Setting the stage for breakthrough discoveries with the nation’s first exascale system, the Argonne Leadership Computing Facility’s (ALCF) Early Science Program (ESP) is offering an open call for proposals to scientists and researchers seeking to use Argonne’s most advanced HPC system – Aurora – anticipated in 2021. According to Argonne, Aurora will be the nation’s first system exceeding an exaflop, or a billion-billion calculations per second. With likely over five petabytes of memory, Aurora will deliver the scientific community a level of computing performance and processing power not before possible. According to Top500.org’s November 2017 list of the world’s most powerful computers, the top system today delivers performance of 93.01 petaflops, a fraction of Aurora’s target.
To prepare for the Aurora ESP, and assist scientists with the completion of proposals, the ALCF has opened registration for their free Simulation, Data and Learning Workshop scheduled for February 27-March 1, 2018. Held at Argonne, the workshop seeks to familiarize users with the ALCF’s existing Theta system, paving the way for use of the Aurora flagship upon deployment.
Theta HPC System
Argonne’s current Theta system, while not capable of the exascale performance levels Aurora will attain in the future, is instrumental in delivering breakthroughs in data science and machine learning. Theta resides among the most influential HPC systems available for open science today, offering performance levels in the 11-petaflop range. Combining Theta and other existing ALCF systems, over 5 billion hours of computing time are committed each year to advanced scientific endeavors.
Theta’s hardware components occupy a massive set of 24 racks, together hosting the HPC system’s 4,392 nodes. Each node depends on a 64-core Intel processor supported by 16 GB of MCDRAM, 192 GB of DDR4 RAM, and a 128 GB solid state drive. The combined 281,088 cores are maximized with Cray interconnect technology, a Lustre-based 10 PB parallel file system, Cray’s HPC software stack, and more. The combination of these technologies enables Theta’s maximum speed of 11.69 petaflops.
Applications for the ALCF Early Science Program
After delivery in 2021, Aurora will see little rest during its tenure. The projects in the ESP will have dedicated pre-production access to the system before it is available to the broader user community. The Aurora ESP, which kicked off with ten projects in 2017, is expanding to align with a new paradigm for scientific computing at the ALCF—the convergence of simulation, data, and learning approaches.
The 2018 call for proposals will bring in 10 additional projects in the areas of data science including data analytics, data-intensive computing, and advanced statistical analysis. Other endeavors in that group will focus on machine learning elements including deep learning, and neural networks. Proposals for cross-cutting projects that involve simulation, data, and learning are also encouraged. While projects will represent a wide range of scientific disciplines, the team at Argonne anticipates significant benefit for fields like materials science, biology, transportation and renewable energy. From an energy-related standpoint, Aurora can help scientists design new classes of materials for more efficient solar panels, batteries, and wind turbines. Other research projects in advanced biology can address complex challenges like seeking organisms which deliver better biofuels or identifying vulnerabilities to help control diseases. Additionally, advanced HPC systems will reveal insights that lead to more aerodynamic vehicles and more efficient engines.
ALCF Simulation, Data, and Learning Workshop
Users wishing to apply for participation in the Aurora ESP have an opportunity to get a leg up by attending the upcoming three-day Simulation, Data, and Learning (SDL) Workshop at Argonne’s facilities in Lemont, Illinois, February 27-March 1, 2018. There, an extensive agenda offers attendees immersion and insight into the Theta system experience, a more in-depth technical dive into workload optimization, and support for developing Aurora ESP proposals to enable cutting-edge scientific endeavors.
Attendees can also tune their applications and develop proposals for awards through the ALCF Data Science Program, ASCR Leadership Computing Challenge (ALCC), or the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program. In the past, these programs enabled pioneering research projects leading to insights for simulation of brain tissue, hybrid photovoltaic materials, and greater understanding of atomic particles.
Timothy Williams, Ph.D., ALCF’s Deputy Director of Science and Program Manager for the Aurora Early Science Program, reinforces the workshop’s critical benefits for attendees. “Aurora will be the first exascale system available for the most advanced scientific endeavors across the three ‘pillars’ of simulation, data, and learning. Emerging usage patterns of HPC systems for computational science reflect today’s convergence of simulation, data, and learning-based methods. SDL workshop sessions will focus on some of the complex computing challenges in simulation like compiling, profiling, and optimizing codes on the Theta architecture. The SDL Workshop will also cover using workflows, containers, deep learning frameworks, and tools relevant to data and learning applications.”
The intensive workshop agenda also offers attendees the opportunity for dedicated hands-on time with industry professionals and ALCF staff experts, and the ability to make reservations for time on the Theta system, providing access for additional computational testing and debugging. Other sessions offer a deep-dive into Python use and scale, utilizing math libraries on Intel’s 2nd generation Intel Xeon Phi processors, debugging, and profiling.
Those interested in the workshop and getting a head start for Aurora’s launch should register today! For additional information, be sure to check out the Aurora ESP webpage and the Proposal Instructions webpage.
# # #
This article was produced as part of Intel’s HPC editorial program, with the goal of highlighting cutting-edge science, research and innovation driven by the HPC community through advanced technology. The publisher of the content has final editing rights and determines what articles are published.