NVIDIA added a new orchestration layer called CUDA-Q Logical to its open source CUDA-Q quantum computing platform, giving hardware teams and national labs a common way to design and test fault-tolerant quantum systems before any of that hardware is built.
CUDA-Q Logical does not run on a quantum computer itself. It is a modeling and orchestration tool: researchers plug in different combinations of algorithms, error-correction codes and qubit hardware, then compare how many physical qubits and how much runtime each combination would need. A single change to any one of those pieces can shift the resource math for an entire application, and teams have historically rebuilt custom infrastructure each time they wanted to test a new combination.
Fermilab is an early adopter of NVIDIA's CUDA-Q Logical. Chief technology officer Anna Grassellino, who directs the lab's Superconducting Quantum Materials and Systems Center, said the tool let her team explore different error-correction and hardware combinations in about three weeks, versus roughly five months of building custom infrastructure before, a change NVIDIA calls a sevenfold speedup.
NVIDIA also pointed to modeling work from Iceberg Quantum, which used CUDA-Q Logical to project how many physical qubits Diraq's silicon-based qubit architecture would need to reach 1,000 logical qubits. The simulation put that figure at about 150,000 physical qubits, roughly a tenth of Diraq's earlier estimate, though the number describes a projected architecture rather than qubits that have actually been built or tested.
Sandia National Laboratories built a separate benchmark, called QUOPS, to track how quantum hardware is progressing toward fault tolerance rather than counting physical qubits alone. A reference implementation now ships with CUDA-Q, and Sandia posted early QUOPS results, covering hardware from Google, IBM and Quantinuum, in a preprint ahead of this month's IEEE Quantum Week. None of those systems are fault-tolerant machines; QUOPS measures how far each is from that goal, which is also how Timothy Proctor, who co-directs Sandia's Quantum Performance Laboratory, described the lab's aim: to track and forecast progress industry-wide rather than certify readiness today.
Beyond Fermilab and Sandia, NVIDIA said hardware and software groups including Infleqtion, IQM Quantum Computers and Quantum Motion are already using CUDA-Q Logical for their own architecture work, while separate NVIDIA tools, NVQLink for linking quantum processors to GPU supercomputers and the NVIDIA Ising model family, are seeing similar early adoption from companies such as Anyon Computing, Quandela and Diraq. CUDA-Q Logical and the QUOPS reference implementation are both available now in NVIDIA's open source CUDA-Q repositories.
