Quantum computing continues to attract an array of technology companies, from computing giants like IBM to pure-play start-ups like IonQ, and many hardware and software firms that don’t even begin with the letter “I”. Among those, Nvidia has been dabbling in the space, but things are getting way more serious.
With all of the other news coming out of the Nvidia GTC Spring event this week, it may have escaped attention that the company also made several quantum-related announcements that more clearly define the role it’s playing in the quantum computing sector today, and how that role might evolve in the future. As with the company broader evolution to becoming a full-stack computing force, software is playing a central role.
A common language
Among these moves, Nvidia announced a new quantum compiler, nvq++, supporting the Quantum Intermediate Representation (QIR) programming specification for a low-level machine language that quantum and classical computers can use to talk to each other. Compilers turn high-level instructions into assembly language for programming. Supporting QIR means Nvidia is backing the notion that computing in general is evolving to a stage during which the high-performance computing (HPC) systems that Nvidia is increasingly fortifying with GPUs and CPUs will work closely with emerging quantum computers, and that this hybrid environment will require common language.
Researchers at Oak Ridge National Laboratory (ORNL), Quantinuum, Quantum Circuits Inc., and others have embraced the QIR Alliance standard, and ORNL will be among the first to use the new software, Nvidia said.
Timothy Costa, Group Product Manager, HPC & Quantum Computing at Nvidia, told Fierce Electronics, “We think quantum will be a co-processor-style accelerator to HPC and scientific computing, so we want to make sure that folks on the Nvidia platform have a highly productive way, a seamless way, to program those systems as advantage starts to arrive,” Costa said. “We want to enable algorithm developers today as they look at this opportunity, but when we do arrive into the future of quantum advantage, and we have quantum processors sitting next to GPU [graphic processing units, which Nvidia designs] accelerated supercomputers, trying to work together, we want to make sure that it’s a seamless experience.”
The aim is to deliver that easier experience not to an HPC or quantum expert, but to a domain scientist or developer in an industry that needs highly advanced computing to solve extremely difficult problems. “We're talking about a GPU accelerated supercomputer and a quantum processor working together so that's a very complex infrastructure, and so the ability of the developer in a very productive way to target it is going to be critical for seeing any advantage,” Costa said.
A new appliance
Until this week, Nvidia’s biggest move in the quantum space was last year’s unveiling of cuQuantum, a software programming model and library aimed at doing for quantum processing what Nvidia’s CUDA platform does for GPUs and parallel processing. Costa said cuQuantum, which already has been adopted by several quantum computing firms, is generally available.
But even more intriguing was a product unveiling this week that brings another dimension to cuQuantum, and again demonstrates how Nvidia is embracing the notion that a hybrid classical-quantum computing era is already underway. Nvidia unveiled the beta release of its cuQuantum DGX Appliance, which Costa described as a software-based container with all the components needed to run cuQuantum jobs–mainly the quantum simulations increasingly being used in industries like finance and pharmaceuticals–optimized for Nvidia’s well-established DGX A100 systems.
Before this week, Nvidia had not played up the DGX A100’s abilities for quantum simulation, but the company’s website notes that “an Nvidia DGX A100 with eight Nvidia A100 80GB Tensor Core GPUs can simulate up to 36 qubits, delivering an orders-of-magnitude speed-up over a dual-socket CPU server on leading state vector simulations.”
Costa said the DGX A100’s GPUs fill a current need for scientists doing algorithmic research and researchers building quantum processors that are still maturing. “The progress that's been made in quantum computing hardware recently is very incredible, but we also know there's a long way to go before quantum processors can provide advantage over classical computing for relevant problems. When it happens it'll be transformational acceleration across a wide range of industries. And that's also what the GPU has been for so many industries already.”
The cuQuantum DGX Appliance is now available in beta.
An ecosystem expansion
Also this week, Nvidia added new names to its list of quantum partners: Classiq, Xanadu and Zapata Computing each are now using cuQuantum, joining a roster of partners that already includes Google Quantum AI, IBM, IonQ and Pasqal.
Classiq supports cuQuantum in its Classiq Quantum Algorithm Design platform, and Zapata is supporting it in its Orquestra platform, while Xanaduhas integrated cuQuantum into PennyLane, its open-source framework for quantum machine learning and quantum chemistry.
"Allowing users to gain HPC performance with cuQuantum through PennyLane opens many doors for our research and development community,” said Lee O'Riordan, Performance Lead for PennyLane at Xanadu. “By using our plugin and the Nvidia library, users can easily move from testing and developing hybrid quantum workloads on their local machines to running them on the fastest computing platforms. This allows users to get the best time-to-solution with a seamless transition, enabling the exploration of research questions that may otherwise have been too complex to examine locally."
All these moves show that Nvidia is moving into quantum computing as aggressively as any firm from the classical computing realm. Unlike IBM, Nvidia isn’t making its own quantum processing units (or at least it hasn’t said so yet), but it is doing its part to enable the QPU evolution.
“If you're a quantum processor company, you're building a QPU, and we want to make sure that we're providing a platform where you can simulate larger designs to push the state of the art forward and then validate those designs. As we know, today's modern [quantum] processors are noisy [error-prone]. So making sure you're building the right thing and you're getting the right results is critical. And we need to have great simulation infrastructure in place to do that.”
While many factors–system noisiness, expensive deployment, ecosystem fragmentation, a shortage of quantum computing talent, and more–could still slow down the quantum evolution, Nvidia doesn’t shy away from identifying itself as a member of the quantum pack.
As Costa said, “I would definitely say Nvidia should be thought of as a quantum computing company.”
RELATED: Nvidia is changing itself into a full-stack computing force