GPUs Transformed AI. Now They’re Here For Quantum.

GPUs Transformed AI. Now They’re Here For Quantum.


While quantum hardware remains immature, companies say they found another way to put complex quantum algorithms to work: running them on the same chips used for powering artificial intelligence.

This process, known as simulation, has in recent years received a boost from the growing scale of computing power that graphics-processing units and other advanced chips offer.

“Nobody thought this was possible,” said Jack Hidary, CEO of quantum software company Sandbox AQ, which spun off from Google in 2022. “We don’t have to wait for a quantum computer. We’re not using a quantum computer, but we’re using quantum equations, quantum software on GPUs. And that’s a big breakthrough.”

GPUs are specialized chips designed to support the heavy load of training and running AI algorithms. Their key role in supporting generative AI propelled GPU maker Nvidia to a trillion dollar valuation earlier this year, although other companies, including Amazon and Google, also make specialized AI chips.

Quantum algorithms are well suited to GPUs thanks to their ability to handle dense math and high bandwidth memory, among other things, said Nvidia’s Director of HPC & Quantum Computing Timothy Costa. “It’s a workload which is a great fit for GPUs for the same reasons that AI is a great fit for GPUs,” he said.

Quantum algorithms have fundamentally different approaches to problem-solving than classical algorithms, but work for certain use cases, including simulating the behavior of natural materials, such as molecules, and optimization problems.

It is possible to run some of these algorithms on the small-scale quantum computers that exist today, which run on quantum processing units, or QPUs—but the technology is still in its early stages and the error rate of these machines remains high.

In the last couple of years there has been an explosion of adoption of GPUs for quantum simulation use cases, said Costa.

“Corporates have played around with pure play quantum algorithms on pure play native quantum chips,” said Markus Pflitsch, founder, chairman and CEO of quantum tech provider Terra Quantum. “Now, they are really interested in: how do I enhance business performance with this stuff?”

Simulation can accelerate research and sometimes, return results better than traditional computing setups can, Hidary said.

Sandbox AQ this month said it was working with battery materials and technology company NOVONIX to use quantum simulation to model the behavior of ions, the charged particles in lithium ion batteries. NOVONIX, which is based in Brisbane, Australia, said the collaboration would allow them to develop new machine-learning models which can accurately predict lithium-ion cell lifetime.

Using that knowledge, the ultimate goal, Hidary said, is to envision and realize new battery chemistry—a much needed innovation since electric cars are pushing lithium-ion battery supply to its limits.

Simulations have limitations in terms of the complexity of algorithms they can run, and future fault-tolerant quantum computers could add better performance and scalability than simulation, Pflitsch said.

When quantum hardware matures, simulation could have an existential crisis, said Heather West, research manager and quantum-computing research lead at IDC, although the question over whether simulation will truly become obsolete or continue to provide some value remains unresolved.

Nvidia’s Costa says GPUs could still have an important role to play on future quantum machines, including doing error correction.

“GPU computing will remain essential for many reasons throughout the lifetime of quantum,” Costa said.

Write to Isabelle Bousquette at isabelle.bousquette@wsj.com

Next Story

Leave a Reply

Your email address will not be published. Required fields are marked *