Overview
- NVIDIA and the Jülich Supercomputing Centre ran a 104-qubit, high-fidelity simulation of magic-state cultivation on JUPITER Monday using 8,192 GH200 GPUs to model realistic hardware noise and fault-tolerant operation.
- Researchers developed and scaled the work first on an NVIDIA GB200 NVL72 system and then used cuQuantum, CUDA-Q, CPU offloading, memory compression, and network optimizations to cut memory and runtime needs for the exascale run.
- NVIDIA says the simulation validates a magic-state approach that can reduce the physical-qubit overhead for fault tolerance by roughly 90 percent compared with prior methods, easing a major barrier to building useful quantum machines.
- The result gives designers actionable data on how noise and error-correction trade-offs behave at scale, but it does not change the exponential classical cost of simulating more qubits and so depends on continued hardware and algorithm advances.
- The demonstration highlights JUPITER’s early use as Europe’s exascale testbed and signals that GPU‑accelerated system and software co-design will guide near-term quantum-hardware decisions while independent peer review is awaited.