And now everyone's $2000 Orins will be stuck forever on Ubuntu 24.04 just like the Xaviers were abandoned on 20.04 and the TX1/2 on 18.04.

Nothing like explaining to your ML engineers that they can only use Python 3.6 on an EOL operating system because you deployed a bunch of hardware shortly before the vendor released a new shiny thing and abruptly lost interest in supporting everything that came before.

And yes, TX2 was launched in 2017, but Nvidia continued shipping them until the end of 2024, so it's absurd they never got updated software: https://forums.developer.nvidia.com/t/jetson-tx2-lifecycle-e...

▲

audiofish a day ago | parent [-]

Same experience here, plus serial port drivers that don't work, bootloader bugs causing bricked machines in the field. This on a platform nearly a decade old! The hardware is great but the software quality is abysmal, when compared to other industrial SoC manufacturers.

	▲	tonyarkles a day ago \| parent \| next [-]
		Ahhhh I see there's someone else who has experienced the serial port driver bugs :). I was responsible for helping them figure out and fix the one related to DMA buffers but still encounter the "sometimes it just stops sending data" one often enough.
	▲	mikepurvis a day ago \| parent \| prev [-]
		I think what's most galling about it is that Nvidia gets away with behaving like this because even a decade later they're still basically the only game in town if you want a low power embedded GPU solution for edge AI stuff. AMD has managed to blunder multiple opportunities to launch something into this space and earn the trust of developers. And no, NUC form factor APU machines are not the answer— both for power/heat concerns and the software integration story being an incomplete patchwork.