NVIDIA simulates 104-qubit fault-tolerant quantum computing on 8,192 GPUs

On Monday, October 19 (US Eastern time), NVIDIA announced that its researchers had carried out a 104-qubit fault-tolerant quantum computing simulation using 8,192 GPUs on the JUPITER supercomputer at Germany's Jülich Supercomputing Centre. According to the report, this marks a scale record for the company in this line of work.

The effort was not only about pushing the simulation to 104 qubits. The team also modeled how a quantum processor would execute fault-tolerant computation under realistic noise, so that researchers can evaluate and refine designs before hardware is actually built. The study additionally validated a technical approach that the report says could cut the number of physical qubits required for the related scheme by about 90%, reducing the resource overhead of quantum error correction.

NVIDIA CEO Jensen Huang said that understanding how complex fault-tolerant systems behave at large scale is key to making practical quantum computing a reality, and that GPU supercomputers can help researchers tackle problems that were previously out of reach. The report also frames the work as part of NVIDIA's broader positioning in quantum computing: beyond supplying GPU infrastructure for AI and high-performance computing, the company is promoting collaboration between GPU supercomputers and quantum processors, using classical systems to simulate, validate and optimize future quantum hardware. Public information currently stops at the announcement of this simulation and the validated approach; the report gives no timeline for follow-up hardware or commercial deployment.