The end of May 2018, saw several important events that continue to advance the Department of Energy’s (DOE) Exascale Computing Initiative (ECI) for the United States. Two of these events, are the passage of exascale budgets by both the full House and Senate Appropriations Committees. These are part of the Energy and Water elements of the Fiscal Year 2019 (FY-19) federal budget. The third event was the submission by industry of their proposals to the CORAL-2 Request For Proposals (RFP) for the Non-Reoccurring Engineering (NRE) and system builds of at least two more exascale computing systems. The country’s first exascale system is projected to be A21 to be installed at the Argonne National Laboratory by 2021. Even though the news was not uniformly good, these three events are more positive steps as the U.S. continues its pursuit of productive exascale computing capabilities to be used in the support of discovery science, national security, and economic competitiveness.
The first event occurred on May 16, when the House Appropriations Committee passed its version of the Energy and Water budget bill. As usual, the high-level bill was accompanied by a detailed report that breaks down the funding levels and provides specific guidance to the agencies. The FY-19 exascale numbers in the report look good, particularly for the Department of Energy’s (DOE) Office of Science (SC) Advanced Scientific Computer Research (ASCR) program.
You may recall that on February 12, 2018, President Trump submitted his FY-19 budget request to Congress that called for a total of $636 million for the DOE’s ECI. The request was divided between the National Nuclear Security Administration’s (NNSA} Advanced Simulation and Computing (ASC) program and the SC ASCR program. The $163 million ASC program request was further divided into $116 million for exascale R&D activities and $47 million for infrastructure work. The $473 million ASCR request put $233 million into the Exascale Computing Project (ECP), which is the SC R&D element of the overall ECI program, and $240 million in facilities investments that will be used to fund the NRE and procurement of the computers.
The May 16th House Appropriations Committee report did not specify a number for the ASC exascale R&D activities, but in that case, the number stays at the request level ($116 million). The report set the ASC infrastructure number at $47 million, which is once again the same as requested. The bottom line for ASC program is that the House numbers are the same as the President’s request, or $163 million. On the SC ASCR side of the ledger, the House reduced the ECP number to $225 million and reduced the facilities number to $225 million. These changes put the overall SC ASCR number to $450 million. In the end, for FY-19, the House Appropriations Committee put the overall DOE’s ECI budget at a still impressive $613 million, but a total of $23 million below the request.
The House report language also identified some important issues. The accompanying language for both the ASCR and ASC programs, voiced concerns that the Energy and Water Appropriations Subcommittee had not received adequate detail about the overall estimated costs of the procurements. For the ASC program, the subcommittee also requested an “analysis of alternatives” to satisfy stockpile stewardship mission needs and to clearly identified threshold requirements for NNSA’s HPC acquisitions. So – while there is clearly support for exascale, the House subcommittee seems to be concerned about its price tag and is starting to ask questions about what will be needed for “beyond exascale” to meet mission needs?
The second important Exascale USA event occurred on May 24th. This is when the full Senate Appropriations Committee passed the Energy and Water element of the FY-19 budget. The NNSA ASC part of ECI ended up exactly in line with the President’s request. The Appropriations Committee approved $116 million for exascale R&D and $47 million for infrastructure preparation for a total of $163 million. On the SC ASCR side, the news was good. The ECI elements received a total of $483 million. The ECP R&D activities were given $233 million in support of their work on applications, middleware software and systems integration. The ASCR facilities part was given a total of $250 million that was split between the Leadership Computing Facilities at Oak Ridge ($105 million) and Argonne ($145 millions). This is especially important for the CORAL-2 procurement because these are the funds that will be used for NRE and the system build work for the SC systems resulting from the RFP. Overall, on the Senate side, the DOE ECI was given a total of $693 million, or just a $10 million increase over the President’s request.
The last major exascale event in May, was the submission of industry responses to the RFP on the 24th. At this point we do not know which companies did, or did not, submit bids. The things we do know is that, like CORAL-1, the proposal preparation process was an aggressive 45 days. Also, like CORAL-1, the proposal preparation requirements were quite extensive. The RFP required the completion of seven volumes that included a number of different configurations and options. Also, the RFP requested that industry run a number of different benchmarks and then estimate the benchmark performance on their proposed systems. The CORAL-2 RFP set the bar very high and made the industry proposal teams work very hard.
With the submission of the proposals, the hard work now shifts behind the scenes to the evaluation process. The DOE and its labs have set an aggressive schedule to quickly get through that process and make awards. As you may recall, the CORAL-2 RFP requested bids for up to three systems. One would be placed at the Oak Ridge Leadership Computing Facility (OLCF). Another one would go to Lawrence Livermore National Laboratory to support NNSA modeling and simulation. This computer could be the same as the OLCF computer but could also be different. The RFP also included an option, if funds are available, for a third CORAL-2 exascale computer to be installed at the Argonne Leadership Computing Facility (ALCF) around 2023. This system would be in addition to the “novel technology” A21 exascale computer that is scheduled to be installed in 2021. That computer could be similar to the LLNL system, but definitely has to be different than the OLCF computer.
Bottom line — on May 24th, the CORAL-2 contracting office at Oak Ridge received hundreds, if not thousands, of electronic pages of proposal material. Now those pages are in the process of getting a thorough examination by researchers at the national labs. The DOE and its labs have promised an aggressive schedule to get that done. In December 2017, the CORAL-2 team held a vendor meeting. At that time, the labs predicted that the RFP would be released in February 2018 and that responses would be due in April. That was supposed to lead to selections being made in May and awards for the NRE work to be negotiated and signed by October. The presentation estimated that the final system build contract awards would begin by the start of 2019. Given that the RFP was released in April rather than February, but that the proposal period was shorten from eight weeks to 45 days, it seems that the RFP is only about a month behind the schedule presented in December. This means that the evaluation teams will be working fast and furious to evaluate all of the data provided in the RFP responses, but that things look good for more exascale systems to start showing up on U.S. lab floors in the 2021 – 2022 timeframe.
All in all, the last few weeks of May were very eventful for the U.S. DOE exascale programs. The news was not uniformly good, but for the most part was very encouraging. Both the House and Senate Appropriations committee reaffirmed the President’s NNSA ASC request and the Senate increased the SC ASCR request. The House voiced some concerns about where things are going and that is going to make some DOE federal employees scramble to prepare reports for Congress. The other big event was the completion of the industry proposal process for the CORAL-2 exascale machines. The procurement process seems to be slightly behind schedule but should allow for the projected delivery of an exascale system to the OLCF in 2021 and its acceptance in 2022. The outlook for Exascale USA continues to look bright. Undoubtedly, China and Europe will keep pressing ahead, but it is clear that the race for computer supremacy is on, and the U.S. is definitely in the running!
About the Author
Alex Larzelere is a senior fellow at the U.S. Council on Competitiveness, the president of Larzelere & Associates Consulting and HPCwire’s policy editor. He is currently a technologist, speaker and author on a number of disruptive technologies that include: advanced modeling and simulation; high performance computing; artificial intelligence; the Internet of Things; and additive manufacturing. Alex’s career has included time in federal service (working closely with DOE national labs), private industry, and as founder of a small business. Throughout that time, he led programs that implemented the use of cutting edge advanced computing technologies to enable high resolution, multi-physics simulations of complex physical systems. Alex is the author of “Delivering Insight: The History of the Accelerated Strategic Computing Initiative (ASCI).”