Hands on with the Jetson TK1
I have recently purchased a Nvidia Jetson to use on a new experimental robot platform. Link This is a review/findings of my first impressions with the Jetson. If you do not already know what the Jetson TK1 is please head on over to a earlier blog about it here. Link
This page is going to be more of a hands on/meat and potatoes first look.
When i first unboxed the Jetson I was impressed that Nvidia took the time to package a development board so well. When I proceeded to hook up the board, however I like many others was unprepared slightly. Mainly the reason being is that I have a ESD table and it would be risky to set it right on the surface, I also did not really want to throw on the rubber feet though I may end up doing that later. So here it rests on top of the box it came packaged in. I proceeded hooked up HDMI, a USB hub, wireless keyboard, mouse and the power. The board started right up and loaded to a log in screen however it did not recognize the wireless keyboard for command line so I used a wired one.
I then logged into the system with the default below.
I followed the README.txt document on the TK1 support site, this had a command you run only once. This step is required before you will get to the Ubuntu GUI. Nvidia Support
This went smoothly and after the restart I was greeted to the Ubuntu desktop. A quick note on the this. Ubuntu and start-up run is very quick and responsive so you are definitely getting your moneys worth with the quad core arm chips with inside the TK1. At this point you are using some of the on board Cuda cores for video output though you will not be fully using it until getting a vision processing/Cuda examples running. I then did some house keeping on the system it self on the suggested items on the community support page.
Although Ubuntu is preloaded on the platform Cuda is not is. Note that you need to sign up for and get accepted to the “CUDA Registered Developer Program” on Nividia site and in order to download the TK1 Cuda 6.0 file. I proceeded to natively install Cuda 6.0 on the board. I just followed the install guide on this page “Link” and was running Cuda examples in no time. A quick note is that I would suggest downloading them on a thumb drive first as the browser on the Jetson crashed a few times while tiring to download Cuda. You however my not run into this issue at all. The Cuda examples did run slower compared to my desktop but keep in mind that a desktop has a lot more Cuda cores and consumes way more power doing so.
I found the whole process to be a bit unrefined however this is a bleeding edge development board so I would say it is rather refined in that aspect. In comparison to the Raspberry pi it is light and day as far as setup and raw power. The Jetson just stomps the Pi into the ground. Is that to say the Pi is not a worthy product no. For the 40 dollar price tag there are still many projects the Pi just fits. It is not really easy to review a 192 dollar board against a 40 dollar embedded computer. My personal suggestion is if you can save up the money for the Nvidia Jetson TK1 do it. It is just more practical in terms of usage and computational speed. Nvidia has already begun placing the TK1 SOC in tablets and you will soon see them in smart phones also. If you are still looking for information on the Jeston I will be posting a soon about loading Open CV and running some of the examples on the Jetson. Also make sure to look at elinux.org as it has been very helpful in learning many aspects of the Jetson.