TERA 1000: CEA completes first milestone towards Exascale

November 12, 2015

The first phase of TERA 1000, the new pre-exascale supercomputer, has been delivered and installed by Atos at CEA/DAM (the CEA’s military applications Division). Completion of this first phase makes it possible to reduce energy consumption by a factor of five, compared to the previous supercomputer TERA 1001, while doubling its peak compute performance.

The compute needs of the CEA’s programs require the implementation of an exascale-class2 supercomputer by 2020. To deliver a system of this size calls for technological breakthroughs. First, to control the electricity consumption of such a system. Secondly, to regulate the data throughputs and to master the huge quantity of data produced by simulations which are ever more precise, in order to model multi-physical and multidimensional phenomena.

To achieve this objective, CEA decided to apply from a very early stage a co-design methodology with Atos, in collaboration with Intel. The aim is to maximize performance by testing interactions between applications and the supercomputer. The CEA’s expertise covers several areas: from the code structure of applications through to code interaction with the supercomputer and up to the architecture of the supercomputer.

The first phase of TERA 1000 installed by Atos at the CEA’s DAM center near Paris includes two compute systems: the first is based on the Intel® Xeon® processor E5 v3 product family and the second is based on the latest generation Bull sequana X1000 system using pre-production Intel® Xeon Phi™ processors (formerly code named Knights Landing). By mid-2016 a very high performance Bull Exascale Interconnect (BXI) will significantly increase the global performance of the system. This system is a forerunner to the future generation of exascale supercomputers that will be implemented by 2020.

The second phase of TERA 1000, which will be up and running by 2017, will consist of 30 Bull sequana X1000 cells, integrating more than 8000 Intel® Xeon Phi™ Knights Landing processors connected by the very high performance BXI interconnect network. It will deliver a peak compute performance of 25 petaflops, with an energy efficiency 20 times better than Tera 100.

Mastering the electricity consumption is a very significant milestone on the roadmap towards the exascale.
1 TERA 100, first system designed and built in Europe to reach the petaflops in 2010, was ranked in 5th position in the worldwide TOP 500
2 exascale : exaflops range computing power (1 Flops = 1 FLoating-point Operations Per Second. The “peta” and “exa” prefixes indicate the number of operations per second : peta = 1015 one million billion ; exa = 1018 or one billion billion).

For François Geleznikoff, Director of Military Applications at CEA, « Tera 1000 is the third generation of supercomputers that stems from the partnership between Bull and CEA-DAM since the beginning of the 2000s. In the meantime, the computing power was increased by a factor of 5000, which contributes very heavily to the improvement of the forecast quality of the digital tools used for defense purposes, but also for industry and research. We are confident that this step opens the way to the exaflops system we need by 2020, with the targeted energy efficiency. »

Philippe Vannier, Executive Vice-President Big Data & Security and Chief Technology Officer of the Atos Group said, « The delivery of TERA 1000 is a key step in the implementation of our Exascale program. It is proof of the capacity of Atos to develop innovative solutions that will significantly reduce the electrical consumption of new generation supercomputers by 2020 – the systems that can reach the exaflops, i.e. a performance of one billion billion operations per second ».

Raj Hazra, Vice President and General Manager, Enterprise and HPC Platforms Group, Intel Corporation, “Intel is collaborating with CEA and Atos on the Tera 1000 project to bring exascale class computing to CEA’s codes. With a key focus on performance and power efficiency, the Intel® Xeon Phi™ processor – code named “Knights Landing” – together with the Intel® Xeon® process​or E5-2600 v3 product family can allow CEA and Atos to enable lower energy consumption while providing breakthrough levels of performance. This is a powerful processor combination that, in systems using the Intel® Scalable System Framework, will be central to meeting the supercomputing challenges of CEA’s Exascale program.”

About the CEA and the Military Applications Division

The French Alternative Energies and Atomic Energy Commission is a public body established in October 1945 by General de Gaulle. A key player in research, development and innovation, the CEA is active in four main areas: defence and security, nuclear energy (fission and fusion), technological research for industry, fundamental research in the physical sciences and life sciences. The CEA’s Military Applications Division (“DAM”) takes charge of defence and security missions. The DAM is responsible for the design, manufacture, through-life support and dismantling of the nuclear warheads that equip France’s sea- and airborne deterrence forces. It is also responsible for the design and manufacture of the nuclear reactors and reactor cores on French Navy submarines and aircraft carrier. It assists the Navy with in-service follow-up and through-life support for these reactors. The DAM is in charge of procuring strategic nuclear materials required for the nation’s deterrence. The DAM also contributes to national and global security through the technical support it provides to the authorities relative to the prevention of nuclear proliferation and terrorism and to disarmament.

Press Contact: Nicolas Tilly– nicolas.tilly@cea.fr