Nvidia researchers have utilized the JUPITER supercomputer, located at the Jülich Supercomputing Centre in Germany, to perform a 104-qubit fault-tolerant quantum computing simulation. This achievement marks the largest simulation of a fault-tolerant quantum processor to date, allowing for the study of reliable computation under conditions mimicking real hardware noise.
The core of the study involved testing the magic state distillation method. Nvidia claims that this approach reduces the requirement for physical qubits by 90% compared to traditional methods. This development aims to lower the hardware resource barriers for building practical quantum computers, which could eventually be applied to drug discovery, materials science, and financial modeling.
The simulation was executed using 8,192 Nvidia GH200 GPUs on the JUPITER system, supported by the cuQuantum software and the CUDA-Q platform. The research team initially optimized the process on a GB200 NVL72 system before scaling it to JUPITER, which reduced the simulation time from several weeks to just a few days.
It is important to note that this result is a simulation performed on a classical supercomputer rather than the creation of a physical 104-qubit fault-tolerant quantum computer. The project provides a framework for researchers to evaluate noise impacts and refine design schemes before committing to the manufacturing of quantum hardware.
