Seeing the Quantum Horizon: A Panel with Google's HPC Team

Admin

— 3 min read
Share
Seeing the Quantum Horizon: A Panel with Google's HPC Team

It's been a busy couple of months, so I'm finally getting around to posting about a great panel I was part of back in August. I joined my old friend whurley of Strangeworks and Bob Sorensen of Hyperion Research for a panel hosted by the Google Cloud Advanced Computing Community. Jay Boisseau led us through a conversation about quantum hardware, hybrid computing, and how to tell whether any of it is becoming useful.

The full title is Seeing the Quantum Horizon: Hype, Hopes, Hybrid Systems, and the Road to Quantum Advantage and Value. The recording runs about 51 minutes.

One distinction I made during the discussion is between the progress of the technology and the progress of the industry. As a scientific instrument, a quantum computer gives us a way to explore quantum systems that become intractable to simulate classically. The scale and quality of those simulations are meaningful measures of progress. But a customer evaluating a quantum service has another set of questions: what can it do for my problem, what does it cost, and how does it compare with the methods I already use?

Bob and whurley kept bringing us back to that customer perspective. A modest improvement can be worth a great deal when applied repeatedly across a large business. It doesn't have to settle a question in computational complexity theory to be worth paying for. Scientific milestones and commercial usefulness both matter, but they answer different questions.

We also spent some time unpacking what people mean by hybrid quantum-classical computing. Bob drew a useful distinction between sending a task to a quantum processor and getting an answer back, and an iterative computation in which the classical and quantum systems have to stay closely synchronized. Those imply different architectures. Using both kinds of compute in a workflow doesn't, by itself, establish a need to put them in the same rack or even the same room.

That's where I remain skeptical of some of the sales pitches for on-premise quantum computers. The asynchronous use cases emerging look promising to me, but I don't see them establishing a requirement for tightly coupled hardware. And classical processing for quantum error correction is part of making the quantum computer work. As Bob pointed out, that's overhead from the customer's perspective. I expect it to become part of what happens inside the box.

The benchmarking discussion raised a related problem. Bob warned about choosing a benchmark that sends the whole sector off optimizing for something users don't need. I pointed to the Unitary Foundation's work on community-driven benchmarking as a way to keep the measurements connected to useful questions. Even something as apparently concrete as a logical qubit needs a definition: what operations can it support, at what quality, and with what practical performance?

For developers wondering how to get involved, my recommendation was open source. The Unitary Foundation's projects offer entry points for people with classical software skills who haven't studied quantum physics. I shared a story from the podcast about a contributor who started by fixing bugs in an error mitigation framework and went on to become a lead author on a research paper. Contributing can be a way to learn the field while doing useful work in it.

There are disagreements and some familiar jokes along the way. If you'd like to jump into a particular part of the discussion:

My final contribution was a qualification to the economic argument: we need to account for externalities, too. What we spend and what we get can leave out energy use, environmental impacts, and social costs. That's work for technology as a whole, including quantum computing.

Thanks to Jay and the Google HPC team for hosting, and to Bob and whurley for the conversation.

Read more