Tuesday, August 27, 2013

Trading bus stops for credit cards: how far embedded computing has come in three decades

Posted by John Keller

THE MIL & AERO BLOG, 27 Aug. 2013. Sometimes I have to stop and marvel at how far embedded computing has come since I started paying attention back in the mid-1980s as I first started out as a trade press reporter.

Those were the days of Cray X-MP supercomputer -- something that literally had benches around it, was taller than a man, and looked a lot like a bus stop. Its theoretical peak performance was 800 million floating point operations per second (800 megaflops).

The Cray X-MP and its successor, the liquid-cooled Cray-2, were considered to be the fastest computers of their day, and their use was confined largely to government research centers for things like nuclear weapons simulation, advanced sonar research, and computational fluid dynamics -- or simulating a wind tunnel in a computer.

There was little, if any, practical use for these kinds of supercomputers for actual deployed military applications. You couldn't fit them on a ship, submarine, or aircraft, and the delicate machines most likely couldn't have withstood the shock, vibration, and other environmental rigors of the field.

There was hope, though. The Holy Grail for DARPA embedded computing scientists was to package one billion floating point operations per second of performance in something deployable. The mantra, at first, was "a megaflop in a shoe box," which evolved to "a megaflop in a coffee can", and eventually to "a gigaflop in a soup can."

Some of the best minds in industry and academia were put to work by research groups like DARPA to make the megaflop-in-a-shoe-box dream a reality.

It would appear their success has surpassed even their wildest dreams. Today we're seeing gigaflop performance on single-board computers and mezzanine modules the size of credit cards that are available off the shelf. We no longer talk about supercomputing, and now describe that kind of technology as high-performance computing (HPC).

Companies like Curtiss-Wright Controls Defense Solutions, GE Intelligent Platforms, Mercury Systems, and others embedded computing firms routinely offer today what DARPA computer scientists were only dreaming about a few decades ago.

Not only are today's gigaflop-performance embedded computing devices setting new speed records, but they also are being developed to be practical for a wide and growing variety of applications. The Curtiss-Wright Fabric40 program is only one example of an emerging ecosystem of embedded computing products with gigaflop performance, and the data throughput to keep these high-performance processors fed with data.

I thought about this earlier this month at the Association of Unmanned Vehicle Systems International (AUVSI) conference and trade show in Washington. For what the best and brightest could only dream about years ago, anyone could walk down those aisles at AUVSI and write a check.

Take a look at the photo above of that Cray X-MP, taken in the 1980s. What that check wold buy today would fit in the guy's shirt pocket, and would be more than 10 times the computing power of the supercomputer behind him.

No comments:

Post a Comment