Time to put the HPC in QC
Quantum computing matters
Today, progress in virtually any field is paced—and limited by—progress in computational performance. Take a deep look at any industry or endeavor or exploration, and you will see a surprising dependency on computational resources. We are seeing, however, a very real stalling of necessary performance advances as Moore’s law flattens and power becomes a significant constraint and limiting factor.
Quantum computing offers the potential to address this impending crisis. But quantum computing is often approached as if it’s an industry in and of itself (and a researchy one at that), and this puts delivery and adoption of these technologies in a reasonable timeframe at risk. Quantum computing should instead be thought of as part of the broader High-Performance Computing ecosystem, and here are some key HPC concepts that can make quantum computing more relevant.
Quantum as accelerator
I remember sitting in a hotel ballroom about a decade ago with representatives from many of the world’s most significant HPC users. The (heated) debate that day concerned the potential for GPU-based computing. There was significant skepticism about whether GPUs could ever be a viable path for the next generation of HPC given the challenges, including whether there would be applications GPUs could address that would justify the paradigm shift required to employ them.
Ultimately the answer was that GPUs were best in a hybrid environment tightly integrated with traditional HPC and acting as “accelerators” on the applications (or portions of applications) for which they were best suited. The fact that GPUs could not show advantage across all applications was not fatal, instead it drove a hybrid approach. A decade later, GPUs have proved to be a game-changing HPC technology, with a significant portion of the world’s most powerful supercomputers utilizing GPUs as accelerators. If back then the promise was that in 5-10 years vendors might begin shipping yet-to-be-defined, all-GPU machines this story would have ended much differently for GPUs.
Customer engagement
With new HPC technologies or new applications of HPC technologies, it has often been critical to provide customers with early-stage product so their teams could explore the technologies and potential applications and use cases. We saw this with AI technologies in HPC. “Killer apps” can come from this deep customer engagement, rather than the engagement just coming from already-identified “killer apps.” This kind of engagement requires “productized” systems that customers can explore more deeply rather than a reliance on only high-latency access to research-oriented resources. Near- and medium-term “NISQ” machines that can be delivered commercially can provide this deeper level of customer engagement leading up to the delivery of error-corrected systems of scale.
The companies (and nation states) that begin investing in adopting these technologies, building internal proficiencies and exploring potential use cases that could benefit from quantum technologies well before they can demonstrate clear “advantage” are the ones that will establish defensible competitive differentiation. It is precisely because of the complexities and challenges of implementing quantum technologies that it will be critical to get a very early start.
You will be assimilated (and that’s ok)
Quantum computing has the potential to reinvigorate the HPC industry. The idea that quantum computing is an all-or-nothing quest for quantum supremacy in comparison to classical approaches is a flawed way to look at investment in and access to quantum technologies. Instead, we need to take a stepped approach to delivering these technologies. And the only way this will be possible is if we develop quantum computing technologies that can truly integrate (from a full system and stack perspective) with current HPC technologies and into existing data centres.
Author:
Mike Piraino, Chief Strategy & Development Officer at ORCA Computing
