Guest Blog: The Convergence of HPC, Artificial Intelligence and Quantum

Author: Per Nyberg, Head of Integrated Quantum Computing and HPC, Quantum Strategy Institute.

The application of high-performance computing (HPC) to use-cases of scientific, industrial, and societal importance is ubiquitous. The ability to simulate the earth’s climate processes, develop new materials at the molecular level or understand fine grained relationships in vast amounts of data would not be possible without HPC.

Since the advent of the first mainframes and supercomputers, these capabilities have at any given time been at the forefront of both technology and computational approaches to solving the most complex problems. The two advance together, leading to the development of novel technologies and increasingly sophisticated applications.

These capabilities are applied by organizations around the world every second of every day on systems located in data centers, supercomputing centers, and the cloud.

Traditionally HPC architectures were designed for simulation-based approaches such as computational fluid dynamics. Conversely, applications were designed to capitalize on the underlying technology that was available to software developers. Today’s HPC systems are composed of multiple processing, memory, communication and storage technologies. This heterogeneity reflects the varied characteristics of applications which have evolved to leverage techniques such as machine learning. This convergence of techniques and technologies has enabled the solution of problems not previously possible and the formulation of entirely new applications.

In the late 2000s the concept of convergence first began to formulate as new “big data” solutions developed. Until that time equation-based simulation on HPC was the primary approach to tackling complex problems. Big data provided a data-driven approach to finding insights in the growing volumes of simulation, observational and digital data.  The first aspect of convergence was that big data and simulation offered complementary approaches to problems that could leverage a common HPC platform. The second aspect was that this common platform would be composed of multiple technologies that could be applied to different applications depending on their needs.

Convergence with AI

The next major development was with artificial intelligence which has made significant advances over the past decade. The availability of rich datasets combined with the advancement of machine learning techniques and HPC computational power have made possible the application of AI to a wide range of areas.

In the late 2010s there was the so-called Cambrian Explosion of specialized processors such as AI specific accelerators for inference and training. As a result, today’s applications can be composed of several approaches, such as simulation and machine learning, leveraging many different technologies such as x86 processors, GPUs, and AI accelerators. Examples can be found across any industry, such as whole vehicle simulation in automotive and fraud detection in financial services.

Weather prediction and climate modeling are often cited in the HPC community as the prototypical applications with an insatiable thirst for computing performance. These are highly complex simulations spanning a broad range of physical and temporal scales. Since the first practical applications in the 1970s, they have tracked and pushed every new HPC technology. Today they can leverage 100,000s of processor and accelerator cores in models which combine simulation and machine learning.

This capability is essential to addressing the grand challenges in science, industry, and society.

The next step is Quantum

The next major advancement is expected to be with quantum computing. We can expect a similar converged trajectory with quantum computing and HPC is already playing a key role in the deployment of the first significant quantum capabilities. While quantum computing has special functional characteristics unlike any other technology, the effective integration of quantum algorithms alongside simulation and machine learning will be essential to its broad adoption.

Ultimately, the benefit of convergence is to bring the best and most applicable techniques and technologies to bear on a use-case. While this places challenges on engineers and developers, this capability is essential to addressing the grand challenges in science, industry, and society.