NVIDIA Expands CUDA-Q With New Platform for Fault-Tolerant Quantum Computing
NVIDIA is expanding its open-source CUDA-Q quantum computing platform with a new orchestration layer designed to help researchers build and test applications for future fault-tolerant quantum computers.
Called CUDA-Q Logical, the new layer gives researchers a programmable way to model and coordinate the algorithms, error-correction methods, hardware architectures and other components required to operate quantum computers using logical qubits.
NVIDIA says the technology is already producing major development-time gains. Researchers at Fermilab used CUDA-Q Logical to reduce a fault-tolerant architecture development workflow from about five months to three weeks — a reported 7× speedup.
What Is NVIDIA CUDA-Q Logical?
Fault-tolerant quantum computing aims to overcome the errors that naturally occur in physical qubits.
Today’s quantum processors are highly sensitive to noise and other sources of error. Building useful systems therefore requires techniques that combine many physical qubits into more reliable logical qubits, along with sophisticated error-correction and control systems.
The challenge is that changing one part of a quantum system can affect the requirements of everything else.
An algorithm, for example, may require different resources depending on the selected error-correction method, quantum processor architecture or other hardware characteristics.
NVIDIA says CUDA-Q Logical is designed to address that codesign problem by allowing researchers to model and orchestrate these components within a single programmable workflow.
That gives researchers a way to compare different configurations and determine which combinations could offer the best performance for systems built around logical qubits.
Fermilab Reports 7× Faster Development
One of the most significant early results comes from Fermi National Accelerator Laboratory.
According to NVIDIA, Fermilab researchers used CUDA-Q Logical to evaluate physical qubits, runtimes and resource requirements across different error-correction approaches and quantum hardware configurations.
The researchers converted complex fault-tolerant system designs into a repeatable computational workflow.
NVIDIA says the approach reduced the time required for fault-tolerant algorithm development from approximately five months to three weeks.
That represents a reported 7× acceleration in the development process.
Fermilab CTO Anna Grassellino said fault-tolerant quantum computing will require researchers to design algorithms, error correction, architectures and hardware together rather than treating each component independently.
The result is important because quantum computing development can involve substantial iteration between software and hardware. Faster modeling could allow researchers to test more possible configurations before committing to physical implementations.
Iceberg Quantum Finds a Potential Hardware Reduction
CUDA-Q Logical is also being used to examine how many physical qubits could be required to build large-scale logical quantum systems.
NVIDIA said Iceberg Quantum used the platform to model a fault-tolerant architecture based on Diraq’s qubits.
The modeling indicated that 1,000 logical qubits could potentially be created using about 150,000 physical qubits, roughly 10 times fewer than Iceberg Quantum’s previous estimate.
The result is a modeling outcome rather than a demonstration of a completed fault-tolerant quantum computer. Its significance is that researchers can use software-based architecture exploration to rapidly assess possible implementations before building the corresponding hardware.
That type of analysis could become increasingly important as quantum companies attempt to scale from today’s noisy processors toward systems capable of executing useful, error-corrected workloads.
Sandia’s QUOPS Benchmark Comes to CUDA-Q
NVIDIA is also adding a new benchmarking capability to CUDA-Q through QUOPS, developed by Sandia National Laboratories.
QUOPS is intended to measure how prepared quantum computing hardware is for practical, fault-tolerant applications.
Historically, quantum computing progress has often been described using measurements such as physical-qubit counts, fidelity and coherence. Those metrics remain important, but they do not necessarily show how close a system is to running useful large-scale applications.
QUOPS takes a different approach by providing a hardware-agnostic method for tracking progress toward utility-scale quantum computing.
Sandia says the benchmark is designed to help quantum hardware developers and users measure and forecast improvements in quantum computing capabilities.
A QUOPS reference implementation is now available through NVIDIA CUDA-Q, according to NVIDIA.
Sandia’s quantum program also focuses heavily on modeling, error correction, benchmarking and quantum hardware development as researchers work toward fault-tolerant systems.
NVIDIA Is Building a Larger Quantum-GPU Ecosystem
CUDA-Q Logical is part of a broader NVIDIA strategy to connect quantum processors with GPU-accelerated computing.
NVIDIA’s NVQLink architecture, announced in 2025, is designed to tightly connect quantum processors with GPU supercomputers, giving researchers access to classical high-performance computing resources for quantum control, calibration and error correction.
The company says organizations across the quantum ecosystem are now using its technologies in areas including quantum error correction, quantum applications, quantum control and hybrid quantum-GPU systems.
Among the developments highlighted by NVIDIA are work involving Diraq, Anyon Computing, Quandela, Quantum Machines, IonQ, MITRE, Phasecraft, UCLA and Caltech.
The broader goal is to move beyond quantum processors operating in isolation and toward quantum-GPU supercomputers, where quantum processors and conventional accelerated computing work together.
Why Fault-Tolerant Quantum Computing Matters
Quantum computers have the potential to tackle certain problems that are extremely difficult for conventional computers, including some simulations involving chemistry, materials and complex physical systems.
But useful quantum computing requires more than simply increasing the number of physical qubits.
Large-scale applications will need reliable logical qubits capable of performing calculations while correcting or suppressing errors. That creates a major engineering challenge because logical qubits can require many physical qubits and sophisticated error-correction systems.
Tools such as CUDA-Q Logical are therefore aimed at one of the field’s central problems: determining how algorithms, error correction and hardware can be designed together efficiently.
For researchers, being able to test these configurations computationally could reduce the amount of time spent building specialized infrastructure for each experiment.
What Happens Next?
CUDA-Q Logical is now available through GitHub, while NVIDIA says the QUOPS reference implementation is also available through CUDA-Q.
The next stage will be seeing how researchers and quantum hardware companies use the tools to evaluate increasingly realistic fault-tolerant architectures.
The technology does not mean practical, fault-tolerant quantum computers are already available. Instead, NVIDIA’s announcement points to a software-development shift: researchers are gaining more sophisticated tools to model the systems that could eventually make those machines practical.
If the reported development-time improvements can be reproduced across different quantum architectures, software-based codesign could become an important part of the race toward useful quantum computing.
