BeagleBone®︎ AI - Technical Specifications

What is the BeagleBone AI?

BeagleBone AI, a new addition to the BeagleBoard.org BeagleBone family, is a high-end Single Board Computer aimed at developers interested in implementing machine-learning and computer vision with simplicity. BeagleBone AI simplify the use of artificial intelligence (AI) in daily application via the TI C66x digital-signal-processor (DSP) cores, two dual-core ARM Cortex-M4 co-processors for real-time control, two dual-core Programmable Real-Time Unit (PRU) subsystems and four Embedded Vision Engines(EVEs) supported through an optimized TIDL machine learning OpenCL API with pre-installed tools. The BeagleBone AI also has an additional dual-core PowerVR SGX544 3D GPU and a Vivante GC320 2D graphics accelerator. Focused on everyday automation in industrial, commercial and home applications.

BeagleBone AI comes in a similar form-fact as BeagleBone Black and compatible with many BeagleBone Cape add-on boards make it easy to extend the functionality.

Processors and RAM

Dual 1.5GHz ARM Cortex-A15 with out-of-order speculative issue 3-way superscalar execution pipeline for the fastest execution of existing 32-bit code
2 C66x Floating-Point VLIW DSP supported by OpenCL
4 Embedded Vision Engines (EVEs) supported by TIDL machine learning library
2x Dual-Core Programmable Real-Time Unit (PRU) subsystems (4 PRUs total) for ultra low-latency control and software generated peripherals

Memory: 1GB RAM and 16GB on-board eMMC Flash

Connectivity

Gigabit Ethernet
802.11ac 2.4/5GHz WiFi via the AzureWave AW-CM256SM
Bluetooth 4.2 and Bluetooth Low Energy via the AzureWave AW-CM256SM

USB Type-C for power and superspeed dual-role controller
USB type-A host

Audio and Video

IVA-HD subsystem with support for 4K @ 15fps H.264 encode/decode and other codecs @ 1080p60
Vivante GC320 2D graphics accelerator
Dual-Core PowerVR SGX544 3D GPU

Additional Features and Connectivity

2x46 expansion headers compatible with many BeagleBone cape add-on boards
- 16-bit LCD interfaces
- 4+ UARTs
- 2x I2C ports
- 2x SPI ports
- Lots of PRU I/O pins
Zero-download out of box software environment
Self-hosted web IDE with local compilers, libraries and examples
Debian distribution (initially version 9.9, 10.0 available)
Linux kernel support (initially 4.14, with 4.19 and 5.2 available)
Connectivity via USB gadget (network/serial), WiFi access point and station, Ethernet, and serial debug header

Click through to our online stores to Buy NowBuy Now

Top Comments

jeffronium over 2 years ago in reply to jomoenginer

At least all the building blocks for some AI.
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
jeffronium over 2 years ago in reply to clem57

I don't know that the Spartan 6 Cape will not work. I just imagine that there are no Capes listed yet because there are none with specific AI board support.
I'd have a crack if I were you. It would be a great addition to the suite of toys the AI has on board!
- Cancel
- Up +1 Down
- Reply
- More
- Cancel
jeffronium over 2 years ago in reply to clem57

I agree. Each of the resources you mention is very capable, especially with some loe level coding.
Using existing tools will be suitable, and sufficient in a lot of cases, but there is so much further you could go at need. A lot of work though. Good luck with that.
Getting them each to contribute to an overall solution without too much overhead in managing the passing of data betwixt them... That will be the trick.
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
cstanton over 2 years ago in reply to jomoenginer

You can feed back directly to BeagleBoard about this here.
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
jomoenginer over 2 years ago in reply to coolkedar

Like the other BBs, just ensure not to connect the sensor directly to a GPIO pin and be mindful that the GPIOs are 3.3v rather than 5v. Derek Molloy in his books suggests to use a transistor or something similar on the GPIO to driver external circuits.

Here's a note from the BB AI reference doc.
https://github.com/beagleboard/beaglebone-ai/wiki/System-Reference-Manual

See section 7.1
- Cancel
- Up +2 Down
- Reply
- More
- Cancel
jomoenginer over 2 years ago in reply to cstanton

cstanton wrote:

There's a fan cape coming, and also one designed by a VCP of element14 presents , so one product, one project
Yeah, but the board is already released. It just seems that the BeagleBoard folks tossed the BB AI out into the wild and then thought, oh wait, maybe we should have something working with it. I understand it is more of a dev kit vs a Raspi, but it seems like it could have used a bit more cook time before being served. I do want to see it do well though.
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
cstanton over 2 years ago in reply to jomoenginer

There's a fan cape coming, and also one designed by a VCP of element14 presents , so one product, one project
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
clem57 over 2 years ago in reply to jomoenginer

Yeah I understand your point. But I have it already and was thinking ht PL would help capture images quicker and allow some image parsing along the way. You are right about testing on a BBB first...
Thanks
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
jomoenginer over 2 years ago in reply to clem57

Considering the age of the LOGI Bone, it might be best to try it with one of the current BB images on say a BB Black to ensure it still works or see if the folks at Valentfx have updated their images. The images I see for that Cape seem to date back to 2014 or 2015. However, I am not sure why folks are still using the Spartan 6 considering this has been EOL for some time now.

I have a BB AI inbound since I am a glutton for punishment. There is a mini Maker Faire in my locale on October 12th, so I'll see if I can get anything ROS or Vision related running on this before then.
- Cancel
- Up +1 Down
- Reply
- More
- Cancel
clem57 over 2 years ago in reply to jomoenginer

jomoenginer your link just poured water on my fire to use a previous cape "Logibone" a small FPGA Spartan 6. Well I certainly will be giving it a college try when/if selected. Thanks for the pointer...
- Cancel
- Up 0 Down
- Reply
- More
- Cancel