ItInsightsWorld General: How JUPITER Helped NVIDIA Validate a 104-qubit Fault-Tolerant Quantum Model

For a 104-qubit NVIDIA study of magic state cultivation, JUPITER became the computational stage. The machine is the first exascale supercomputer in Europe, and the method is described as using 90% fewer qubits than previous fault-tolerance approaches.

The work ran on a system of NVIDIA GH200 superchips at Jülich Supercomputing Centre and is presented as the largest comprehensive simulation of a fault-tolerant quantum processor to date, because it models realistic hardware noise at record scale.

Why Magic States Matter Before Hardware Exists

Qubits operate at the limits of physics, and the interactions needed to control them introduce unavoidable, error-inducing noise. Fault-tolerant quantum computing therefore depends on magic states: a specific arrangement of qubits continually consumed during computation.

Because machines must produce those states repeatedly, any optimization in magic-state cultivation sharply lowers the difficulty of building a useful quantum computer. NVIDIA's simulation under realistic noise validated a state-of-the-art approach and gave researchers tools for developing and testing future improvements.

The earlier Jülich work simulated 50 qubits under ideal, noiseless conditions. This new model instead focuses on magic-state cultivation under realistic hardware noise, helping researchers understand the impact of errors before a processor is built. It also supports predictions for useful quantum applications in fields like drug discovery, financial modeling and materials science, while reducing reliance on hardware trial and error.

From Four GPUs to Exascale Tests

Simulating this process can evaluate and improve designs before building quantum hardware, but the memory required can grow exponentially with the number of qubits, putting large-scale runs beyond conventional systems. For the JUPITER project, an NVIDIA GB200 NVL72 system at the NVIDIA Accelerated Quantum Research Center (NVAQC) served as the development platform.

That platform enabled researchers to scale tests from four GPUs to hundreds before moving to 8,192 GPUs on JUPITER. The center accelerated development and testing of the simulation methods, let teams compare results with JUPITER, and showed how accuracy scaled with additional computing resources. It helped reduce a weekslong simulation effort to days while preparing the team to use JUPITER’s exascale computing capacity effectively.

Software Layers Behind the Run

The simulations used NVIDIA cuQuantum software for highly optimized performance and scale. They also used the NVIDIA CUDA-Q platform’s fault-tolerant logical layer to express the magic-state circuit.

Stage	Configuration	Purpose
Early testing	four GPUs	start the scaling path
Center validation	NVIDIA GB200 NVL72 at NVAQC	compare methods and results before exascale use
JUPITER run	8,192 GPUs on JUPITER	model a 104-qubit fault-tolerant processor under realistic noise

The final point, for this site, is practical: larger quantum devices will make simulation harder, not easier, so pre-validating methods on accelerated clusters such as those at NVAQC matters before world-class supercomputers carry the heaviest calculations.