Nvidia open‑sources 'Ising' models

Quantum computers use fragile bits called qubits, and Nvidia on April 14 released open artificial intelligence models meant to help keep those qubits working long enough to be useful. (nvidia.com) The new family is called Nvidia Ising, and Nvidia said it includes pre-trained models, a training framework, and a cookbook for two jobs: processor calibration and error-correction decoding. (nvidia.com) Calibration is the tuning step that sets a quantum chip’s controls, like adjusting thousands of radio knobs before a signal comes through cleanly. Nvidia said its Ising Calibration models automate that tuning and are available through its quantum computing platform. (nvidia.com) Error correction is the cleanup step that tries to spot and fix mistakes before they spread, because qubits are noisy and today’s best systems still make errors about once every thousand operations, Nvidia said. The company said useful machines would need error rates closer to one in a trillion. (nvidia.com) Nvidia said its Ising Decoding models run quantum error-correction decoding up to 2.5 times faster and 3 times more accurately than traditional approaches. Those figures came from Nvidia’s April 14 launch materials and have not yet been independently verified in the company’s announcement. (nvidia.com) The release is open source, which means researchers and hardware companies can use the models, retrain them, and fine-tune them instead of building everything from scratch. Nvidia said it is publishing the models with data, tooling, and deployment guidance. (nvidia.com) The market reaction was immediate on April 15. Bloomberg reported that South Korean software and cybersecurity stocks including Axgate and ICTK briefly hit their daily 30 percent trading limits, while China’s GuoChuang Software and QuantumCTek and Japan’s Fixstars each rose at least 8 percent. (bloomberg.com) That move reflected a wider bet that artificial intelligence could become part of the plumbing around quantum machines, not just a separate business line for chipmakers. Nvidia’s own description of Ising frames the software as a way to scale quantum systems with classical computing and machine learning. (nvidia.com) Nvidia is not building a general-purpose quantum computer here. It is selling picks and shovels for the companies that are, by offering software to tune chips faster and decode errors in real time. (nvidia.com) The next test is whether outside labs can reproduce Nvidia’s performance claims on real hardware. For now, the company has moved one of quantum computing’s hardest engineering problems into the open-source artificial intelligence race. (nvidia.com)