NVIDIA simulates 104-qubit system on JUPITER supercomputer
The Jülich Supercomputing Centre hosted the run, which used magic state cultivation to cut the physical qubits needed for error correction by 90%.
NVIDIA researchers ran a 104-qubit quantum simulation on the JUPITER exascale supercomputer at Germany's Jülich Supercomputing Centre, according to digitaltoday.co.kr. The run modelled realistic noise conditions rather than an idealised environment.
Using a technique called magic state cultivation, the team reported needing 90% fewer physical qubits for error correction than conventional approaches. The work was implemented with NVIDIA's cuQuantum and CUDA-Q software platforms.
The simulation ran on NVIDIA GH200 processors and scaled to as many as 8,192 GPUs. The researchers say simulations that previously took weeks were reduced to days.
First tested on GB200 NVL72
The technique was first tested at an NVIDIA accelerated quantum research centre on a GB200 NVL72 system, starting with 4 GPUs and scaling to several hundred.
The resource savings have only been shown in simulation and still require verification on real quantum processors. Jülich previously ran a 50-qubit simulation under ideal conditions.
Quick answers
What did NVIDIA simulate on the JUPITER supercomputer?
A 104-qubit quantum system under realistic noise conditions, using magic state cultivation for error correction.
How many GPUs did the simulation use?
The work used NVIDIA GH200 processors and scaled to as many as 8,192 GPUs.
Have the savings been confirmed on real quantum hardware?
No. The 90% reduction in physical qubits has only been shown in simulation and still requires verification on real quantum processors.