More than a dozen leading scientific supercomputing centers globally are adopting NVIDIA’s NVQLink, a universal interconnect designed to integrate quantum processors with accelerated computing platforms. This technology connects quantum processors to NVIDIA GPUs, enabling large-scale quantum-classical workflows powered by the CUDA-Q platform, and delivers a GPU-QPU throughput of 400 Gb/s with latency under four microseconds. Institutions including Japan’s AIST and RIKEN, Korea’s KISTI, and CINECA in Italy are joining U.S. facilities in advancing the integration of quantum and classical hardware, overcoming control and error-correction challenges to unlock new scientific discoveries.
NVQLink: Connecting Quantum and Accelerated Computing
NVIDIA’s NVQLink is a new, universal interconnect designed to link quantum processors with accelerated computing systems. This technology is being adopted by over a dozen leading supercomputing centers globally, including facilities in Asia, Europe, and the U.S., to advance quantum computing research. NVQLink aims to unite the strengths of quantum and GPU computing, enabling a future where supercomputers combine quantum simulation capabilities with the programmability and massive parallelism of GPUs.
NVQLink delivers significant performance metrics crucial for hybrid quantum-classical workflows. It achieves 40 petaflops of AI performance using FP4 precision, coupled with a GPU-QPU throughput of 400 Gb/s and a latency of less than four microseconds. Quantinuum successfully used NVQLink with its Helios quantum processor to demonstrate real-time scalable decoding for quantum error correction, exceeding performance requirements by 32x with a reaction time of 67 microseconds.
The open system architecture of NVQLink facilitates integration with existing quantum control systems and NVIDIA’s CUDA-Q software platform. This allows for flexible, configurable decoders capable of massive parallelism, and accessible real-time application programming interfaces. Researchers can easily scale classical compute resources alongside expanding quantum processors and applications using NVQLink’s Ethernet-based connectivity.
Adoption by Leading Supercomputing Centers
More than a dozen supercomputing centers globally are adopting NVIDIA’s NVQLink to integrate quantum processors with accelerated computing. Facilities across Asia – including those in Japan, Korea, Taiwan, Singapore, and Australia – and Europe & the Middle East are joining U.S. national laboratories in this effort. NVQLink aims to advance research by linking quantum and classical hardware, enabling large-scale quantum-classical workflows powered by the CUDA-Q platform. This widespread adoption signals a significant move toward hybrid quantum-GPU systems.
NVQLink delivers substantial performance metrics crucial for quantum computing. It provides 40 petaflops of AI performance at FP4 precision, coupled with a GPU-QPU throughput of 400 Gb/s and a latency of less than four microseconds. Quantinuum’s Helios quantum processor utilizes NVQLink to demonstrate real-time decoding for quantum error correction, achieving a reaction time of 67 microseconds – 32x faster than the two-millisecond requirement. This highlights NVQLink’s ability to support complex quantum computations.
This interconnect’s architecture allows for flexible and configurable decoders with massive parallelism, crucial for tackling quantum error correction. Developers can access these capabilities through real-time APIs within the NVIDIA CUDA-Q software platform. The use of Ethernet also allows researchers to scale classical compute resources as quantum processors and applications expand. NVQLink is now available for quantum builders and supercomputing centers interested in access.
In the future, supercomputers will be quantum-GPU systems – combining the unique strengths of each: the quantum computer’s ability to simulate nature and the GPU’s programmability and massive parallelism.
Jensen Huang, founder and CEO of NVIDIA
NVQLink Performance and Technical Specifications
NVIDIA’s NVQLink is a universal interconnect designed to link quantum processors with accelerated computing systems. It delivers substantial performance metrics, including 40 petaflops of AI performance at FP4 precision, a GPU-QPU throughput of 400 Gb/s, and latency of less than four microseconds. This high-speed connection is crucial for hybrid quantum-classical applications, allowing tight integration with quantum control systems and NVIDIA’s CUDA-Q software platform, and enabling scalable real-time decoding for quantum error correction.
NVQLink has been adopted by over a dozen supercomputing centers globally, including facilities in the U.S., Asia, and Europe. These centers are utilizing NVQLink to advance research and development in quantum computing. Quantinuum’s Helios quantum processor successfully integrated with NVIDIA GPUs via NVQLink, demonstrating the first scalable, real-time decoding for qLDPC quantum error correction codes with a decoder reaction time of 67 microseconds—exceeding a two-millisecond requirement by 32x.
The architecture of NVQLink is designed for flexibility and scalability. Utilizing Ethernet, researchers can readily expand classical compute resources as quantum processors and applications grow. Access to NVQLink is available for quantum builders and supercomputing centers, with developers benefiting from real-time application programming interfaces within the NVIDIA CUDA-Q environment. This allows for easy building and testing of approaches to quantum error correction and quantum-GPU applications.
Real-World Applications of Quantum Error Correction
NVIDIA’s NVQLink is enabling real-world applications of quantum error correction by tightly integrating quantum processors with NVIDIA GPUs. Quantinuum’s Helios QPU successfully deployed scalable, real-time decoding for qLDPC codes using NVQLink and the CUDA-Q platform. This integration protects quantum information from noise, a major challenge in quantum computing, and achieved a reaction time of 67 microseconds – exceeding a two-millisecond requirement by 32x.
NVQLink delivers the performance necessary for advanced quantum error correction, offering 400 Gb/s throughput and latency under four microseconds. This is made accessible to developers through real-time application programming interfaces within NVIDIA CUDA-Q, simplifying the building and testing of quantum-GPU applications. The system’s use of Ethernet also enables researchers to easily scale classical compute resources alongside expanding quantum processors.
More than a dozen supercomputing centers globally—including facilities in the U.S., Asia, Europe, and the Middle East—are adopting NVQLink to advance quantum computing research. These include national labs like Brookhaven and RIKEN, as well as centers like CINECA in Italy and the Pawsey Supercomputing Research Centre in Australia, all seeking to harness the power of integrated quantum-classical workflows.
