Summary

NVIDIA has expanded its open-source CUDA-Q platform with CUDA-Q Logical, an orchestration layer for designing fault-tolerant quantum-computing systems. The platform also now includes a reference implementation of Sandia National Laboratories’ QUOPS benchmark.

NVIDIA said on September 14, 2026, that it has expanded its open-source CUDA-Q platform with CUDA-Q Logical, an orchestration layer for designing and testing applications intended for fault-tolerant quantum computers. The release also adds a reference implementation of QUOPS, a benchmark developed by Sandia National Laboratories to measure progress toward practical quantum-computing applications.

CUDA-Q Logical is now available through GitHub. NVIDIA says the platform is already being used by quantum-computing companies and research laboratories including Fermilab, Infleqtion, IQM Quantum Computers, QCDesign Quantum Motion and Sandia National Laboratories.

Contents

What CUDA-Q Logical changes

Quantum processors use physical qubits, which are vulnerable to noise and errors. Fault-tolerant quantum computing aims to manage those errors by encoding information in logical qubits and coordinating the physical hardware and error-correction processes needed to operate them reliably.

Designing such a system requires several elements to be developed together. A change to a quantum algorithm, error-correction code, hardware architecture or another quantum-processing-unit component can alter the number of physical qubits, runtime and other resources required by the application.

CUDA-Q Logical is designed to represent and coordinate these parts in one computational workflow. Researchers can switch between different algorithms, error-correction approaches and hardware configurations to evaluate how they affect a system built around logical qubits. NVIDIA describes the layer as programmable and verifiable, allowing researchers to explore integrated configurations rather than building separate specialised tools for each design.

The practical change is therefore in system design and evaluation. The announcement concerns a software platform for exploring fault-tolerant architectures, rather than the launch of a new quantum processor.

Early results and the QUOPS benchmark

Fermilab researchers used CUDA-Q Logical to validate earlier results and assess physical-qubit requirements, runtimes and other resources across different error-correction approaches and quantum hardware. Fermilab reports that its team reduced the time needed to develop and evaluate fault-tolerant algorithm designs from about five months to three weeks, which NVIDIA describes as a sevenfold speedup.

Iceberg Quantum used the platform to model a fault-tolerant architecture for Diraq’s qubits. The model indicated that 1,000 logical qubits could be created with 150,000 physical qubits—roughly 10 times fewer than Diraq’s previous estimates. This result is an architecture-level resource estimate produced through modelling.

QUOPS adds a separate way to assess progress. Developed by Sandia National Laboratories, the benchmark is intended to be hardware-agnostic and cross-platform, with a focus on the capabilities needed for utility-scale applications. Quantum-computing progress has often been described through physical-qubit counts, fidelity and coherence time. QUOPS is intended to track progress toward practical workloads across complete fault-tolerant systems.

Sandia has shared early QUOPS results for quantum-processing units from Google, IBM and Quantinuum in a preprint, and a QUOPS reference implementation is now available in CUDA-Q.

The examples in the announcement cover design and modelling stages: Fermilab’s result concerns the time required for a computational workflow, while Iceberg’s result concerns the estimated resources for a proposed architecture. Together, they show how an orchestration layer can be used to compare possible system configurations before a complete fault-tolerant machine is built.

A software layer for quantum-GPU systems

NVIDIA is positioning CUDA-Q Logical alongside tools intended to connect quantum processors with conventional accelerated computing. Its NVQLink system architecture is designed to tightly couple quantum processors with GPU supercomputers. NVIDIA says Anyon Computing has used it for a quantum-control system, Quandela for a QPU-GPU architecture and Quantum Machines for an integration demonstration at the Israeli Quantum Computing Center.

The company also says Diraq used NVIDIA Ising models to calibrate a silicon-based qubit processor. Other CUDA-Q integrations described in the announcement cover quantum error correction, error mitigation, quantum-application deployment and GPU-accelerated simulation.

Taken together, CUDA-Q Logical and QUOPS address two linked problems in quantum computing: how to design the large collection of components needed for fault-tolerant operation, and how to compare the resulting systems using measures closer to useful applications than raw physical-qubit counts alone.

Sources