NVIDIA and DARPA work to bring the Terminator one step closer to reality

NVIDIA and DARPA work to bring the Terminator one step closer to reality

Summary: Giving autonomous devices more processing power and capability is a perfect idea.

TOPICS: Hardware

NVIDIA has been awarded a five-year contract from the Defense Advanced Research Projects Agency (DARPA) focused on developing new, higher efficiency embedded processor technologies that would increase the power performance ratio at least 75 fold.



With a goal on delivering at least 75 GFLOPs per watt of processing power, up from today’s achievable number of 1 GFLOP per watt, NVIDIA contract (worth up to $20 million), part of the Power Efficiency Revolution For Embedded Computing Technologies (PERFECT) program aims to make the 75 GFLOP processing power available that is estimated to be what is necessary for automated warfighter technologies in autonomous devices.  With this level of power/performance the expectation s that machines will be able to collect and analyze significantly more data, in real time, than is currently possible.

The NVIDIA piece of the PERFECT project, known as Project Osprey, will make use of novel techniques and NVIDIA experience in heterogeneous computing and parallel processing to deliver massive heterogeneous processing concurrency along with methodologies that allow for the system to deal with increases in soft errors that will result from the significant increase in overall computing speed.

Project Osprey is part of Phase Two of the PERFECT program, focused on developing the technologies and processes to deliver the 75 GFLOP power efficiency.   Proof of concept will be delivered as either a simulation or equivalent demonstration. Phase One was simply to determine if the concept was worth investigating, and Phase Three will develop the required technologies to the point where they can be implemented.

Topic: Hardware

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.


Log in or register to join the discussion
  • The jusm of performance per watt is overestimated

    For example, NVidia says that current limit is 1 GFLOPS per Watt, but Apple's A6X SoC (32 nm) consumes at peak 8 Watts at about 67 GFLOPS, equalling 8 GFLOPS per Watt.

    (Though it should be said that Apple's chip is by far the best what semiconductor technology can master in terms of performance/Watt -- many times better than what NVidia offers currently. Probably NVidia took their own chips to make this 1-to-75 GFLOPS/Watt jump-comparison.)

    Of course, it is quite a long way before 75 GFLOPS per Watt: 20 nm will already make possible 20 GFLOPS per Watt. 14 nm -- 40 GFLOPS per Watt. 10 nm will do the trick, achieving up to 80 GFLOPS per Watt.
    • *jump (thankfully, it is not "jusm")

      The subject.