Alasdair Allan


First announced back in February at Embedded World in Nuremberg, Germany, the BeagleBone AI started shipping at the tail end of last week.

The new BeagleBone AI. (📷: BeagleBoard.org)

It may share the familiar BeagleBone form factor, but the new BeagleBone AI isn’t like the BeagleBone boards we’ve seen before. This isn’t really a general purpose single-board computer (SBC) instead, as you can tell from the name, this board is intended for machine learning inferencing at speed.

Built around the Texas Instruments AM5729, a dual-core 32-bit Arm Cortex-A15 processor running at 1.5GHz, the board looks at least initially to be underpowered when compared to the new Raspberry Pi 4 which has a quad-core 64-bit Arm Cortex-A72 also clocked at 1.5GHz. But unlike the Raspberry Pi 4, the board that at the moment I’d recommend as the ‘best in class’ for getting started with machine learning, the BeagleBone AI has some extras.

As well as the main processor, the board features two dual-core Programmable Real-Time Unit (PRU) subsystems and four Embedded Vision Engines(EVEs), plus an additional dual-core PowerVE SGX544 3D GPU and a Vivante GC320 2D graphics accelerator.

I’m really not sure how the new BeagleBone AI is going to measure up to the either the Raspberry Pi 4, or the new generation of custom silicon intended to speed up machine learning inferencing at the edge. It’s going to be fascinating to find out, although the results might well depend on how well those four Embedded Vision Engines are supported by TensorFlow, if at all?

TI Deep Learning API examples on BeagleBone AI. (📹: Jason Kridner)

So I’m rather looking forward to getting my hands on the new BeagleBone AI as, along with the new RockChip RK1808-based USB accelerator hardware, it is one of the boards I’m most intrigued about right now. But more on that as soon as I get my hands on the hardware, which should hopefully now happen sometime next week.

The BeagleBone AI costs $125 and is now shipping.





Source link