Chelson a

Reviving this because I had exactly this situation — S905W box, eMMC completely dead — and it **is** possible to boot from SD with no working eMMC. Firstly i had to get the UART all setup to communicate with the board just to find out it was stuck on a boot loop because the emmc was dead.. i was going to replace this later Ideally you need serial communication to get this board working. this will allow you to talk to the hardware otherwise you are just guessing! I have one running Armbian 26.08 (kernel 6.12.103-ophub) right now, booting entirely from the card with nothing attached to the PC. SteeMan is right that the stock aml-s9xx-tvbox builds chainload from the factory u-boot in eMMC. The fix is to put a real Amlogic bootloader in the card's raw sectors so the mask ROM boots the card directly. **The part that isn't obvious** is why the sensible-looking attempts fail. It isn't the boot method — it's the DDR parameters. Every board's memory training blob (`acs.bin`) lives in its own bootloader, and on a TV box that means inside the eMMC. When the eMMC dies, the correct parameters die with it. Borrow a bootloader from another board and BL2 says: DDR3 chl: Rank0+1 @ 912MHz - FAIL DDR3 chl: Rank0 @ 912MHz - FAIL DDR3 chl: Rank0 16bit @ 912MHz - PASS. It falls back to half width, passes a quick self-test, then corrupts memory under real load. You get a boot that reaches systemd and then panics with oopses at random addresses. I chased that for hours; it is not a clock speed issue (768 MHz/CL11 corrupts exactly like 912 MHz/CL13). **What worked:** a u-boot built for the Tanix TX3 Mini, which is also a p281 S905W board, in the matching RAM size: https://github.com/arnarg/tx3-mini-uboot-build release v2019.01 u-boot-v2019.01-tx3-mini-1g.bin (1 GB boards) u-boot-v2019.01-tx3-mini-2g.bin (2 GB boards) Already in raw-sector format, so: dd if=u-boot-v2019.01-tx3-mini-1g.bin of=/dev/sdX conv=fsync bs=1 count=444 dd if=u-boot-v2019.01-tx3-mini-1g.bin of=/dev/sdX conv=fsync bs=512 skip=1 seek=1 One more thing to watch: ophub's `boot.scr` only scans `mmc 1`, and with a dead eMMC your SD may come up as `mmc 0`, so it looks straight past the card. Write your own `/extlinux/extlinux.conf` on the boot partition (the image ships only an `extlinux.conf.bak`), pointing at `/zImage`, `/uInitrd` and `meson-gxl-s905w-p281.dtb`, with the APPEND line copied from `uEnv.txt`. Also keep `video=HDMI-A-1:1920x1080@60e` on the kernel command line — without a forced mode u-boot can hang at console init when no display is detected. Full write-up with the failure modes of every bootloader I tried, and the serial logs: Credit to @arnarg for building and publishing those bootloaders — that repo is the thing that made it work. u-boot-tx3-mini-1g.bin README.md scripts.zip

This is an awesome effort from you all! I purchased a v3. 1 2gb ddr3 version off ebay for 25usd and tried out the mate graphical image and it worked like a charm. Just some feedback for you all and others like me that read this forum. 996mb ram I think it is as reported by other users also. Vulkan graphics seem to be working, WiFi and usb ports all working out of the box. Bluetooth appears but I didn't try to connect to any devices yet. I even tried the tiny stories 656k ollama llm to see if that would work and its pretty fast with some bad stories haha even the 1b parameter model loaded but it's barely working. I will try llama.c next to try get some. Kind of acceleration from the gpu. Supertuxkart runs very smoothly. I installed on a 32gb SanDisk tf card with no worries flashing using etcher. It's quite a capable little machine. I even managed to compile a game I'm building at the moment for Linux + dos called deadseas which runs rather well. I havnt tried sound or anything yet as I'm direct into an old monitor via hdmi > dvi. So much better than the stock Android that the x96q comes with. Anyway thanks for your efforts!