No technology evolution gets talked about more at Nvidia’s GTC events than that of AI, and the latest Spring GTC conference was no different, but take a closer look and you might notice that Nvidia is not so quietly ramping up its efforts to support the emerging field of quantum computing.
Truly valuable quantum computers may be “solidly a decade away” or more, Nvidia co-founder, president and CEO Jensen Huang said this week. He isn’t the only one saying that, as quantum computing systems with sufficient error correction to effectively attack big computation tasks in a reliable way remain a work in progress. Current quantum computers, and even those to arrive in the coming years, most likely will be connected to classical computers in some way, creating hybrid classical-quantum computing solutions to tackle new problems and advance the science of quantum computing to become more commercially viable.
In this hybrid relationship, quantum computers have much to gain from connecting via low-latency links to high-performance GPUs, which can help accelerate error correction and improve the fidelity of quantum processors. At GTC this week, Nvidia took steps in that direction with partner-involved announcements of a "GPU-accelerated quantum computing architecture" and new software to support these hybrid quantum efforts.
In the first case, Nvidia unveiled the DGX Quantum system, which pairs the company’s Grace Hopper GPU and its CUDA Quantum programming software with the quantum orchestration and system control platform of Israel-based partner Quantum Machines, with low-latency PCIe being used to establish a link between Nvidia’s GPU and the quantum processing unit of another vendor (Nvidia previously has partnered with several QPU developers).
“When you're trying to build a [quantum] computer, you have to use the most revolutionary computer at the current time to create some truth to know whether the quantum computer is generating the right answers,” Huang said, adding that classical GPUs also are ideal for conducting necessary research into new quantum algorithms before they can be used by quantum computers. He further stated, “The quantum computer will likely be connected to a classical computer and that classical computer will likely be an accelerated computing system [for the quantum computer].”
Regarding the second announcement, Nvidia announced this week that its CUDA Quantum programming model is now open-source, a move timed with a related announcement from Nvidia partner Quantum Brilliance, and Australian quantum computing firm. Quantum Brilliance released a new version of its own open-source Qristal quantum compiler software that it said is designed to work with Nvidia’s Cuda Quantum. Essentially, this means that the Qristal software is able to compile quantum programs written in CUDA Quantum, and that those programs can be run on Nvidia’s GPUs, as well as other processors
Pat Scott, Software Lead at Quantum Brilliance, stated, "By providing the capability in Qristal to compile quantum programs written in CUDA Quantum, we have made it possible to build quantum software that runs seamlessly across NVIDIA graphics processing units (GPUs), central processing units (CPUs) and quantum processing units (QPUs). Incorporating CUDA Quantum into Qristal also means users can run large-scale supercomputer simulations of future hybrid quantum-classical computers that simultaneously exploit quantum processors, classical CPUs and GPUs."
Nvidia also announced at GTC this week that even more quantum ecosystem companies plan to support CUDA Quantum. These include quantum hardware companies such as Anyon Systems, Atom Computing, IonQ, ORCA Computing, Oxford Quantum Circuits, and QuEra; quantum software companies Agnostiq and QMware; and even three groups that host supercomputing centers–the U.S.-based National Center for Supercomputing Applications, Japan’s National Institute of Advanced Industrial Science and Technology, Finland’s IT Center for Science.