"This whitepaper is restricted and covered by the Raspberry Pi Ltd non-disclosure agreement (NDA). It should not be copied, shared, or duplicated without permission."
I guess it gets some bump from the shared platform with the Pi, but… there's enough SoMs with good platform support at this point. And it's not like they're notably cheaper than those either?
(ed. I guess they are still cheaper than competing SoMs with roughly equal performance… but you pay the price of a poor closed platform instead)
You’re talking about the new (and expensive) Turing RK1 and the Nvidia Jetson as alternatives?
I agree the really well supported platforms are both older as well as more expensive (e.g. i.MX & Sitara platforms.) However, I won't call the RK1 "expensive" considering it outperforms the Raspberry Pi (5) by a larger factor than it is pricier by (roughly.)
[Ed.:] Actually, no, the RK1 isn't better either, given there doesn't even seem to be a datasheet available for the module as a whole.
I guess I'm just living in my expensive but well-supported world of NXP, TI and ST SoMs that I can actually debug…
Wondering if CM5 will offer enough of a boost to allow on-device LLM processing on a Home Assistant Yellow.
Guessing the answer is no, but it might be worth trying.
[1] https://www.tomshardware.com/raspberry-pi/raspberry-pi-compu...
[2] https://pip.raspberrypi.com/categories/945-forward-guidance
https://www.home-assistant.io/yellow/
https://www.jeffgeerling.com/blog/2023/time-card-mini-adds-p...
I have another one on a Bicool Mini Base Board (A) which has a lot more I/O, albeit only one HDMI port and no USB3. It's suitable for desktop type applications with the caveat that the CM4 is barely suitable for that stuff.
Also I've read some e-scooter sharing company used CMs for the smarts in their vehicles. Edit: might have been RPi's not CMs (but the latter wouldn't be out of place).
In short: industrial & embedded uses.