising open-source AI models for quantum computing.
- Ising Decoding is up to 2.5x faster and 3x more accurate.
- Models and data available on GitHub, Hugging Face, build.nvidia.com.
