Read part one of this conference recap here.
Industry Program Director Brendan McGinty welcomed guests to the second day of the annual National Center for Supercomputing Applications (NCSA) Industry Conference, October 8-10, on the University of Illinois campus in Urbana (UIUC). One hundred seventy from 40 organizations attended the invitation-only event.
The program opened with the NCSA director’s address; William Gropp’s talk was titled, “Challenges and Opportunities in the Next Generation of HPC Systems.”
Gropp became NCSA’s fifth director in June 2017, and has served as NCSA’s chief scientist since 2015. He is co-principal investigator (PI) on the Blue Waters supercomputer, and PI of the National Science Foundation’s Midwest Big Data Hub. Gropp holds the Thomas M. Siebel Chair in Computer Science (CS) and leads an active research program in the UIUC CS department. Prior to joining NCSA, Gropp held appointments at Yale University, Argonne National Laboratory, University of Chicago and the Institute for Advanced Computing Applications and Technologies. He holds a Ph.D. in CS from Stanford University.Gropp opened his talk with a chart from a 2004 Scientific American magazine that illustrated microprocessor components between 1993 and where conventional wisdom of the time thought they would be in 2019. The chart plotted clock speed (gigahertz), transistors (millions) and gate length (nanometers). The graph was illustrating Moore’s Law; trends that, at the time, everyone believed would continue.
Moore’s Law predicted the number of transistors on a microchip would double every two years and the cost would be halved. But the expected trajectory began to flatten in 2005, which challenged Moore’s Law, and marked-end of architectural stability.
“Moore’s Law will fade; features will keep getting smaller, but it will take longer and longer to achieve each reduction in size,” said Gropp. “There won’t be a point (at least not for the next 10 years) when Moore’s Law will be dead (at the end of any improvement), but as it gets harder and harder to achieve the same factors of reduction in size, engineers and scientists will need to innovate other ways to continue to improve performance,” he added.
“This is when new architectures were needed to increase performance; this is when GPUs, highly-parallel, simpler cores and specialized elements entered the scene,” said Gropp. “Everyone has been pushing toward extreme-scale architectures that are becoming more heterogeneous,” he added. For example, Gropp noted that China’s next-generation systems present a diversity of accelerator choices. The U.S. Department of Energy’s Sierra system features 4320 nodes of IBM POWER9 CPUs, with NVIDIA Volta Graphics Processing Units (GPUs). NCSA’s Deep Learning System has 16 nodes of IBM POWER9, with 4 NVIDIA Volta GPUs, plus FPGAs (Field Programmable Gate Arrays).
Gropp noted that it isn’t only processors and platforms that have changed. Since 2005, storage prices have fallen. A proliferation of sensors created a new way to look at the world, but they have also presented challenges with the amount of data they create, and our networks must facilitate larger data transfers.
But the biggest problem, according to Gropp, was the end of software stability. Fortran has been around for more than 40 years; C and C++ about 25 years. “New architectures and big-data demands have seen the birth of many new programming languages that don’t seem to last as long,” he said, and added, “Note the rise of Python; Perl was the Python of its day.”
While there are advantages to rapid innovation, the disadvantages include difficulty in finding where versions and components intersect. “I consider this a failure of software engineering,” said Gropp.
The end of Dennard scaling made our algorithms imperfect; they no longer exploit the performance that’s possible. Dennard once ensured the future was predictable, “but that free ride is over,” he added.
What is NCSA doing to stay ahead of this problem?
“We’re building teams of complementary expertise,” said Gropp. Anchor projects, such as the Center for Artificial Intelligence (AI) Innovation, Center for Digital Agriculture and the Large Synoptic Survey Telescope (LSST), have drawn specialists with a wealth of skills and strengths. NCSA has a new software directorate that employs more than 30 developers with expertise in many program languages. This workforce is especially attractive to NCSA Industry partners who can subcontract skills needed for special projects (contracts range from .25 to 6 FTE’s and for six months up to years in length).
NCSA Center for AI Innovation
During the Center for Digital Agriculture kick-off and first day of the NCSA Industry conference, it was evident that the 40 organizations represented, including NCSA Industry partners and tech companies who sponsored the event, are either active or soon to be active with AI; and just about every sector was represented. Therefore, few were surprised to learn that NCSA is launching a new Center for AI Innovation. Co-founders include NCSA Industry Director Brendan McGinty, Research Scientist Eliu Huerta, and Senior Research Scientist Volodymyr Kindratenko. Focused on the needs of industry, research and scholarship, “the new center will serve as a single umbrella for all things AI,” said McGinty. He also mentioned that a special announcement would be made during SC19.
Rising to the demands of the communities they serve; the NCSA Center for AI has been two years in the making. Initially forming with an investment from the National Science Foundation and Department of Energy (DoE) physics grants; each totaling $1.8 million, many more disciplines, domains and sectors are represented, including multi-messenger astrophysics, high-energy physics, medicine, agriculture, and financial services, to name a few. UIUC faculty affiliates from Agriculture, Electrical and Computing Engineering, Physics, the Beckman Institute, and Carle Clinic will foster an interdisciplinary scholastic AI experience, and they’re partnering with other universities. The advisory board has representation from national laboratories and industry.
The center’s education and outreach initiative is led by Kindratenko, who has found through experience with NCSA’s Students Pushing Innovation program, or “SPIN,” that many undergraduate scholars who enter the program are already acquainted with machine and deep learning.
To test and strengthen their skills, an AI hackathon was sponsored by NVIDIA and co-organized by the Center for AI, Gravity Group, Innovative Systems Lab and NCSA Industry immediately prior to the Industry Conference. To prepare for the hackathon, students completed IBM tutorials that acquainted them with the POWER9 platform; workshops continue every Wednesday from 3-5:00 p.m. at NCSA. “Two days, Three problems, Five teams, Twenty participants, and one dream,” said Huerta.
AI Hackathon winners are encouraged to apply for internships with companies at UIUC’s Research Park in the spring where they can help tackle real-world problems for research and industry partners. “This hands-on experience at the undergraduate level is critically important if we hope to answer the call for an AI-savvy digital workforce,” said Kindratenko.
The balance of day two included a session on emerging technologies chaired by NCSA Technical Program Manager Dan LaPine. Representatives from four tech giants presented case studies where AI methodologies support research:
- Thomas Henson (Dell Office of the CTO, Data Engineering Advocate) presented, “Scaling AI Initiatives from POC to Large Scale Production Deployments.”
- Mark Fernandez (HPE, Americas HPC Tech Officer & SpaceBorne Computer Payload Developer) talked about the first HPC system to be deployed into outer space: “The Dawn of HPC and AI Above the Clouds.”
- Roger Goff (DDN BDM & Senior Solutions Engineer): “Making the Best User of Flash for HPC with Lustre.”
- Tom Gibbs (NVIDIA Director of Developer Relations) presented a talk titled, “The Convergence of HPC and AI for Grand Challenge Science Problems.”
Photos by Leake and NCSA.
About the Author
HPCwire Contributing Editor Elizabeth Leake is a consultant, correspondent and advocate who serves the global high performance computing (HPC) and data science industries. In 2012, she founded STEM-Trek, a global, grassroots nonprofit organization that supports workforce development opportunities for science, technology, engineering and mathematics (STEM) scholars from underserved regions and underrepresented groups.
As a program director, Leake has mentored hundreds of early-career professionals who are breaking cultural barriers in an effort to accelerate scientific and engineering discoveries. Her multinational programs have specific themes that resonate with global stakeholders, such as food security data science, blockchain for social good, cybersecurity/risk mitigation, and more. As a conference blogger and communicator, her work drew recognition when STEM-Trek received the 2016 and 2017 HPCwire Editors’ Choice Awards for Workforce Diversity Leadership.