Once completed, the 64-core version of the Parallella computer would deliver over 90 GFLOPS of performance and would have the the horse power comparable to a theoretical 45 GHz CPU [64 CPU cores * 700MHz] on a board the size of a credit card while consuming only 5 Watts under typical work loads. For certain applications, this would provide raw performance than a high end server costing thousands of dollars and consuming 400W.
This is a really interesting board from Adapteva, who are aiming to bring a cheap parallel computing board to the masses. They recently completed a successful Kickstarter campaign for the $99 board that packs a heck of a lot of bang for your buck.
There is increasing interest in Imaging and Machine Vision and the imminent release of the camera module for the Raspberry Pi has been whipping up a bit of a fervour. Unfortunately, I think there may be a little bit of disappointment with the Raspberry Pi solution concerning the grand plans that people have for it in this area. You have to remember that the Raspberry Pi has the equivalent processing of a 486 computer and until the image processing algorithms are ported and optimised for the graphics chip, the RPi is a tad underpowered for vision applications.
Machine Vision
Machine Vision / Image Processing is a very computationally intensive science and has struggled until relatively recently to start to fullfil it’s potential. This has been largely down to a lack of processing power. Some of the great challenges in Image Processing is designing algorithms that are robust at detecting / extracting / tracking features of interest across different scenes and lighting conditions. One of the ways that you can improve this is to overlay multiple algorithms, but of course this increases the number of horses required under the hood. Back in the days of the 486 computer, Machine Vision used analogue cameras with resolutions in the 768 x 576 pixel range and were also monochrome. Less than 0.5MP and if you want to go colour, you triple the amount of data. A Raspberry Pi camera module is 2592 x 1944 pixels or around 5MP AND colour. You are just not going to be able to process that amount of data at reasonable frame rates, let alone 30 frames a second. Optimised algorithms will help, but the basic realities remain.
Parallela
Most Image Processing algorithms lend themselves very well to being split up and performed across an image in parallel. This is what makes the parallela so well suited to Machine Vision. With the small form factor and $99 price, it’s one to really keep and eye on. If you didn’t get in one their KickStarter campaign (as I didn’t), then you can sign up for updates on the Adapteva homepage.