NVIDIA has announced that JUPITER, Europe’s first exascale supercomputer, has completed a record-breaking quantum computing simulation, allowing researchers to explore and validate the cultivation of magical states.
The 104-qubit simulation, described by NVIDIA as the largest comprehensive simulation of a fault-tolerant quantum computing system to date, ran on NVIDIA GH200 GPUs at the Jülich Supercomputing Center in Germany.
Magic states are essential for performing complex calculations on fault-tolerant quantum computers. However, creating them with sufficiently low error rates remains a major challenge.
Traditional methods require significant hardware resources. Some estimates suggest that up to 95 percent of a quantum computer’s physical qubits could be used to create these states, leaving relatively little for calculations.
NVIDIA’s announcement comes days after Google reported that its Willow quantum processor achieved a 40x error reduction in preparing specialized quantum states.
Unlike NVIDIA’s work, which uses classical supercomputing hardware to simulate quantum processes, Google’s experiment involved preparing magical states on an actual quantum processor. The two approaches address different challenges in developing fault-tolerant quantum computers. Simulations help researchers evaluate designs before testing them on physical hardware.
Simulation of real quantum errors
Last year, JUPITER simulated a complete 50-qubit universal quantum computer under ideal, noiseless conditions, surpassing the previous 48-qubit record set by Jülich researchers in 2019 on the Japanese K computer.
However, the latest 104-qubit simulation takes realistic hardware noise into account, allowing researchers to study how errors affect the cultivation of the magical state.
In contrast to ideal simulations, which assume that quantum operations can be carried out without errors, the new work models the disturbances that physical quantum processors are subject to during calculations. This provides a more realistic way to assess whether proposed methods can work reliably under practical conditions.
By simulating the process under these conditions, NVIDIA researchers validated an advanced approach to cultivating the magical state, providing a way to test future improvements. The results could also help identify vulnerabilities in proposed designs and guide researchers to more efficient methods before committing resources to building quantum hardware.
Scaling to 8,192 GPUs
For the new project, NVIDIA researchers initially developed and tested their methods on a GB200 NVL72 system at the company’s Accelerated Quantum Research Center.
They scaled the tests from four GPUs to hundreds before moving to 8,192 GPUs on JUPITER. According to NVIDIA, the development work helped reduce the simulation effort from weeks to days.
The team used NVIDIA’s cuQuantum software to optimize simulation performance and its CUDA-Q platform to model the fault-tolerant quantum circuit.
The detailed simulation could allow researchers to predict how future quantum computer designs might work before physical hardware is built. This could reduce the need for repeated hardware testing and help accelerate the development of quantum applications in drug discovery, financial modeling and materials science.
“Useful quantum computing requires incredibly complex, fault-tolerant systems – and understanding how they behave at scale is important to building,” said Jensen Huang, CEO of NVIDIA. “Our results from Jülich show how quantum processors and GPU supercomputers can work together to tackle research that would otherwise be unattainable, thereby accelerating the discovery of transformative quantum applications.”
