Hacker News Reader

Comment by Alupis | Hacker News Reader

BeagleBoard is akin to a Raspberry Pi. It's not meant to be a powerhouse PC, or in this case, a powerhouse AI computing platform.

It's meant for embedding, tinkering, and learning.

To that end, 4GB of RAM on an AI Accelerator board is fine - the expected workloads will not consume a lot of RAM. This also makes the lack of NVMe sufficient as well.

For more "horsepower" there is also the BeagleBone AI-64[1], which claims up to 8 TOPS.

[1] https://www.beagleboard.org/boards/beaglebone-ai-64

View on HN

Looks like 1/10 of TOPS and 1/3 of mem bandwidth of Jetson Orin Nano (the TI chip seems able to run 3733MT/s at 32 bit, so 15GB/s) but much cheaper.

For reference:

- BeagleY-AI - 4 TOPs - $72.00

- BeagleBone AI-64 - 8 TOPs - $187.50

- Jetson Orin Nano Dev Kit - 40 TOPs - $499.95

The 20 TOPs Jetson Orin Nano seems to only be available in a commercial module, not in a dev kit. The commercial module is priced at $259.00.

Keep in mind that once you've prototyped a product on the BeagleY-AI board - you can actually BUY the TI AM62A74 chips in volume for ~ $25. Good luck sourcing the SoC on that Jetson Orin Nano board...

The 5-year old Google Coral Edge TPU does 4 TOPS for $40-130. Comparing exact features, the $70 BeagleY-AI probably most directly compares to Coral's $130 option.

https://coral.ai/products/

https://www.aliexpress.us/item/3256806327903926.html?src=goo...

You can buy a full desktop GPU for the price of a Jetson Orin dev board. Not really a good comparison point for a $70 SBC.

A full desktop GPU would drain your battery and take up too much space so it is inferior in almost every aspect for the use cases that the Jetson Orin is built.

Talk about Laptops with dedicated GPUs, that's the most Tops/Watt you're going to get without any of the software stack problems.

Probably the solution to look for if you need >30 Watts of compute.

The Orin boards can do what a desktop GPU can do while drawing 10-20x less power. At the cost of terrible software support and hardly any memory bandwidth.

The first PC where I installed Slackware 2.0, had 16 MB of RAM....

Not a powerhouse indeed.

Your first PC probably couldn't run any modern AI programs. I know there are times when it's reasonable to question whether we really need so much hardware for the things we're running, but I'm pretty sure currentcurrent-gen AI is actually a place where we're using the hardware to its limit.

Reply on news.ycombinator.com