Cray to develop insanely fast $600M ‘El Capitan’ supercomputer for U.S. nuclear simulations

9

Seattle-based supercomputer maker Cray has been tapped to develop a new $600 million system for the U.S. Department of Energy, capable of conducting 3D simulations at unprecedented speeds to better assess and maintain the country’s nuclear weapons capability.

The new system, dubbed “El Capitan,” will be an “exascale” supercomputer, referring to systems that can conduct at least one quintillion (a billion billion) calculations per second, on a level believed to be capable of simulating the human brain. El Capitan will actually be capable of 1.5 quintillion calculations per second, enabling advanced artificial intelligence and modeling capabilities, Cray says.

Peter Ungaro CEO Cray
Peter Ungaro. (Cray Photo)

El Capitan will be more than seven times faster than the world’s current top supercomputer, and more powerful than the top 100 fastest systems in the world combined, said Cray CEO Peter Ungaro in a phone briefing with reporters. Government officials say scientists need these computing capabilities to ensure that U.S. weapons remain a deterrent in the absence of underground nuclear testing, which has been under a national moratorium since 1992.

The new system, scheduled to begin officially operating in 2023, will enable “simulations and calculations at resolutions that are difficult, time-consuming or even impossible using today’s state-of-the-art supercomputers,” said Lisa Gordon-Hagerty, DOE undersecretary for Nuclear Security and administrator of the National Nuclear Security Administration. “It will provide scientists and weapon designers the computational tools to explore the use of new materials and components, improve robustness and safety, reduce maintenance costs, and reduce manufacturing and production costs.”

It’s a giant contract for Cray at a time of major change for the Seattle company. Hewlett-Packard Enterprise is acquiring Cray for $1.3 billion in a deal expected to close in late 2019 or early 2020. On the call with reporters, Cray’s Ungaro made a point of saying that the level of computing challenge faced by U.S. nuclear scientists is similar to those big companies are grappling with.

“The future will be all about the fusion of workloads like IoT, analytics, AI, simulation and modeling, all into one business-critical workflow, operating at unprecedented scale and even in real time, and often this convergence will happen within a single application,” Ungaro said, describing these capabilities as key for next-generation data centers. “Everything is getting bigger and more challenging.”

El Capitan will be created for the National Nuclear Security Administration and Lawrence Livermore National Laboratory, also serving the Los Alamos National Laboratory and Sandia National Laboratories. The system will run on Cray’s new Shasta supercomputing architecture. In a separate announcement, Cray said it has secured nearly $1.5 billion in contracts for Shasta-based systems.

Cray is also developing two other exascale supercomputers for the DOE, for Argonne National Laboratory and Oak Ridge National Laboratory.

UPDATE, 7:30 a.m.: Cray detailed the financial terms of the agreement in an SEC filing this morning, laying out the timetable and milestones for decisions on whether to proceed with the project.

LLNL System Delivery

On July 3, 2019, we entered into a fixed price Subcontract with Lawrence Livermore National Security, LLC under which we will provide items including an exascale-class high performance computer system and services including system maintenance and analyst services (the “LLNL Build Contract”). The LLNL Build Contract consists of several system deliveries, culminating in a final system known as “El Capitan.” The total estimated price to Cray for performance of the LLNL Build Contract is approximately $523 million, inclusive of both products and services and which is subject to and conditioned upon incremental funding and adjustments as described below. This means that the entire contract value is not currently funded but is expected to be incrementally funded as performance progresses and LLNL receives ongoing funding from the U.S. government, with approximately $5 million of the contracted amount funded by the U.S government as of the signing of the LLNL Build Contract. The LLNL Build Contract also contains various options for LLNL to purchase additional products and services.

The systems to be delivered by us under the LLNL Build Contract includes Early Access Systems (EAS) and associated services, an Early Delivery System (EDS) and associated services, a Capacity Tier System (CTS) and associated services, an Intermediate Delivery System (IDS) and associated services and a final system and associated services. The EAS, valued at approximately $5 million with associated services, are planned for delivery by mid 2020 and will provide LLNL an opportunity for early evaluation of system architecture at a prototype level. Following completion of aspects of the NRE and COE activities and a prototype checkpoint, the LLNL Build Contract provides for a “Go/No-Go” decision point to determine whether or not LLNL wishes to proceed with the purchase of the EDS, CTS, IDS, and final system, with such “Go/No-Go” planned to occur in 2021, subject to change depending on factors such as program milestone progress and configuration changes. In the event a “No-Go” decision is made then LLNL reserves the option to wind down the program with no further deliveries by Cray, and this “unwind” option may be exercised by the parties within 15 days of a “No-Go.” If either of the parties exercises the “unwind” option, this would represent a cancellation of the program. In the event a “Go” decision is made, it is expected the parties will negotiate an updated system configuration and performance targets taking into account the outcome of the NRE efforts and projected state of technology at time of system delivery. Because of the “Go/No-Go” decision point, the value of the “El Capitan” system would not be included in Cray’s backlog until after a “Go” decision.

As the “Go/No-Go” decision and system deliveries will occur in the future we have negotiated a mechanism designed to minimize the financial risk of cost fluctuations in key system commodity components that currently include: Memory DIMMS (DDR5), High-Bandwidth Memory (HBM2e), solid state drives (SSD) and hard disk drives (HDD). The final price of the system and/or or the final configuration will be adjusted based on the ultimate prices of these components. Should the price of the risk-sharing components be less than the price assumed in the contract, the total price of the system will be less than the estimated system price above or the customer could decide to increase the size of the configuration. If the total price of the risk-shared components were to exceed the costs assumed in the contract, then the customer could decide to reduce the size of the system from the original configuration or increase the funds to be spent, or a combination of both, to accommodate the cost increase. Any changes to the final price of the system as a result of the risk-sharing provisions will be made dollar-for-dollar based on changes to the covered commodity component costs. The potential total impact of the risk-sharing adjustment could exceed $100 million.

The final system is planned for acceptance in March 2023. Although LLNL currently prefers to acquire the systems with direct funding the LLNL Build Contract contains an option for LLNL to utilize third party leasing services. Additionally, given the size of the final system, the subcontract is structured such that Cray will receive milestone payments based on completing discrete elements of each system delivery (such as, for example, completion of the EDS) and/or purchase of certain components during the build process. As a result, over $400 million may be subject to early payment in the roughly six months prior to final system acceptance to provide earlier cash flow for Cray. Each system is subject to acceptance tests as agreed to by Cray and LLNL. Cash payments may be due following completing discrete elements and “acceptance” of each system delivery even if revenue for such discrete element is not recognized by Cray at that time.

As the LLNL Build Contract is subject to federal funding, Cray has agreed with LLNL to certain termination rights for the convenience of LLNL. In the event of termination for convenience by LLNL, Cray will be paid a percentage of the price reflecting the percentage of the work performed prior to the notice of termination, plus any reasonable charges resulting from the termination which we are able to substantiate to the satisfaction of LLNL, such amounts not to exceed the total value of the subcontract. As the subcontract deals with delivery of cutting-edge technology LLNL has agreed, prior to pursuing any termination remedies, to engage in a review of the subcontract and system requirements and attempt renegotiation of the requirements and schedule



i am as a writer and blogger...


Leave a Reply

Your email address will not be published. Required fields are marked *