Rockchip RK1808 AI Compute Stick Launched with Linux SDK

Orange Pi Development Boards

Rockchip RK1808 looks like a nifty and inexpensive little chip for artificial intelligence applications delivering up to 3.0 TOPS at low power, and the company has already released documentation and a Linux SDK with Caffe and Tensorflow framework support for the chip.

So the main hurdle now is to get hardware to play with. Some people are selling (samples?) of the official RK1808-EVB on Taobao, but it costs close to $500 US.  There’s still no RK1808 development board, but Rockchip has discreetly launched the RK1808 AI Compute Stick a few weeks ago.

RK1808 AI compute stick

RK1808 AI Compute Stick specifications:

  • SoC – Rockchip RK1808 dual core Cortex-A35 processor with NPU
  • AI inference performance – 3 TOPS for INT8, 300 GOPS for INT16, 100 GOPS for FP16
  • Video – 1080p60 H.264 decoding, 1080p30 H.264 encoding
  • Camera, ISP – 2MP camera support, ISP with BT.601/656/1120 support
  • Host Interface – USB 3.0 port
  • Power Supply – Via USB port
  • Dimensions – 60 x 19 mm

It works just like the already released Intel Neural Compute Stick 2 and Orange Pi AI Stick, and you just need to connect it to a host computer or board like Raspberry Pi 3.

RK1808 AI Compute Stick Specifications

Sadly that’s about all the information I get. It’s very hard to get people answer simple questions like download links for tools and Linux SDK, and pricing. The company’s open source Wiki still does not have anything about the Rockchip AI compute stick.

I first found out about this new platform on Rockchip Korea Facebook account, and if you are interested, they said people could contact them by email to purchase a sample.

Support CNX Software - Donate via PayPal or become a Patron on Patreon

15
Leave a Reply

avatar
3 Comment threads
12 Thread replies
1 Followers
 
Most reacted comment
Hottest comment thread
5 Comment authors
dgpbennitroblu Recent comment authors
  Subscribe  
newest oldest most voted
Notify of
blu
Guest
blu

Hopefully they price it competitively to the jetson nano..

Jon Smirl
Member

This is a $6 chip. As soon as the dev board companies start making things the USB stick should be $25 and a dev board $30. Maybe even less than that.

nitro
Guest
nitro

RK1808 + PMIC is $12 (I doubt that the PMIC is $6…). Factoring in LPDDR3 (presumably), I don’t think the USB stick can be <$50

Jon Smirl
Member

We have quote of $5.89 for the RK1808. PMIC is usually under $1

nitro
Guest
nitro

That’s very interesting. Our quote was $12… Is this recent?

Jon Smirl
Member

Feb 21. I just sent email asking for PMIC price.

nitro
Guest
nitro

Ours was this month. Maybe they changed the pricing meanwhile / you were quoted for higher volume?

Jon Smirl
Member

That is a volume quote, samples are $12 a pair.

Jon Smirl
Member

I don’t have a schematic for the USB stick, but it should only need three main parts – SOC, PMIC, 1GB LPDDR3. $6 + $0.50 + $4.50 = $11. You might need to add a SPI flash for $0.25. $25 retail from Chinese dev board company should not be a problem.

dgp
Guest
dgp

>That is a volume quote,

10K?

Jon Smirl
Member

I believe this is the software for it:
https://github.com/rockchip-linux/rknn-toolkit

Jon Smirl
Member

Check out the toolkit, there is an emulator that runs on the PC so that you don’t have to have hardware.

https://github.com/rockchip-linux/rknn-toolkit/blob/master/doc/RKNN-Toolkit%20User%20Guide%20V1.0.0.pdf

Jon Smirl
Member

I browsed the toolkit docs and there is a section in there about accessing the RK1808 over USB, so this is the software that is used with the USB stick. It should also be possible to put a dev board into USB device mode and use it in the same way. So I’d just buy a dev board and skip the USB stick, this is assuming someone starts selling a RK1808 dev board.

ben
Guest
ben

So is the idea you boot into this usb device, or it meant to be accessed as an attached device from your main linux machine?

What kind of applications are people most likely to use this for?

Jon Smirl
Member

There are two ways to do it when attached to Linux machine, USB stick which is hardwired to boot into USB device mode and talk to the Linux host, or a dev board with software which turns it into a USB device which the Linux host can talk to too. Note that the dev board will also work like a normal dev board for standalone use. This trick should work with a RK3399Pro dev board.

The idea is to use your much larger Linux desktop to develop your neural network. You develop it using the rknn toolkit and the simulator. After you get it working use the USB attached RK1808 to test it on real hardware. Finally move it over to the standalone RK1808 (or RK3399Pro) using the C SDK for running it.

If you don’t care about Rockchip hardware and just want to play with neural networks your money would be better spent on a NVidia RTX 20 class GPU for your desktop. It has special hardware for tensorflow and it plays games too.

RK1808 is well suited for robotic vision systems.