Exascale Computing Project Midyear Update: Pushing Forward on All Fronts
As we move into the second half of 2017, I am happy to report that the Exascale Computing Project (ECP) has achieved a number of early milestones and is progressing at a lively pace on all fronts.
Our goal of a first exascale system by 2021 is very much in our sights, but bringing all elements of that first ecosystem forward in harmony is challenging. Co-design and integration are always on our minds as we continually evaluate our progress, reassess the obstacles and risks, and focus on the key tasks necessary to achieve this goal.
It is important to note—despite what some of the press coverage in trade publications and other news media implies—that our focus is not on competing with China to get there first. We focus on achieving exascale capability because of the many advances it will enable for the US scientific and industrial communities. We began a decade ago, in 2007, by identifying first the potential benefits and then the challenges that have to be surmounted to field a capable exascale system. Our priority in creating the ECP ecosystem is to bring together all the pieces to drive our national agenda in support of national security, scientific discovery, and economic competitiveness.
Most recently, as I noted in the June ECP newsletter, we announced the ECP’s PathForward program had awarded hardware development research and development (R&D) contracts to six vendors—AMD, Cray, HPE, IBM, Intel, and NVIDIA. The ECP is providing $258 million, and the selected vendors will contribute 40% of the total costs, bringing the total investment to at least $430 million. The PathForward program was created to accelerate the pace of necessary hardware technology beyond the previously planned roadmaps for high-performance computing (HPC) hardware. The contracts support R&D aimed at advancing both system and node architectures, closing gaps in vendor technology roadmaps, or accelerating time-to-market to address ECP performance targets. Through the PathForward contracts, we seek solutions that will improve application performance and developer productivity while maximizing the energy efficiency and reliability of an exascale system.
PathForward program funding is the third leg of building an exascale infrastructure, along with Software Technology and Application Development. Those areas received funding earlier in the project and already have numerous R&D projects underway.
Projects in software technology are continuing to make progress toward the development of a comprehensive and coherent software stack that will enable application developers to productively write highly parallel applications that can portably target diverse exascale architectures. We are working closely with vendors and other US Department of Energy (DOE) facilities to ensure coverage across the software stack and avoid overlap. A gap analysis is underway to identify and address gaps in the current portfolio, based on the updated needs of ECP applications, the needs of DOE HPC facilities, and our expectations of what vendors will develop and provide on the system.
We’ve achieved a number of project milestones with our application development (AD) efforts. The 25 AD projects, 5 co-design projects, and AD productivity and training projects are now firing on all cylinders, having completed more than 100 milestones since the fourth quarter of FY 2016. Included in these milestones is an impressive list of accomplishments and deliverables, including software releases, science-capability demonstrations and benchmark simulations, detailed planning and assessment, algorithm development, performance analysis and improvement, and software design and implementation. Progress on AD projects is being shared with program office stakeholders; these positive interactions are helping to build and maintain advocacy for ECP products and solutions. Finally, a new AD project, the Application Assessment Project, has been initiated to assist application developers in getting optimal performance out of their applications, coordinating and integrating with other Software Technology activities, and developing best and sustainable software practices.
Steady progress has been made in ECP co-design and integration efforts—which is vital to stitching all this together. Application, hardware, and software researchers are organized and working hand-in-hand to integrate all the necessary pieces to ensure the evolution of a holistic US exascale ecosystem.
Collaboration among ECP stakeholders from government, academia, and industry has never been stronger. Guided by our ECP Industry Council, we will shape our exascale R&D efforts to deliver technology and system capabilities of paramount importance to the nation’s industrial sectors as well as to national security and scientific discovery endeavors.
The year 2017 is shaping up to be one of tremendous progress.
Thank you for your continued interest in the ECP. We look forward to seeing many of you in person at upcoming HPC community events.
Director, Exascale Computing Project
Helping Scientists Save Time and Strengthen Software Development
A mathematical, or numerical, software library is a collection of algorithms and software that can make the development of scientific applications better, faster, and cheaper. The Extreme-Scale Scientific Software Development Kit project coordinates and enables dependable mathematical software libraries designed and developed for exascale platforms.
Berkeley Lab Researchers Share Best Paper Award at Intel Workshop
Lawrence Berkeley National Laboratory is participating in the four-year Exascale Computing Project “NWChemEx: Tackling Chemical, Materials and Bimolecular Challenges in the Exascale Era,” led by Pacific Northwest National Laboratory. Relatedly, two Berkeley Lab researchers coauthored a performance evaluation paper on NWChem that received a best paper award at the Intel Xeon Phi User’s Group workshop held recently in conjunction with the 2017 International Supercomputing Conference in Frankfurt, Germany.
OpenMP Tutorial Webinar
The Exascale Computing Project on June 28 conducted a tutorial webinar on OpenMP presented by Oscar Hernandez of Oak Ridge National Laboratory and Tom Scotland of Lawrence Livermore National Laboratory. Along with slides, a video of the presentation is now available on YouTube.
Intermediate Git Webinar
The IDEAS Productivity project, in partnership with several US Department of Energy (DOE) Computing Facilities and the DOE Exascale Computing Project, resumed its webinar series July 12 on Best Practices for HPC Software Developers with a webinar on intermediate Git commands.
Watch the Intermediate Git webinar presentation on YouTube.
A New Co-design Center Website Goes Live
The Center for Efficient Exascale Discretizations (CEED) has launched a website. CEED is one of the Exascale Computing Project’s five co-design centers established to overcome performance barriers by targeting common patterns of computation and communication, known as “application motifs.”
Argonne’s Theta Supercomputer Goes Online
Theta, a new production supercomputer located at the U.S. Department of Energy’s Argonne National Laboratory, is officially open to the research community. The new machine’s massively parallel, many-core architecture advances Argonne’s leadership computing program toward its future Aurora system.
Explaining Computing’s Vast Invisible Software Infrastructure
Rajeev Thakur, director of the Exascale Computing Project’s Software Technology focus area, has authored a special guest feature post in the high-performance computing blog insideHPC that describes the complexities of developing a software stack for exascale.
Berkeley Lab Algorithms Extract Biological Structure from Limited Data
A team of researchers from Lawrence Berkeley National Laboratory has developed a new algorithmic framework that uses advanced techniques to determine 3D molecular structure from very sparse sets of noisy, single-particle data. The accomplishment stems from an Exascale Computing Project collaboration.
PathForward and More on Federal News Radio
Exascale Computing Project Director Paul Messina talks about the PathForward program and various aspects of the ECP in an interview with Federal News Radio.
Lab Partner Article Discusses Centers of Excellence
The driving force behind faster, more robust computing power is the need for simulation and codes that are higher resolution, increasingly predictive and incorporate more complex physics. It’s an evolution that is changing the way the national labs’ application and code developers are approaching design. To aid in the transition and prepare researchers for pre-exascale and exascale systems, Lawrence Livermore National Laboratory has brought experts from IBM and NVIDIA together with Lab computer scientists in a Center of Excellence, a co-design strategy born out of the need for vendors and government to work together to optimize emerging supercomputing systems.