Nvidia goes quantum open‑source

Quantum computers store information in qubits, which are fragile enough that even routine operations can introduce errors. Nvidia said on April 14 it is releasing an open-source model family called Ising to help tune those machines and catch mistakes as they happen. (nvidianews.nvidia.com) Nvidia said useful quantum systems need two things that today’s hardware still struggles with: calibration, which is the repeated tuning of a processor, and error correction, which is the constant cleanup of bad results. The company said Ising launches with one model for calibration and another for decoding the error-correction signals. (developer.nvidia.com) The calibration model is a 35 billion-parameter vision-language model, which means it reads the images and charts quantum labs already use to diagnose hardware. Nvidia said that model can cut calibration work from days to hours by helping an automated agent decide what settings to change next. (nvidia.com) The decoding models are smaller three-dimensional convolutional neural networks, a type of pattern-finding system built for fast decisions. Nvidia said they run quantum error-correction decoding up to 2.5 times faster and 3 times more accurately than traditional approaches. (nvidianews.nvidia.com) The underlying problem is scale. Nvidia said leading quantum processors now make an error about once every thousand operations, while useful machines will need error rates closer to one in a trillion. (developer.nvidia.com) That gap is why Nvidia is pairing quantum hardware with classical computing it already sells. The company said Ising is part of its quantum-GPU supercomputing stack alongside CUDA-Q software, the cuQuantum simulation toolkit, and the NVQLink interconnect between quantum chips and graphics processing units. (nvidia.com) (nextplatform.com) Nvidia is also making the package open rather than shipping a closed service. The company said Ising includes pretrained weights, datasets, training frameworks, and tools for fine-tuning and deployment, with permissive licensing and documented data provenance. (nvidia.com) That matters for quantum labs because each machine has its own noise patterns and proprietary measurement data. Nvidia said users can retrain or specialize the models for their own processors without moving sensitive quantum processing unit data off-site. (developer.nvidia.com) The early adopters Nvidia named include Academia Sinica, Fermi National Accelerator Laboratory, Harvard John A. Paulson School of Engineering and Applied Sciences, Infleqtion, IQM Quantum Computers, Lawrence Berkeley National Laboratory’s Advanced Quantum Testbed, and the United Kingdom National Physical Laboratory. Northwestern University said data from Fermilab’s NEXUS facility helped train and benchmark the calibration model. (nvidianews.nvidia.com) (quantum.northwestern.edu) Nvidia’s quantum lead Sam Stanwyck told The Next Platform that the point is to make artificial intelligence the “control plane” for quantum hardware, with open models that researchers can keep improving. For now, the release is less about selling a finished quantum product than about embedding Nvidia software and graphics processors deeper into the workflows quantum developers already use. (nextplatform.com)