ARM exec counsels massively parallel patience
His chips may drive your car. Eventually
Fusion Summit Don't expect parallel-processing power to invigorate consumer devices for a few more years. But when it does, it may save your child's life – or at least protect you from a nasty lawsuit.
OpenCL, which enables a GPU to share processing chores with a CPU, will be slow to affect the consumer market, seeing as how OpenCL-capable SoCs (system-on-chip processors) are only starting to wend their way to the development labs of device manufacturers.
"The perils of the IP business world means that you produce some IP, you sell it to a certain partner, he makes a chip, he sells the chip to a partner, and yadda-yadda-yadda," ARM Fellow and vice president of technology Jem Davies explains. "It takes a couple of years."
Davies' call for patience came during a causal tête-à-tête with reporters at AMD's Fusion Developer Summit this week, after his keynote presentation expressing solidarity with AMD in the two companies' commitment to a massively parallel future.
Due to the IP business' time-consuming yadda-yadda-yadda, Davies explains, it's going to be a few more years before we see "properly" OpenCL-capable devices employing ARM-based chips.
"I think you know what I mean by 'proper'," Davies said, "compared to some of the sort of skimpy implemetations we see at the moment."
Although it will take a couple of years until those "proper" devices appear, Davies says, there's a good deal of prototyping now going on among embedded-device manufacturers and developers.
Image-processing applications, Davies contends, will be one of the first widespread implementations of OpenCL-on-ARM, due to to the fact that image-processing algorthms are "a fertile ground for parallelism."
Davies wasn't merely talking about red-eye reduction or facial-recognition in images taken by cheesy smartphone and tablet cameras. His sights are set higher – augmented reality is one "obvious case," he suggests.
Of more interest to Davies, however, is an application area with an even longer development cycle than consumer electronic devices: automotive management and control.
Among the image-processing applications that the automotive industry is looking at are white-line detection for "lane-departure detection", and the use of cameras to measure the distance between vehicles.
"Obviously, you can detect if you're about to run into the car in front of you with radar," he says, "but that's expensive. If you can just do it with a camera, that's cheaper."
He also notes that there's "quite interesting" work being done on detecting sudden, right-angle intrusions into a car's path, such as a kid running out into the street in front of your two tons of steel.
But such lifesavers – and other, more mundane smartphone and tablet apps – will take a while to appear. The tipping point that should inspire an explosion of OpenCL-based consumer app development, Davies says, will be – you guessed it – OpenCL-capable consumer devices.
"Once you put a billion devices a year in the hands of consumers and developers," he says, "suddenly people are going to start saying 'Ooo! I think I'm going to do this'."
Reluctantly reaching into his bag of clichés, Davies summed up: "Really, I hate to use this phrase, but build it and they will come." ®