Intel, NVIDIA, MIT, Sandia to Create Petaflops Supercomputer

16 Years Ago EricMack 2 694 Views

The DARPA geeks are at it again . On Friday, the Defense Advanced Research Projects Agency announced it is developing what it calls an . The project is part of what DARPA refers to as its Ubiquitous High Performance Computing (UHPC) program -- that's Pentagon speak for "badass," which the agency itself says has the modest goal of simply "re-inventing computing."

It (the UHPC program) plans to develop radically new computer architectures and programming models that are 100 to 1,000 times more energy efficient, with higher performance, and that are easier to program than current systems.

And DARPA says the trickle-down effect of having ultra quick, efficient and powerful computers that don't melt under the heat generated by their own awesomeness will amount to......at least 50-times greater energy, computing and productivity efficiency, which will slash the time needed to design and develop complex computing applications.

Specifically though, here's the specs DARPA is looking for, courtesy of the UK's Channelregister :

...DARPA wants a petaflops supercomputer, including networking, storage, and compute elements as well as cooling, to be crammed in a space a little larger than a standard server rack - 24 inches wide by 78 inches high and 40 inches deep - and consume only 57 kilowatts to power and cool the device.

The machine has to deliver a peak petaflops of performance and 50 gigaflops per watt sustained power efficiency while running the Linpack Fortran number-crunching test. The system has to be able to do single-precision and double-precision math, 16-bit, 32-bit, and 64-bit integer math, and chew on a streaming sensor array...

Or at least that's the dream. None of the systems have actually been built yet, but the teams that will be working to build them between now and 2018 have been chosen. Intel, NVIDIA, MIT's Computer Science and Artificial Intelligence Laboratory, and Sandia National Laboratory in Albuquerque, N.M. will all be working on prototypes.

No hard numbers on the amount of cash involved, but it has been reported that NVIDIA will receive at least $25 million for the project.

Photo courtesy Argonne National Laboratory, used under a Creative Commons license.

Dani AI

Generated 10 Months Ago

As noted, DARPA’s UHPC (ExtremeScale) competition picked a small set of heavyweight performers to pursue radically denser, more power-efficient supercomputing designs and to seed new execution models for extreme parallelism. The selected teams were charged with producing prototype designs that rethink hardware, cooling, storage and the software stack — a long-term, milestone-driven effort that DARPA scoped toward demonstrable prototypes by 2018. (davidbader.net)

Practical follow-up: NVIDIA was awarded a lead-team contract (reported at about $25M) and assembled an industry/DOE/academia consortium; other performers include Intel, MIT/CSAIL and Sandia National Laboratories leading complementary efforts. Sandia’s lab announcement lists partner and university collaborators and describes its focus on architecture and system integration. These awards were intended to fund multi-phase research toward compact, high-efficiency prototype systems. (nvidianews.nvidia.com)

Why this matters technically: DARPA’s targets pushed several interlocking problems — orders-of-magnitude energy-efficiency, very high sustained per-watt throughput, compact thermal management, and programming models that hide extreme parallelism from application developers. Comparing targets to contemporary practice underscores the gap: the most energy-efficient large machines of the early 2010s achieved a few GFLOPS/W on Linpack, so DARPA’s ambitions were intentionally disruptive. Expect the most useful artifacts to be new runtime/OS ideas, benchmarks and component-level innovations rather than an immediate drop-in commercial rack. For historical tracking, watch vendor reports, peer-reviewed papers from the award teams, and later Top500/Green500 entries for measurable progress. (insidehpc.com)

Be the first to reply

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.