New BeagleY-AI opens up design in familiar form factor

Pronounced “Beagle - Y - A - I”, it is a new family of BeagleBoards that are designed to open up a familiar form-factor with an open hardware reference and built-on components available through distribution.

BeagleY-AI provides 4 trillion-operations-per-second of deep-learning algorithm performance in a form-factor compatible with numerous peripherals already available.

OSHWA certification: US002616

Details and order information at:

Announcement at: BeagleY-AI: Making Computers Open Again - BeagleBoard

Quick start guide at: BeagleY-AI Quick Start — BeagleBoard Documentation

TI Overview page ( j722s technical docs ) : BEAGLEY-AI Evaluation board |


give it another few minutes.

1 Like

Will this run Vulkan and OpenCL?

If it does not run Vulkan what graphics package has this been tested on?

Tried to order from both sources and they are on backorder, do you have a source with stock so we can get a couple of boards?

Any plans for a “cape” with NVMe support and another ethernet port?

Does the GPU require the purchase of a license so it can be used?

What is the “out of the box” touchscreen solution for this board. We don’t want to mess around, just plug it in and light it up.

TI has GPU working in their yocto/arago build… It’s also Imagination based…

NVMe, we have it wired up, just need to get the R5’s to generate the correct pcie clock, but yes, i have this plugged in my alpha board…


1 Like

Why would the R5’s generate the pcie clock?

When the full datasheet comes out it’ll make more sense… But the clock mux for that is controlled by the r5/secure domain, the a53 domain just gets to use it.


Texas Instruments AM67A Arm-based vision processor

I have not been able to find the SDK for this, did find AM62A7 and a 68 version, please post a link to the SDK.

@foxsquirrel the sdk does not exist in public yet… beat at announcing their own product line! :wink:


What’s the difference between this and the regular Beaglebone AI I have?
I was under the impression that all the beaglebone stuff was open hardware?

Original BeagleBone AI was in near BeagleBone form-factor, like BeagleV-Fire and had 32-bit Arm cores (A15) with 2 TOPS engine with only TIDL (C) support.

BeagleBone AI-64 has Arm 64-bit cores (A72) and an 8 TOPS engine with Python (TensorFlow, ONNX) support in an extended BeagleBone form-factor with USB on the header and an M.2.

BeagleY-AI is 64-bit Arm and 4 TOPS in a new form-factor (for us) with a 40-pin male header rather than 2x 46-pin female. PCIe is on ribbon, rather than M.2.

It is, but other boards in this new form-factor (for us) are not certified open source. I won’t name names, but others might.

What did I miss???

Can you elaborate on this.

I think this refers to the various fruit themed boards which are quite popular along with many other similar boards.

1 Like

I will have to look into that.

Concern here is the use of the GPU / NPU.

Value in these boards is not derived from doing “blinky” projects. It is being able to roll out a flashy UI, a UI that uses the GPU.

Just looked into that other company, I am pretty sure everyone can now officially label me an “Old Fool”.
That person’s company, that is selling little boards has over 100 employees, never imagined that market was so big. WoW, that is all I can say…

And the 40 pin header. It looks very familiar, is this compatible with the fruit-themed boards? I haven’t found a pin-out chart yet.

I assume it is, and if so, that means we’d have no ADC inputs? (I would’ve preferred the old P9 header)

Are there any plans to work on a BalenaOS version for this board? Run containers on embedded edge devices - balenaOS

Docker might be nice for a few use cases

I am also curious about real GPU support. I’ve been shopping through all the sbc’s and soms, this is a real pain point that is nearly impossible to get clarity on.

For instance, I think I saw that even the beaglebone ai didn’t actually render using the GPU and reverted to software based.

A yocto layer/entire project that users could build from that actually worked would be huge. Hopefully the change in headers can also get away from the craziness of the capes and dto.

Yes, we fell into the trap, not with BB but a competitor of Ti. The SoM / SoC we went with was a very good choice. It had all right demo graphics on it, vulkan is fast, every thing else was fine. Cool graphics on the SoM was the decision point.

Now in our cycle its time for the GUI, guess what, could not get vulkan to launch. It gets worse after this point…Its a big soap opera.

Point is they all hide the fact that the GPU requires a license or the other negative is they cannot get it certified to run vulkan graphics and math. No disclaimer, I assumed the license followed the chip since they had installed actual demos. Hard lesson learned is its best to get the UI dialed in first then work on the rest of the system.


One approach I was using was if anyone could run chrome with hardware acceleration. I found a patch for the som on the chip as the beaglebone ai. I never tried it. Basically it feels like in the world of soms we are better off using the eval boards directly from the mfg since you can pray you can at least replicate their demos.

The imx8 seems to be the best supported for GPU. Wonder what makes beaglebone stick with TI. Maybe they are sponsored and I never looked into it.

I think the greatest contribution would be beaglebone creating a tutorial on how to build using yocto and the meta-ti layer. There is a great udemy course on this actually for the black. The sdks from the mfg always feel like “lock in” on something supposed to be portable. Additionally using the images provided ends up being a crutch as it relies on Nelson being a linchpin.

I will say though the beagley-ai looks really great. I can’t really see any reason to get the beagleplay when this would be a better option. I wíll be buying it as a dev board

I agree, it does have potential. If the GPU is online that would be great too. If the header is P2P with the Pi that will be a fair marketing tool, that is one nice thing about the pi they have plenty of 3P sellers with cool break out boards.

2 weeks later it will be obsolete…
We are running weston compositor on a I7 test machine with a yocto build and so far its doing very well. Also do yocto on the bbb, sometimes it is so much simpler to appreciate all the work that goes into the pre-built images.


I found some neat “Capes” for that form factor. I do not quite say now it can definitely work but…

GitHub - mjbots/pi3hat

The first link was incorrect.

Anyway, there is a “Cape” with that form factor that may be worth porting if people get around to it. I tried on the BBAI-64 and without further exploration, I stopped development.

Fail on my part in a big way but now with the new form factor, it may very well be a possibility. Do not get me correct in saying I am affiliated with him or his maker ability. I probably know less.


P.S. I just thought since the newly revised board addition is making some “waves,” one could try (me…maybe) or someone with very, expert sensibility. Anyway, notta from me lately. On with it!

1 Like