Convergence of High Performance Computing, AI and Big Data Analytics in the Exascale Era
We’re in the midst of a never-ending need for larger, more flexible high performance computing (HPC) systems to help science further its rapid pace of discovery. In many fields, answering deeper questions requires higher-fidelity simulations and AI training that typically gobble up computational resources at scale. Researchers and data scientists have a great opportunity to make major strides with the Department of Energy’s Aurora supercomputer to be located at the Argonne National Laboratory (ANL). Its exascale-level scalability and speed will help the scientific research community answer more challenging questions faster than ever before. In the coming months and years, we will see previously unattainable progress in science and exciting innovations.
Advancing science
Exascale computing isn’t just about theoretical science — it has many practical applications for the betterment of society. It can help researchers identify new materials for improved batteries, generate a new and sustainable type of solar cell, or find excellent catalysts to speed chemical reactions. In molecular research, Aurora will also help the pharmaceutical industry identify new and effective medicines. The combination of predictive analytics and computational chemistry has the potential to reveal which drug molecules are chemically and physically viable. From there, researchers can use that experimental data to identify the few best drug candidates for more comprehensive testing in the laboratory. We’ll also see engineers combining computational fluid dynamics and machine learning to construct databases that model things like industrial processes or simulate the airflow over an airplane wing for greater fuel efficiency. Exascale-level simulation and modeling also help find new materials for use in a variety of other applications, too. Research groups can then use that information to create the material in a lab and demonstrate its practical applications.
Historically, these tasks proved exceptionally difficult — if not impossible — since the various workloads can require hundreds of thousands, or even millions, of calculations. With the aid of today’s machine learning methods, HPC systems can comb through the output and mine it for the most useful information. As the examples above illustrate, with the aid of Aurora’s prowess for automated workflows, scientists can tackle more challenging scenarios and bring innovative ideas to life much faster and at a lower cost.
As the scientific community starts using Aurora in the near future, they will undoubtedly use all its compute capability, and in the process, uncover new questions that expand their work. With exascale computing it’s difficult to predict all the problems they can solve until they start running experimental workloads on it and see what we can do. The process of experimentation opens many new doors for scientific discovery. The sky’s the limit.
Making access to exascale easier for researchers
There are many in the development community who look to Intel and ANL for help with the challenges associated with programming for an accelerated, exascale topology like Aurora’s. With all of Aurora’s technological advancements in the exascale system, the teams at Argonne and Intel are committed to making system usage as easy as possible.
The ANL team is already working hard to educate the scientific community about Aurora’s hardware and software stack. Aurora is projected to exceed two exaFLOPS of double precision compute performance with high-speed networking, Distributed Asynchronous Object Store (DAOS) for fast I/O, new Intel Xeon processors codenamed Sapphire Rapids with high bandwidth memory, and new Intel Data Center GPUs codenamed Ponte Vecchio.
Argonne also seeks to ways help members of the developer community prepare their applications for accelerated programming models. With modern tools, it’s now easier to take large, legacy-based applications and software stacks and move them to Aurora’s GPU accelerator model which will offer a fantastic performance benefit.
There are diverse ways in which developers can code for the GPU using various programming models like Intel oneAPI DPC++, OpenMP, and Kokkos. The framework and accelerated math libraries create many more opportunities for developers to make use of exascale architectures. Plus, since Aurora’s nodes will provide a unified memory architecture it will be easier for researchers to move existing workloads to the Aurora system. In other words, the Argonne team wants researchers to worry less about migrating memory between CPUs and GPUs and focus instead on getting their advanced science done.
Developers who have used other HPC systems can make the most of their coding expertise to gain the enormous benefit GPU acceleration provides. When they log in to use Aurora, the supercomputing interface itself will remain familiar, but users will experience a lot more compute power than they could access previously. With a bit of coding modification, many legacy applications can achieve up to a hundred-fold speed increase when they run on a system like Aurora. That means scientists can cut the time needed to get the answers they need, saving them hours, weeks, or potentially even months.
Christopher Knight leads the chemistry and materials science group at the ANL facility and manages a team of computational scientists that work with users to accelerate their research efforts. At a recent talk at an American Chemical Society event in Chicago, Knight discussed how to prepare code for exascale. Reflecting on the event, Knight described the audience’s enthusiasm. “Six people came up after the presentation and asked how they could get time on Aurora. It’s inspiring to have conversations like these because there are so many creative and innovative ideas out there, and we’ll have the compute power to make those visions a reality.”
There’s no secret process for getting access to Aurora once the machine is online and available. ANL invites anyone to submit their proposal to the Argonne Early Science program. The team endeavored to simplify the proposal process and has already received many proposals from researchers queuing up for their time on the system once it goes online.
When a proposal is green-lighted and assigned time on Aurora, those researchers will have the full support of Argonne facility staff. Even if an individual is not an expert in parallel computing, it’s no problem. Knight noted, “We have software engineering resources devoted to helping with things like compiling an application with oneAPI or assisting in evaluating unexpected results.”
Find out more
If you’re interested in tapping Aurora’s potential, check out the Aurora web page to learn more about its architecture. There you’ll find resources like ANL’s early adopter webinars and many stories about teams preparing for their time on the system. The oneAPI site offers another great resource for learning more about heterogeneous parallel programming approaches that can make the most of an exascale system.
Added Knight, “We’re thrilled to see such excitement for the new system. As we get closer and closer to launch, we’re very eager to see real science performed on the system. We anticipate the supercomputer will have a profound impact and we’ll undoubtedly see major scientific breakthroughs with its assistance. We look forward to seeing everyone’s proposals for time on Aurora!”
Listen to a podcast interview with Christopher Knight or read more about the Aurora exascale system.
Rob Johnson spent much of his professional career consulting for a Fortune 25 technology company. Currently, Rob owns Fine Tuning, LLC, a strategic marketing and communications consulting company based in Portland, Oregon. As a technology, audio, and gadget enthusiast his entire life, Rob also writes for TONEAudio Magazine, reviewing high-end home audio equipment.
# # #
This article was produced as part of Intel’s editorial program, with the goal of highlighting cutting-edge science, research and innovation driven by the HPC and AI communities through advanced technology. The publisher of the content has final editing rights and determines what articles are published.
