Interesting anecdotal datum:
This morning, I was running an OpenCV program, (on my GoPiGo3 / RPi 3), that allows user to determine useful RGB or HSV masking values, by looping showing the input image, a set of sliders, and the masked image.
I stepped away for a while, leaving the program running, and upon returning I noticed the processor temperature had risen to 78degC which is approaching the 80degC temperature thresholding limit.
The program shows 60 to 65% in the CPU Usage Monitor of the Raspbian For Robots noVNC Desktop and uptime reports average load of 3.5.
Another interesting datum: the additional load caused my smart charger to falsely detect full charge status, causing Carl to get off his dock early (only 2 hours of usual 3 hours). He will have a shortened playtime and will return early for his next fill-up.
This experience may suggest caution when designing and testing “real-time” OpenCV applications.
My plan will use periodic single-image analysis which I don’t expect to be a problem on the GoPiGo3.
Not surprising, given the fact that it’s probably beating the living [censored!!] out of the Pi’s GPU, and GPU’s tend to get hot.
Actually, there are a number of software packages that will beat the camel-dung out of your system.
in the July issue of Linux Format, there’s an article on how to spin light/embedded Linux images for a number of different systems, including the 'Pi, using a utility called Yocto.
(Custom embedded Linux images, Tam Hanna, Linux Format, issue 251, pp 52-55).
@Francesco_F might be especially interested in this, as it seems to be particularly suited to Ubuntu, and it can spin images for the Pi. Even on dissimilar hardware like a PC - which is essentially mandatory unless you’re a bloody masochist.
As far as stress is concerned, the author mentioned in a side-box to the main article that they had purchased “a high performance workstation based on an FX8320 octa-core processor from AMD, along with 16GB of RAM and a 128GB SSD” for testing purposes. When all was said and done, not only had they consumed every byte of storage on the SSD, the processing demands brought this octa-core behemoth to it’s knees! As a parenthetical comment, after noting how resource intensive the Yocto process was, he suggested that die-hard overclockers and gamers might find it useful for stress-testing their machines. . . . With this in mind, though you can run Yocto on the 'Pi, it’s really not a great idea - unless you’re a glutton for punishment and have a lot of vacation time saved up!