Orange Pi AI Stick Lite $20 Neural Compute Stick Ships with Free Training Tools

Shenzhen Xunlong Software launched Orange Pi AI Stick 2801 neural compute stick last year for $69. It was interesting as an alternative to Intel Neural Compute Stick. But once the documentation and SDK were released a few days later we discovered you could only run demos with the stick for that price, and the training tools would cost an extra $149.

The excellent news is that the company has now launched an updated, yet similar Orange Pi AI Stick Lite for just $19.99, and best of all the PLAI training tools are now entirely free.

Orange Pi AI Stick LiteOrange Pi AI Stick Lite specifications:

  • ASIC – Gyrfalcon SPR2801S Lightspeeur series processor with 2-dimensional Matrix Processing Engine (MPE) with AI Processing in Memory (APiM) delivering up to 2.8TOPS @ 300mW, 5.6 TOPS @ 100 MHz (peak performance)
  • Storage – eMMC 4.5 flash with 68 MB/s read, 84.69 MB/s write
  • Host interface – USB 3.0/2.0 port
  • Power Consumption – 5V @ 200mA
  • Dimensions – 66.5 x 20.5 x 10.8 mm
  • Temperature Range – Operating: 0 to 40°C; storage: -20 to 80°C

The hardware specifications look the same as last year’s model. Supported frameworks include Tensorflow, Caffe and PyTorch, and VGG & SSD models can run on the stick.

The company mentions compatibility with Allwinner based Orange Pi SBC’s, but the SDK is offered for Arm and x86 together with 1-2 model demos. The Plai training tools (aka model transformation tools) which allow customizing the models running on the stick are now provided for free, but for some unknown reason, the company does not provide a public download link and instead tells people to contact them to get it.

All related public documentation available on Orange Pi website was released in November 2018.

Share this:

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress

ROCK Pi 4C Plus
Subscribe
Notify of
guest
The comment form collects your name, email and content to allow us keep track of the comments placed on the website. Please read and accept our website Terms and Privacy Policy to post a comment.
9 Comments
oldest
newest
dimtass
4 years ago

Interesting. Have you bought this? Does it really working as advertised?

dimtass
4 years ago

Yes, I’ve seen that. Also it states that the model transformation tools only support CNN from caffe.
That’s a bit restrictive as most of the people are moving to Keras or Tensorflow. OK, sure you can transform models between those higher level tools, but still this is an additional effort.
I’m triggered, as I was waiting for a cheap stick, but because I’ve already have too many useless HW in my drawers, I’ll probably keep watching on it and see if anyone publishes a review.

dimtass
4 years ago

That would be great Jean-Luc! The price is really good and I’m curious how it performs. Thanks!

pingi
pingi
4 years ago

That would be awesome, than you!
I am keeping an eye on the market, specially for a tool compatible with yolo and feasible to be deployed in nautical settings, therefore power consumption would be interesting.

willy
willy
4 years ago

Yes with a howto serving as an intro for all of us still having zero idea where to start from.

Bruce
Bruce
4 years ago

The reason you can’t download the tools seems to be because they are ‘for sale’ by the chipmaker: https://dev.gyrfalcontech.ai/register

Khadas VIM4 SBC