There is currently a tremendous amount of interest in quantum computing, which can potentially solve some of the world’s biggest problems. However, despite the feverish efforts of cloud providers, hardware manufacturers, and software industry, quantum computing is still on the drawing board.
It will happen someday. But, for now, we’ll have to settle for quantum simulations. But those seem challenging, too.
Today graphics processing unit leader Nvidia Corp. announced a partnership with PennyLane to enable quantum simulations. Before the announcement, I had the opportunity to talk to Timothy Costa, director of high-performance computing and quantum at Nvidia, as the company was prepping today’s announcement of its connection of its cuQuantum with Xanadu’s PennyLane. In this newly developed environment, quantum simulations can run on high-performance clusters of Nvidia GPUs.
Costa said he’s looking forward to the days when the hardware is ready to run. “But we want to make sure that when those machines arrive, they’re able to be used productively so that the software is ready, the algorithms are ready, and different institutions know how to use such a platform productively as it arrives,” he said. “That comes down to simulating very large quantum systems for that algorithm.”
Doing that at the scale required for productive quantum computing requires a multinode, multi-GPU supercomputing scale simulation. Costa says this is a significant challenge to solve. “Programming individual processors is hard enough,” he says. “Doing it well across many nodes and many GPUs is a hard problem.”
So, what Nvidia is announcing today is the integration of new multinode, multi-GPU application programming interfaces from the company’s cuQuantum software development kit into the industry-leading PennyLane quantum programming framework. With this new development, a researcher using PennyLane will be able to run a GPU supercomputer, with few changes, at the scale of future of quantum computing.
Costa says academic institutions such as Stony Brook and Brookhaven and large companies like Rolls Royce and Volkswagen already utilize this connection between cuQuantum and PennyLane.
If you develop a jet engine simulation on current computer gear, you won’t be able to take that code and use it in a quantum environment. Maybe the most impressive feature of Nvidia’s development is that whatever you run in the quantum simulation will be a simple lift-and-shift to quantum computers as they become available. In fact, Rolls Royce has developed that jet engine simulation in this environment.
When will quantum computing be available in the real world? Estimates vary wildly. So that makes Nvidia’s work here seem like a bit of a gamble. What if the quantum computer we get in five or ten years is markedly different from the one we’re simulating today? Costa says not to worry.
“These frameworks are simulating high-level, gate-based architectures,” he said. “Whether we’re talking about superconducting qubits, neutral atoms, photonics, or all the different modalities of qubits, they all fit into this same framework of how you would program such a system.”
Aside from jet engine simulations, Costa tells me that applications range from energy and drug discovery to optimization and logistics. He said a significant percentage of Global 100 companies have launched quantum research. “We do think some areas are more likely to bear fruit in the near term,” he said. “Most are those in which simulation of chemical systems is the core is at the core.”
Historically, Nvidia has been on the forefront of democratizing technology. With the sophistication and expense of quantum, how will the company keep that heritage? In time, using CUDA Quantum, Costa sees the democratization of quantum happening. For now, running Nvidia’s cuQuantum with Xanadu’s PennyLane can happen on-prem or in a number of cloud environments, such as Google Cloud Platform, Amazon Web Services, Microsoft Azure or the company’s own DGX Cloud.
In a collision of buzzwords, Costa told me that AI and quantum are complementary. “There’s two angles to this,” he said. “How does quantum help AI and make AI more productive? And how does AI help quantum and make quantum more productive?”
He said Nvidia had been more focused on the latter. “PennyLane is the leading framework for quantum machine learning,” he said. “It’s designed to help people leverage quantum computers in ML and AI. And it’s an unanswered problem at this point whether quantum is going to be an accelerant to AI or not. But this is a framework in which people can explore and find answers to that question.”
Costa said he’s starting to see customers couple the two in a workflow, for example, where a large language model plays a significant role in a workflow and quantum computing steps in later.
Nvidia is well-known for doing audacious things. In this case, it’s audacious yet a bit amorphous, too. Where is it all heading? When will quantum computing materialize? Will Nvidia’s approach be right, or will a competitor steal the glory? Good questions. Only time will tell.
But this approach plays well into Nvidia’s tendency to take a full-stack approach. Given the lofty heights Nvidia sits at today, I am constantly hearing industry chatter of other of how other silicon manufacturers are set to take a bite out of Nvidia.
What makes the company unique is the full-stack approach, where they bring together all the components necessary to deliver a solution, and this often includes third parties, sometimes competitors. In many ways, the more complex the problem, the more it requires something turnkey. Quantum is amongst the hardest to solve, which should tip things Nvidia’s way and act as a catalyst for growth for years to come.