Tensorflow with Jetson TK1

I’ve just ordered my Jetson TK1 kit from Amazon, and wanted to start researching while waiting for it to be delivered. Is there a native build of Tensorflow for this platform available, or a tutorial / reference available for setting up and building the source from scratch? I’d like to compile custom ops separately from the main Tensorflow build, so if there are any references / tips to get me started on this platform, it would be highly appreciated!

Are there any native tools / dev environment specifically for the TK1 available, or most people just use the standard Linux build environment?



For deep learning use-case, it’s recommended to use Jetson TX2.
Our powerful inference engine, TensorRT, is available on 64-bit OS. But TK1 only supports 32-bit Ubuntu.

The available tool for TK1 can be found here: https://developer.nvidia.com/linux-tegra-r216
Tutorial for building TensorFlow on TK1 can be found here: http://cudamusing.blogspot.tw/2015/11/building-tensorflow-for-jetson-tk1.html

Our latest AI libraries for TX2 can also be found here: https://developer.nvidia.com/embedded/jetpack


Thanks for the indepth reply, AastaLLL. I will be moving onto the Jetson TX2, but just wanted to start out on the lower end hardware (TK1) to initially get started and experiment with.

yeah so expensive, I’m also starting with tk1 waiting for the edge tpu