NVIDIA expands open-source CUDA-Q with a layer for fault-tolerant quantum apps

NVIDIA expanded its open-source CUDA-Q platform with CUDA-Q Logical, a software layer for designing and testing applications meant for fault-tolerant quantum computers — machines that use logical qubits, which are error-checked groups of physical qubits, so they can run larger jobs.

Useful quantum computing still depends on beating the errors that single qubits make. NVIDIA put an open tool on GitHub so labs can try those error-checked designs without each shop building its own stack.

NVIDIA’s newsroom published “NVIDIA Expands Open Source CUDA-Q Platform for Fault-Tolerant Quantum Computing,” dated September 14, 2026, with a Toronto GlobeNewswire stamp. That company page is the filing event. CUDA-Q is NVIDIA’s open software for writing programs that mix quantum processors and ordinary graphics chips. CUDA-Q Logical is a new orchestration layer — software that lets researchers line up the pieces of a fault-tolerant design and swap them. Fault-tolerant here means the machine can keep going when individual qubits make errors. A logical qubit is a group of physical qubits plus error-checking, used as one more-reliable bit. This is a software drop. It is not a claim that a large fault-tolerant quantum computer is already running.

NVIDIA says researchers can now design and orchestrate the many parts a fault-tolerant application needs, then switch options to find better setups for logical qubits. Timothy Costa, NVIDIA’s vice president and general manager of quantum, said quantum computing is “maturing into an era of logical qubits,” and that researchers need an open platform that can represent a whole fault-tolerant system. He said CUDA-Q Logical lets teams explore integrated systems no matter the qubit type. File those as his words. This desk did not run the software.

NVIDIA says CUDA-Q Logical is already used by quantum-chip makers and labs including Fermi National Accelerator Laboratory, Infleqtion, IQM Quantum Computers, QCDesign Quantum Motion, and Sandia National Laboratories. A QPU is a quantum processing unit — the quantum chip. File the roster as NVIDIA’s. This desk did not audit those labs.

Iceberg Quantum, as NVIDIA tells it, modeled a fault-tolerant design for Diraq’s qubits and said 1,000 logical qubits can be made with 150,000 physical qubits — about ten times fewer than Diraq’s earlier estimates. Those are partner figures on NVIDIA’s page. This desk did not rerun the model.

Fermilab researchers used CUDA-Q Logical to check earlier results and compare physical qubits, runtimes, and other resource needs across error-correction methods and hardware. NVIDIA and Fermilab say that cut the work from five months to three weeks — a 7× speedup, meaning about seven times faster. Anna Grassellino, Fermilab’s chief technology officer and director of the Superconducting Quantum Materials and Systems Center, said the team explored those combinations in three weeks versus about five months of building specialized infrastructure. File the 7× and the quotes as NVIDIA and Fermilab. This desk did not time the work.

Sandia National Laboratories built QUOPS, an independent benchmark that asks how close a machine is to useful fault-tolerant work, without tying the score to one chip brand. NVIDIA says QUOPS is now in CUDA-Q. Sandia posted early results in a preprint before IEEE Quantum Week, with first numbers for Google, IBM, and Quantinuum quantum processors. Timothy Proctor, co-director of Sandia’s Quantum Performance Laboratory, said QUOPS is meant to track and forecast what quantum computers can do. File the benchmark and the preprint as Sandia via NVIDIA. This desk did not rerun QUOPS.

The same newsroom page also lists other quantum-GPU items — NVQLink hookups, Qedma and QCentroid error-correction work, Diraq using NVIDIA Ising to calibrate a silicon qubit chip. Those are same-day ecosystem notes on the same release. They are not this filing. Install pins and compiler details stay in Sources.

CUDA-Q Logical is on GitHub, NVIDIA says. A QUOPS code drop and two papers — one on CUDA-Q Logical, one on QUOPS — are linked from the newsroom page. This desk does not treat a download count or a package version as the news.

CONFIRMED here: NVIDIA’s September 14, 2026 newsroom page — PRIMARY company source. NOT claimed: that a large fault-tolerant quantum computer exists, that the Fermilab 7× or Iceberg ten-times-fewer figures are independently audited, a stock move, or investment advice. Distinct from the already-filed openai-gpt56-sol-quantum-experiments — an MIT lab using a coding agent on qubit calibration, not this NVIDIA platform drop.