Overview
- Nvidia’s Ising family, announced Tuesday, offers open-source AI models for tuning quantum processors and for decoding errors fast enough to keep qubits usable.
- The suite spans Ising Calibration plus two Ising Decoding convolutional networks that target either lower latency or higher accuracy.
- Nvidia reports decoding that runs up to 2.5 times faster and reaches up to three times higher accuracy than pyMatching, and it says training needs drop by about 10x.
- The code and weights are available on GitHub, Hugging Face, and build.nvidia.com, though real-time deployment relies on Nvidia’s CUDA-Q software and its NVQLink GPU–QPU link.
- Early users include Harvard, Fermilab, Lawrence Berkeley’s Advanced Quantum Testbed, and IQM, and following Wednesday’s coverage Nvidia shares rose about 3.8% as IonQ and D-Wave climbed in premarket trading.