I've also used the Raspberry Pi + Coral TPU. For a while it was almost impossible to get the USB as it was always out of stock, but those issues seem to be gone. I would not, however, recommend the devboard as the documentation is quite sparse (runs on an NXP processor). RPi+Coral is a good combo, and the Pi5 supports the PCIe accelerators.

The BeagleBoard BeagleY-AI that was just released costs $70 at 4TOPS, equal to the Coral which is $60 for the USB alone disregarding the cost of a Pi [2]. It just came out, but BeagleBoards have quite a good community reputation and because the hardware is more open, easier to develop into a commercial product. The older BeagleBone AI-64 has a processor that supports 8TOPs, and TI has processors that go up to 32TOPs if you have the skills to create your own board (I don't).

I would not recommend developing with the NVIDIA Jetson unless you have particularly deep pockets and like to use an out-of-date toolchain. The latter isn't too unfamiliar to the embedded world though.

[1] https://pineberrypi.com/products/hat-ai-for-raspberry-pi-5 [2] https://www.beagleboard.org/boards/beagley-ai

Thank you so much for sharing your detailed insights and experiences! I'm particularly intrigued by the Raspberry Pi + Coral TPU combo, and it's reassuring to hear that the availability issues seem to have been resolved. The mention of the newly released BeagleBoard BeagleY-AI also caught my attention, especially given its cost-effectiveness and open hardware approach. I've appreciated the support from the BeagleBoard community in past projects, so the BeagleY-AI seems like an attractive option to explore.

Regarding the NVIDIA Jetson, your comments affirm some of my reservations about the costs and the currency of the toolchain. It seems like it might not be the best fit for my needs at this time.

Do you have any experience or tips on getting started with either the Raspberry Pi + Coral setup or the BeagleY-AI, particularly? Any advice on kickstarting a project and navigating potential pitfalls would be invaluable. Also, if you have any recommendations for resources or communities focused on these platforms, I would be extremely grateful.

Out of curiosity, was this comment generated by an LLM (at least partially)?

The wording gives that kind of feel and now I'm wondering if my feeling around it is at all accurate.

> Out of curiosity, was this comment generated by an LLM (at least partially)?

Same author as https://news.ycombinator.com/item?id=39876247

Definitely. I don't understand why people feel they need to do that, especially here.

I wanted to address the comment you made. It's important to understand that large language models have become integral tools for skilled prompt engineers. This isn't an oddity but rather a natural evolution for this new professional field. In my view, your response seemed to carry a tone of prejudice and skepticism, which I find questionable. The advancement and integration of these models into various workflows are indicative of progress and innovation, not a cause for derision or doubt. I hope we can maintain a more open and respectful dialogue moving forward.

Thanks for the suggestion. I just ordered a Coral USB to hopefully play around with.

Related to the Coral accelerator, is it usable as a general accelerator outside AI workflows? When I looked into it, there was no general SDK on using the Coral, only an "ingest ML model" compiler.

Why not go with the Raspberry Pi 5?