Nvidia has completed a 104-qubit simulation of a fault-tolerant quantum computer on JUPITER at Germany’s Jülich Supercomputing Centre. Described by the company as the largest comprehensive simulation of such a system to date, the work modelled a machine capable of detecting and correcting errors during operation. Crucially, the entire simulation ran on conventional computing hardware, not a quantum processor, while incorporating the noise that disrupts real quantum devices.
The researchers evaluated magic state cultivation, a technique for preparing specialised quantum states required for certain calculations. Producing these states can consume substantial hardware resources; Nvidia says its approach requires 90% fewer qubits than earlier methods. By incorporating realistic disturbances rather than assuming error-free operations, the simulation allowed researchers to assess the technique under more practical conditions. It builds on a November 2025 simulation in which Jülich researchers modelled 50 qubits on JUPITER under ideal, noise-free conditions.
The latest project used 8,192 Nvidia GH200 GPUs. Before scaling up to JUPITER, researchers developed and tested their methods on a smaller system at Nvidia’s Accelerated Quantum Research Center, reducing simulation work from weeks to days, according to the company. JUPITER, which opened in September 2025 as Europe’s first exascale supercomputer, can perform more than a billion billion calculations per second. Such simulations could help researchers identify weaknesses and evaluate proposed quantum computing designs before committing resources to physical hardware, potentially reducing the need for repeated hardware testing.
