hardware accelerators

Question

Hi all,

I'm running rtabmap from a raspberry pi 3b, equipped with a kinect. The publishing rate is pretty low, and I guess it's because of the cpu which is high.

I'm evaluating if to replace the raspberry with something stronger like a Nvida Jetson Nano, or if to get a google coral usb accellerator

Do you have suggestions? If I buy a new hardware, which one do you suggest? ARe these usb devices to accellerate the computations helpful in the context or ROS?

Cheers

---

Originally posted by Batstru on ROS Answers with karma: 1 on 2020-06-15

Post score: 0

---

Original comments

Comment by TurBot on 2022-04-19:
You might consider publishing the relevant topics over wlan and process them on your laptop. In case it has an Nvidia GPU there are also plugins from Nvidia Isaac SDK you could use. It needs some effort building the system properly but it is possible.

Answer 1

The problem is the software. Will your processing software actually use the GPU on the Jetson? It won't unless it was specifically written to use the Jetson or the USB accelerator.

One very simple thing to try is to replace the Pi3 with a Pi4 that has perhaps 4GB RAM. That might give you a 2X speed bump. But it looks like perhaps you need a 10X bump? Then you need to adapt your software.

In general, ROS middleware is not going to do anything with a GPU.

---

Originally posted by chrisalbertson with karma: 136 on 2020-06-15

This answer was ACCEPTED on the original site

Post score: 0

---

Original comments

Comment by jordan on 2020-07-03:
As a point of reference: pi4 with mapping, navigation, and frontier exploration all running concurrently went from occasional max cpu to < 25%, compared to pi3. That said, one should analyze the cause of any high cpu load before just throwing more resources at it. Having optimized the nav stack for pi3, performance on pi4 is outstanding!

Comment by Batstru on 2020-07-03:
Cool! Now please point me out to any documentation to let me to optimize the software :-)

Cheers