Nvidia uses JUPITER to simulate error-correcting quantum computer
The 104-qubit simulation included hardware noise, while Nvidia said its approach required 90% fewer qubits than earlier methods.
Nvidia has modeled a 104-qubit quantum computer capable of correcting errors as it runs, using Germany’s JUPITER supercomputer. Nvidia called the work the largest full simulation yet of a machine with that capability.
The experiment ran entirely on conventional computers rather than a quantum processor. It is intended to help researchers evaluate quantum computing designs before committing to building the hardware.
The team examined magic state cultivation, a technique that creates special quantum states needed for many operations. Making those states can require numerous extra qubits; Nvidia said its approach reduced that requirement by 90% compared with earlier methods.
The model included interference affecting actual quantum hardware. Qubits are sensitive to such disturbances, including noise generated by control equipment. An earlier Jülich simulation, conducted on JUPITER in November 2025, covered 50 qubits but assumed conditions without noise.
Located at Germany’s Jülich Supercomputing Centre, JUPITER opened in September 2025 and was Europe’s first exascale supercomputer. The latest experiment employed 8,192 GPUs with Nvidia GH200 chips. Nvidia said preliminary development and testing on a smaller system in its quantum lab shortened the work from weeks to days.
