Last week, Nvidia announced that NVQLink - its new quantum interconnect architecture - had been adopted by some of the world’s leading supercomputing canters and national research labs.

The NVQLink interface is an open interconnect architecture designed specifically for hybrid quantum-classical computing. It provides a very high bandwidth, low latency interface while tightly integrating QPUs and GPUs together under the control of Nvidia's CUDA-Quantum software stack. This tight integration is crucial for real-time quantum error correction, which is essential to build fault-tolerant quantum processors.

Hybrid Quantum-Classical Computing

Quantum processors produce and ingest enormous amounts of data, but they cannot process that data themselves. With fully fault-tolerant quantum computers years away, all quantum computing in the near and medium term will be based on hybrid systems, where classical computers carry out pre- and post-processing and quantum processors are used as accelerators. At present, this hybrid architecture is the only path to scalable, fault-tolerant quantum computing (Figure 1).

Bandwidth and latency to classical compute are major bottlenecks in quantum computing, particularly for error correction. NVQLink enables a very high throughput of 400Gb/s, which allows a lot of data to be streamed between GPUs and quantum processors and very fast, sub 4 micro-second latencies, approaching speeds where classical compute can react within quantum coherence times. NVQLink also leverages standardized RoCE[1] Ethernet networking technology. RoCE has very low latency and low overheads and its use eliminates the need for proprietary alternatives.

NVQLink uses the CUDA Quantum software, a bespoke software platform for hybrid quantum-classical computing. It allows developers to write code in C++ as part of a GPU-hosted application, supporting error correction and enabling new hybrid algorithms.

Major Ecosystem Partners

Nvidia announced that more than 24 of the world’s leading national scientific computing centers are adopting the NVQLink interface, including Lawrence Berkeley, Los Alamos, Oak Ridge and MIT Lincoln labs in the US; Japan’s Riken, Korea’s KIST and Taiwan’s NCHQ in Asia, as well as major European labs such as France’s GENCI, Germany’s JSC and the NQCC in the UK.

This has major implications for scientific computing. Solving many of the biggest challenges in science will need quantum and classical compute. For national labs and for industry, this announcement signals that infrastructure is being built, which helps de-risk investments into quantum workflows as there is now a path to integrate them with existing GPU infrastructure.

Figure 1: Nvidia’s Hybrid Quantum-Classical Computing Concept

Analyst Viewpoint

This is a major strategic announcement by Nvidia and shifts the narrative from fully fault-tolerant quantum computers to quantum-GPU based hybrid systems, essentially saying that the first scientific and industrial breakthroughs will be made by hybrid quantum-classical systems, where massive classical GPU compute is tightly integrated with quantum computing:

Competitive Differentiation – this enables Nvidia to stay relevant as quantum computing matures. By building the interconnect plus software layer (CUDA-Quantum) for quantum and GPU hybrid systems, Nvidia leverages its core strengths (i.e. GPU compute, high-performance interconnects) and positions itself in the quantum ecosystem without needing to build the qubits.
Market Catalyst - Nvidia’s quantum moves could accelerate the development of practical quantum systems, particularly hybrid systems. Nvidia’s tools such as CUDA-Quantum and cuQuantum are enabling quantum research in academia and in industry. If successful, Nvidia could become a critical infrastructure provider for quantum computing in the same way as it is for AI today. This could drive increased investment into quantum research, particularly around hybrid applications, calibration and error correction.
Ecosystem Adoption - adoption by many prestigious supercomputing centres and national research laboratories provides Nvidia with credibility in the quantum infrastructure space. For partners, seeing global research institutions commit to NVQLink helps reduce risk. If these ecosystem partnerships gain real traction, Nvidia could become the de facto interface layer between classical supercomputing and quantum processors, thus broadening Nvidia's TAM to include quantum infrastructure.
Competitors - alternatives to NVQLink have been developed by Google, IBM as well as start-ups such as Quantum Machines, IonQ and Rigetti. However, none of these have exactly what NVQLink offers, which is: the combination of a high-throughput, low-latency, standardized interconnect, tight integration with GPU compute and a broad ecosystem based on the CUDA-Quantum platform.

In most cases, users will select their preferred interconnect designs based on numerous factors, including level of latency required, how much classical compute is needed, decoders being run, hardware architecture, etc. However, the existence of multiple incompatible interconnects will inevitably lead to fragmentation, which will slow interoperability.

Even though many QPU developers and controller vendors – including Quantinuum, Rigetti and IonQ - are adopting NVQLink, the biggest players with strong vertical strategies, such as IBM and Google, will most likely continue with their own technologies and strategies.

Counterpoint Research has also published a Research Brief “Nvidia’s NVQLink Adopted by Top Global Supercomputer Centers” which provides further details and analysis and is available for subscribers of Counterpoint’s AI360 service.

[1] Remote Direct Memory Access (RDMA) over Converged Ethernet

Author