Next-gen supercomputer will run 1,000 times faster than the 20-Petaflop Sequoia.[/caption] The market for supercomputers may have shifted more toward Big Data analysis of business and the softer sciences, but its next generation will depend at least partly on people who own nuclear weapons. The Department of Energy (DoE) announced it has given five high-performance computing (HPC) vendors contracts worth a total of $25.4 million to make sure the next generation of supercomputers gets far enough into the exascale era to satisfy the weapons testers who are still among the biggest customers for multi-petaflop computing systems. IBM, AMD, Cray, IBM, Intel and NVIDIA all got pieces of a two-year project called DesignForward, whose goal is to develop super high-speed interconnects that slash the time it takes to shift data among thousands or millions of processors. Making communications faster among masses of computing nodes and individual processors in turn renders the whole system faster – as much as 1,000 times faster than current systems, according to the DoE. "Exascale computing is key to NNSA's capability of ensuring the safety and security of our nuclear stockpile without returning to underground testing," according to a published statement from DoE quoting Robert Meisner, director of the National Nuclear Security Administration (NNSA)'s Office of Advanced Simulation and Computing program, which awarded the contracts. "The resulting simulation capabilities will also serve as valuable tools to address nonproliferation and counterterrorism issues, as well as informing other national security decisions." The work will also address the problem of how to make supercomputers even faster while using less energy – a lot less energy – than earlier models. "Energy constraints are causing our building blocks, microprocessors and memory to change dramatically," Sudip Dosanjh, director of the National Energy Research Scientific Computing Center at Lawrence Berkeley National Laboratory (Berkeley Lab) said in the announcement. "We need to collaborate with computer companies to ensure that future supercomputers meet DOE's mission needs in science, energy and national security." Under the contract, Intel will lead the effort to create more powerful, more efficient interconnect architectures and physical designs. Cray is responsible for developing faster open-network protocol standards; AMD will work on interconnect architectures and execution models. IBM is responsible for making the interconnects more energy efficient and improving intrasystem communication. NVIDIA will focus on tuning interconnects for massively threaded processors and software. The DesignForward contracts are part of DoE's existing FastForward project, which has already awarded $62.5 million to four companies working with DoE on new approaches to extreme-scale computing and R&D projects that might eventually be funded. The goal of the project is to produce by 2020 a supercomputer able to run faster than 1 exaflop/s (1 quintillion operations per second (10 to the 18th power, or 1 followed by 18 zeros). DoE specs require that its exascale computer use no more than 20 megawatts during operation, compared to Sequoia, the 1.6-million-core, 20-petaflop, NNSA nuclear-weapon-simulating IBM supercomputer installed at Lawrence Livermore National Laboratory in 2012, which uses about five megawatts. Image: DoE National Nuclear Security Administration Fast Interconnect is Key to Next-Gen Supercomputer
[caption id="attachment_14226" align="aligncenter" width="600"] Next-gen supercomputer will run 1,000 times faster than the 20-Petaflop Sequoia.[/caption] The market for supercomputers may have shifted more toward Big Data analysis of business and the softer sciences, but its next generation will depend at least partly on people who own nuclear weapons. The Department of Energy (DoE) announced it has given five high-performance computing (HPC) vendors contracts worth a total of $25.4 million to make sure the next generation of supercomputers gets far enough into the exascale era to satisfy the weapons testers who are still among the biggest customers for multi-petaflop computing systems. IBM, AMD, Cray, IBM, Intel and NVIDIA all got pieces of a two-year project called DesignForward, whose goal is to develop super high-speed interconnects that slash the time it takes to shift data among thousands or millions of processors. Making communications faster among masses of computing nodes and individual processors in turn renders the whole system faster – as much as 1,000 times faster than current systems, according to the DoE. "Exascale computing is key to NNSA's capability of ensuring the safety and security of our nuclear stockpile without returning to underground testing," according to a published statement from DoE quoting Robert Meisner, director of the National Nuclear Security Administration (NNSA)'s Office of Advanced Simulation and Computing program, which awarded the contracts. "The resulting simulation capabilities will also serve as valuable tools to address nonproliferation and counterterrorism issues, as well as informing other national security decisions." The work will also address the problem of how to make supercomputers even faster while using less energy – a lot less energy – than earlier models. "Energy constraints are causing our building blocks, microprocessors and memory to change dramatically," Sudip Dosanjh, director of the National Energy Research Scientific Computing Center at Lawrence Berkeley National Laboratory (Berkeley Lab) said in the announcement. "We need to collaborate with computer companies to ensure that future supercomputers meet DOE's mission needs in science, energy and national security." Under the contract, Intel will lead the effort to create more powerful, more efficient interconnect architectures and physical designs. Cray is responsible for developing faster open-network protocol standards; AMD will work on interconnect architectures and execution models. IBM is responsible for making the interconnects more energy efficient and improving intrasystem communication. NVIDIA will focus on tuning interconnects for massively threaded processors and software. The DesignForward contracts are part of DoE's existing FastForward project, which has already awarded $62.5 million to four companies working with DoE on new approaches to extreme-scale computing and R&D projects that might eventually be funded. The goal of the project is to produce by 2020 a supercomputer able to run faster than 1 exaflop/s (1 quintillion operations per second (10 to the 18th power, or 1 followed by 18 zeros). DoE specs require that its exascale computer use no more than 20 megawatts during operation, compared to Sequoia, the 1.6-million-core, 20-petaflop, NNSA nuclear-weapon-simulating IBM supercomputer installed at Lawrence Livermore National Laboratory in 2012, which uses about five megawatts. Image: DoE National Nuclear Security Administration
Next-gen supercomputer will run 1,000 times faster than the 20-Petaflop Sequoia.[/caption] The market for supercomputers may have shifted more toward Big Data analysis of business and the softer sciences, but its next generation will depend at least partly on people who own nuclear weapons. The Department of Energy (DoE) announced it has given five high-performance computing (HPC) vendors contracts worth a total of $25.4 million to make sure the next generation of supercomputers gets far enough into the exascale era to satisfy the weapons testers who are still among the biggest customers for multi-petaflop computing systems. IBM, AMD, Cray, IBM, Intel and NVIDIA all got pieces of a two-year project called DesignForward, whose goal is to develop super high-speed interconnects that slash the time it takes to shift data among thousands or millions of processors. Making communications faster among masses of computing nodes and individual processors in turn renders the whole system faster – as much as 1,000 times faster than current systems, according to the DoE. "Exascale computing is key to NNSA's capability of ensuring the safety and security of our nuclear stockpile without returning to underground testing," according to a published statement from DoE quoting Robert Meisner, director of the National Nuclear Security Administration (NNSA)'s Office of Advanced Simulation and Computing program, which awarded the contracts. "The resulting simulation capabilities will also serve as valuable tools to address nonproliferation and counterterrorism issues, as well as informing other national security decisions." The work will also address the problem of how to make supercomputers even faster while using less energy – a lot less energy – than earlier models. "Energy constraints are causing our building blocks, microprocessors and memory to change dramatically," Sudip Dosanjh, director of the National Energy Research Scientific Computing Center at Lawrence Berkeley National Laboratory (Berkeley Lab) said in the announcement. "We need to collaborate with computer companies to ensure that future supercomputers meet DOE's mission needs in science, energy and national security." Under the contract, Intel will lead the effort to create more powerful, more efficient interconnect architectures and physical designs. Cray is responsible for developing faster open-network protocol standards; AMD will work on interconnect architectures and execution models. IBM is responsible for making the interconnects more energy efficient and improving intrasystem communication. NVIDIA will focus on tuning interconnects for massively threaded processors and software. The DesignForward contracts are part of DoE's existing FastForward project, which has already awarded $62.5 million to four companies working with DoE on new approaches to extreme-scale computing and R&D projects that might eventually be funded. The goal of the project is to produce by 2020 a supercomputer able to run faster than 1 exaflop/s (1 quintillion operations per second (10 to the 18th power, or 1 followed by 18 zeros). DoE specs require that its exascale computer use no more than 20 megawatts during operation, compared to Sequoia, the 1.6-million-core, 20-petaflop, NNSA nuclear-weapon-simulating IBM supercomputer installed at Lawrence Livermore National Laboratory in 2012, which uses about five megawatts. Image: DoE National Nuclear Security Administration 