I’m trying to implement some computer vision algorithms for my robot, based on beagle board black.
vision processing is very CPU intensive, so I was wondering if there are any existing frameworks for running parallel computations on GPU , particularly on beagle board
thanks
Yang
Do you think it’s more powerful than the central processor?
For parallel/backround processing the two PRUs may be more suitable. But I’d guess you would have to implement the code for your own and will not see large speed improvements when you often have to access main RAM (but any GPU computation - if possible - would suffer from the same problem).
I’m a new BBB user. I’ve been trying to figure out the advantages between the BBB, the Rasberry Pi 2, and the Jetson TK1 for low-cost platforms. The Jetson TK1 Development is currently available with a promo code for $99.
The OpenCV framework has a few routines that have been optimized to use CUDA. Will these run on an nVidia Jetson TX1 or Jetson TK1? Could you program CUDA or OpenGL directly?
https://developer.nvidia.com/embedded/develop/software
From what I can tell, the unique advantage of the Beaglebone Black is it provides mainline Linux with ultra-low power consumption (2 W) integrated with two 200 MHz PRUs managed with the PRU-ICSS and the ability to interface with various capes. Great for interfacing GPIO and/or buses.
So many considerations: physical size, weight, power requirements, preemptive/nonpremptive OS, integration, etc.