TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
NVIDIA says it used 8,192 GPUs on Europe’s JUPITER supercomputer to simulate a 104-qubit fault-tolerant quantum processor under realistic hardware noise. The work tested a magic-state cultivation approach that NVIDIA says could require 90% fewer qubits than previous methods; the simulation is not a demonstration of a working quantum computer.
NVIDIA said it used 8,192 GPUs on the JUPITER supercomputer at Germany’s Jülich Supercomputing Centre to simulate a 104-qubit fault-tolerant quantum processor under realistic hardware noise. The company says the calculation let researchers examine a method of producing magic states that may use 90% fewer qubits than previous approaches; it was a computer simulation, not a run on a physical quantum processor.
The simulation modeled magic-state cultivation, a method for preparing quantum states needed to carry out fault-tolerant calculations. According to NVIDIA, the team used realistic noise assumptions to study how errors affect the process, rather than modeling an ideal, noiseless system. The company described the work as the largest comprehensive simulation of a fault-tolerant quantum processor to date, a characterization that has not been independently verified in the source material.
NVIDIA said the computation ran on JUPITER, which it identifies as Europe’s first exascale supercomputer, using NVIDIA GH200 superchips. Researchers used cuQuantum software and the fault-tolerant logical layer of NVIDIA’s CUDA-Q platform to represent and run the simulated circuit. The work was developed and tested before the full-scale run using an NVIDIA GB200 NVL72 system at the company’s Accelerated Quantum Research Center.
The development process, NVIDIA said, scaled tests from four GPUs to hundreds before the team moved to JUPITER’s 8,192 GPUs. The company said this preparation helped cut a simulation that would have taken weeks down to days. The source report does not give the exact runtime, detailed noise parameters, or independent measurements of the simulation’s accuracy.
Testing Quantum Designs Before Construction
Large simulations can help researchers compare proposed quantum processor designs before committing to costly hardware builds. In this case, modeling noise matters because errors are a central obstacle to reliable quantum computation: a design that works in ideal conditions may behave differently when realistic imperfections are included. The work offers a way to investigate that gap using GPU computing alongside quantum-computing research.
Magic states are consumed during fault-tolerant computation, so the resources required to produce them affect the size and complexity of a future machine. NVIDIA’s claimed 90% reduction in qubit requirements, if borne out and applicable to practical designs, could point to a more resource-efficient approach. The announcement does not establish that useful quantum applications are now available, or that the simulated method has been demonstrated in a built processor.
The project also illustrates a growing role for classical supercomputers in quantum research. Exascale resources can model systems that are difficult to assess experimentally, but simulations remain dependent on their assumptions, algorithms, and available computing capacity. Their value is in informing design and testing—not replacing physical hardware validation.
NVIDIA GPU high performance computing
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Ideal Models to Noisy Simulations
The Jülich work follows an earlier effort that, according to NVIDIA’s announcement, simulated 50 qubits under ideal, noiseless conditions. The new project instead modeled magic-state cultivation with realistic hardware noise, expanding the kind of behavior researchers could study. The comparison is between different simulation conditions as well as different system sizes.
Fault-tolerant quantum computers use error-correction methods to perform computations despite noise in qubits and their control. Magic states are a specialized resource used in these schemes, and producing them repeatedly can demand substantial hardware. The JUPITER project focused on simulating this resource-production step; it did not report construction or operation of a 104-qubit physical quantum computer.
NVIDIA said its development system allowed the researchers to test methods and examine how results scaled before using JUPITER. That staging helped prepare the software and computation for a large run, while the supercomputer supplied the resources for the reported simulation.
“Useful quantum computing will require fantastically complex, fault-tolerant systems — and understanding how they behave at scale is essential to building them.”
— Jensen Huang, NVIDIA founder and CEO
quantum computing simulation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits of the Reported Results
The announcement does not provide a peer-reviewed paper, independent replication, full technical results, or a detailed account of how the 90% qubit reduction was calculated and compared with earlier methods. The figure should therefore be treated as NVIDIA’s claim about the approach, not as an independently established result in the material available.
It is also unclear how closely the noise model matches the conditions of a future physical processor, how the simulation’s accuracy was checked, and whether the result can be extended to larger circuits or other fault-tolerant designs. No physical processor performing the simulated computation was reported. The source does not specify when complete methods or research data will be made public.
As an affiliate, we earn on qualifying purchases.
Further Validation and Hardware Tests
The next steps will depend on researchers publishing enough methodological detail for other teams to assess the simulation and test its findings. Further work could compare the approach with alternative magic-state methods, examine additional noise conditions, and evaluate how resource estimates change as circuit size grows.
Any practical claim about performance or reduced hardware needs will ultimately require continued testing against physical quantum hardware and updated error models. NVIDIA and Jülich have not, in the source material, announced a specific date for a paper, follow-up simulation, or hardware demonstration.
As an affiliate, we earn on qualifying purchases.
Key Questions
Did NVIDIA run a 104-qubit quantum computer?
No. NVIDIA reported a computer simulation of a 104-qubit fault-tolerant processor. The source does not describe a physical quantum processor running the calculation.
What did the simulation study?
It modeled magic-state cultivation, a process used in fault-tolerant quantum computing, under realistic hardware-noise assumptions.
What does the 90% figure refer to?
NVIDIA says the magic-state cultivation approach studied could require 90% fewer qubits than previous methods. The source does not provide independent verification or the detailed comparison behind that figure.
Where and how was the calculation performed?
The simulation ran on JUPITER at the Jülich Supercomputing Centre in Germany, using 8,192 GPUs and NVIDIA’s cuQuantum and CUDA-Q software, according to the company.
Does this mean useful quantum applications are ready?
No. The announcement describes research intended to help evaluate future designs. It does not report a quantum computer performing useful applications such as drug discovery or financial modeling.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
