Nvidia Simulates 104-Qubit System on JUPITER, Cutting Physical Qubit Needs by 90%

Quantum processors are vulnerable to environmental interference, which introduces errors and makes accurate calculations harder as systems grow larger. Fault tolerance methods aim to maintain accuracy despite errors in individual physical components. One approach produces specially prepared quantum resources called magic states, which are consumed during operations required for fault-tolerant quantum calculations. Producing these resources requires substantial computing capacity.

Nvidia's simulation incorporated realistic hardware noise, allowing researchers to investigate errors that earlier simulations under ideal conditions could not adequately represent. "Useful quantum computing will require fantastically complex, fault-tolerant systems—and understanding how they behave at scale is essential to building them," said Jensen Huang, Nvidia's CEO and founder. "Our results from Jülich show how quantum processors and GPU supercomputers can work together to tackle research that would otherwise be out of reach—accelerating the discovery of transformative quantum applications."

Earlier research at Jülich simulated 50 qubits under ideal conditions, while the latest experiment incorporated realistic hardware noise. The experiment used classical computing infrastructure to simulate quantum operations rather than testing the technique on a functioning quantum processor.

JUPITER, located in North Rhine-Westphalia, Germany, provided the computing capacity needed to model complex quantum operations. The supercomputer used Nvidia GH200 processors, with researchers scaling their methods to 8,192 GPUs. Before moving to JUPITER, the team developed and tested its simulation methods using Nvidia's GB200 NVL72 system at the company's Accelerated Quantum Research Center. Researchers expanded initial tests from four graphics processors to hundreds, evaluating performance and accuracy before deploying the methods on the larger machine. These preparations reduced simulations that previously required several weeks to just days.

The team also used Nvidia's cuQuantum software and CUDA Q platform to simulate quantum circuits and represent the operations involved in magic state cultivation. The results could help researchers assess quantum processor designs before constructing hardware, potentially reducing the need for costly physical experiments. TechRadar reported the advance could improve drug discovery, financial modeling and other fields. However, Nvidia's result remains a simulation, and further experimental work must establish whether the technique delivers comparable benefits on actual quantum hardware.