Alice & Bob Decodes LDPC-Cat Codes on GPUs With NVIDIA NVQLink 

Insider Link

• Alice & Bob demonstrated GPU decoding of its LDPC-cat error-correction codes using NVIDIA NVQLink and synthetic data, a step toward scalable error correction for cat-qubit quantum computers.
• The proof of concept used Quantum Machines’ control hardware and showed improved decoding efficiency when processing multiple sets of error measurements in parallel.
• The demonstration did not achieve real-time decoding, and Alice & Bob plans to optimize data transfer and processing to keep pace with its quantum computers.

PRESS RELEASE — Alice & Bob, a leader in fault-tolerant quantum computing, today shared results from implementing GPUs to decode its LDPC-cat architecture using NVIDIA NVQLink. The proof of concept using synthetic data is the first time LDPC-cat codes have been decoded on GPUs via the low-latency, high throughput NVQLink architecture, and moves the company a step closer to fast, scalable error correction for its cat-qubit quantum computers. 

NVQLink is NVIDIA‘s open system architecture for tightly coupling the performance of GPU supercomputing with quantum processors (QPUs) to build accelerated quantum supercomputers. NVQLink specifically supports tasks that require fast communication between quantum and classical hardware, such as error-correction decoding, calibration, and hybrid quantum-classical applications across any qubit modality. 

Alice & Bob assessed the feasibility of GPU decoding for LDPC-cat codes using an NVQLink connection with Quantum Machines’ OPX1000 and Open Acceleration Stack, a modular framework that couples the controller directly to GPUs and other classical processors with microsecond-scale roundtrip latency. In particular, researchers assessed whether measurement data could be transferred from the control electronics to the GPU. LDPC-cat codes were developed by Alice & Bob for its cat-qubit architecture and first introduced in the paper “LDPC-cat codes for low-overhead quantum computing in 2D”  published in January 2025.  

Though a proof of concept and not constituting real-time decoding, the results show how decoding can scale by becoming more efficient as several syndromes are decoded in parallel. 

Alice & Bob will now optimize the pipeline, the chain of steps that carry the measurement data through NVQLink to the GPU, to enable the performance needed for real-time decoding. 

“Error correction is what will turn quantum computers into useful machines, and decoding is one of its most demanding classical tasks. Showing that our LDPC-cat codes can be decoded on NVIDIA GPUs with existing tools is encouraging for scaling, because it means decoding capacity can grow with the machine. The next step is making it fast enough to keep pace with the quantum computer, and we’re working with NVIDIA to get there,” said Juliette Peyronnet, US GM Alice & Bob.  

“A major challenge with present day quantum computing is errors build up faster than they can be corrected, and traditionally it has been a challenge to bring state of the art compute to this problem for faster qubit modalities like cat qubits,” said Sam Stanwyck, Director, Quantum Product, NVIDIA. “Alice & Bob’s work shows how NVQLink can in fact accelerate the real-time decoding needed to turn cat qubits into fault tolerant quantum-GPU supercomputers.” 

This collaboration builds on the existing relationship between the two companies. In March 2026, the companies announced a runtime of 1 hour 57 minutes for the decoding of simulated syndrome data of its Elevator Codes with the NVIDIA CUDA-Q platform. In June 2025, the companies announced integration of NVIDIA CUDA-Q into Dynamiqs, Alice & Bob’s QPU simulation library, enabling GPU-accelerated quantum simulations.