From an OS's point of view, only a Mesa build with Teflon and Rocket driver support is needed (available since Mesa 25.3). Inferences can then be executed with ai-edge-litert. I have been experimenting with this for some time on all my devices equipped with Rockchip RK3588/RK3588s.

@savznkvo Installing headless (blind / via network) on a board this new will fail probably. Get a TV and connect an HDMI display or get a serial adapter and connect that for debug. I'll check later with a spare NVME, but I am pretty sure it works (provided, that you compiled the correct firmware). HTH // Sven-Ola

@sven-ola I tried this way too, no luck. Or at least I can't connect to SBC via SSH after.

Hello @savznkvo and @Logan. Directly booting from NVME is not supported by the ROM software burned in the RISCV chip. Instead, the OrangePi RV2 chip is able to boot from SD, eMMC, and Flash / MTD (in this order). You need to prepare the board's internal flash for NVME boot with Boot Armbian from SD card. As root, start armbian-install and select "Install to MTD". Transfer a fresh Armbian image to your NVME. If you do not have the "armbianized" MTD boot config installed, @c0rnelius above hints on booting apply. HTH // Sven-Ola

Hello @1111Windows! Having a portrait HDMI display is rather unusual. Normally, one can buy landscape HDMI displays and simply rotate them via software or video card hardware. However: if reading out the EDID via I2C / HDMI cable does not work correctly, the HDMI display driver may simply fallback to 1920x1080@60 using the compiled-in EDID. You can try to get the EDID from the display with the apt install read-edid tool. If I recall correctly, one can provide a custom EDID binary as file to the kernel via some boot cmd parameter, see https://www.marcusfolkesson.se/blog/custom-edid-in-linux/ for details. Simply configure some mode via video= when the mode is not available in the EDID will not work (as you already found out). It should also be possible to add a custom modeline, but this is deep dive as stated here: https://nyanpasu64.gitlab.io/blog/crt-modeline-cvt-interlacing HTH // Sven-Ola