Tuesday, August 20, 2013

USB3 Vision versus GigE Vision

If you couldn’t wade through the presentation I offered in my last post, “Understanding USB3 Vision” here’s a graphical comparison of the two interface formats.

I think the message here is that USB3 is a good choice when you have large images from single camera systems and you don’t need a long cable length. However, most single camera machine vision applications are, in my humble opinion, best met with smart cameras. If you go towards a PC solution you’re probably using either linescan or multiple cameras, in which case I think you’ll want CameraLink or GigE.

In other words, I’m still on the fence about the benefits of USB3 in machine vision.


Ivan Klimkovic said...


As far as i know USB3 Vision works rather effectively with Multiple camera setup (even 26 units or more), plus the cable lenght is increasing ever since and with active repeaters can easily reach 100m. Some companies are working on several hundred meter cables now.
Or did you have something else in mind ?

John Phillips said...

Curious to understand how you determined the values on the "cable security" and "CPU load" axes.

They should be equal on cable security, as there are widely available locking connectors/cables for GigE. Locking cables can be certified under the USB3 Vision standard, but the standard does not mandate their use. I'd almost argue that GigE cables are more secure, given that they have a little tab that prevents them from being pulled out - assuming you haven't broken the tab already :)

On the CPU usage front, yes, USB uses DMA which avoids some CPU cycles. But most GigE Vision SDKs have some sort of driver which helps reduce CPU usage below what the standard (Intel, Marvel, etc.) Windows driver uses on its own. Plus, with today's modern multi-core PCs running at high clock rate, the overhead in transferring GigE data into application memory is basically noise.

Disclosure - I work at Pleora, a manufacturer of both USB3 Vision and GigE Vision products.