NVIDIA Ising: The AI Models Tackling Quantum Computing's Scaling Barrier
NVIDIA unveils Ising, open-source AI models for quantum processor calibration and error correction, integrated with CUDA-Q to unlock large-scale quantum computing.
NVIDIA has released Ising, an open-source family of AI models built to solve the two hardest unsolved problems blocking quantum computers from scaling into reliable production systems: qubit calibration and real-time error-correction decoding. Both issues are essentially high-dimensional signal-processing challenges, which is exactly the class of problem modern AI models excel at.
The suite includes Ising Calibration, a 35-billion-parameter vision-language model that reduces multi-day, physicist-driven calibration workflows to automated runs measured in hours. Ising Decoding, a pair of 3D convolutional neural networks, targets real-time quantum error correction and reportedly delivers up to 2.5x faster throughput and up to 3x higher accuracy than pyMatching, the current open-source decoding standard — a meaningful gain since the decoder sits on the critical path of every quantum gate operation.
Shipped under NVIDIA's Open Model License with weights, training tools and deployment recipes included, Ising is positioned as a shared foundation layer for the quantum industry rather than a proprietary product. It deploys via CUDA-Q on GPU infrastructure (A100/H100) and connects to quantum processors through NVQLink, enabling sub-millisecond latency for live decoding.
For engineers, the significance lies in turning two of quantum computing's hardest scaling bottlenecks into tractable AI inference problems, giving hardware builders a more concrete path from hundreds to potentially millions of qubits.