Identify your R36S board from its .dtb file

The table below is the whole of what the database holds on this hardware: 15 boards against 10 systems. Rows are its boards, columns its systems, and each mark is its verdict, not a test we ran. We have not booted each of these systems on each of these boards, and we are not going to imply otherwise by presenting the grid without saying so.

✓ Fully supported~ Community-tested✗ Not supported (暂不支持)blank: the database does not list this combinationnothing on the published list is marked ✗

The three tiers are the database's own wording, kept rather than flattened into a yes/no: 社区 support means someone reported it working, which is not the same as the project having tested it. A board with a blank cell is one the database does not mention that system for at all.

⚠️ Not verified: none of this grid is ours. It is a community-maintained database; we ported and translated it, checked that our copy matches the original field by field, and left every verdict as it stands. Treat a ✓ as “someone reported this works”, not as a warranty.

The table carries notes, and they are the most useful part of it — they are where a project maintainer wrote down what actually bites. 42 of its 87 board/system pairs carry nothing more than “untested by us”, so they are counted here in one line instead of repeated. The rest are below, in full.

• Read the download notes carefully before installing.

  15 entries: Arkos4Clone — R36S V12 clone · R36S G80C V1.0 · R36S V20 clone · R36S budget variant v03 · R36S budget variant v04 · R36S original, P4 panel · R36S original, newer mainboard · R36S original, P0 panel · R36S G80CAMB V1.2 20250422 · R36S G80CAMB V1.2 20250423 · R36S · R36S Power

• This is a WIP build. If you hit stick or audio problems, regenerate the dtbo with the ROCKNIX web dtbo generator and put it in the overlays folder on the ROCKNIX partition.

  10 entries: Rocknix WIP — R36S V12 clone · R36S G80C V1.0 · R36S V20 clone · R36S G80CAMB V1.2 20250422 · R36S G80CAMB V1.2 20250423 · R36S · R36S Power

• If audio is broken, change load mmc 1:1 ${dtb_loadaddr} rk3326-k36-clone.dtb to load mmc 1:1 ${dtb_loadaddr} rk3326-k36-clone-v2.dtb An OTA update overwrites boot.ini, so you have to edit it again; later UnofficialOS releases can be switched from the EmulationStation menu.

  10 entries: UnofficialOS — R36S V12 clone · R36S G80C V1.0 · R36S V20 clone · R36S G80CAMB V1.2 20250422 · R36S G80CAMB V1.2 20250423 · R36S · R36S Power

• The right stick is detected inverted.

  8 entries: ArkOS K36 — R36S V12 clone · R36S G80C V1.0 · R36S V20 clone · R36S G80CAMB V1.2 20250422 · R36S G80CAMB V1.2 20250423 · R36S

• Supported from the 2026-01-12 build onward.

  2 entries: Rocknix MOD — R36S budget variant v03 · R36S budget variant v04

⚠️ Not verified: these notes are the database author's, passed through unchanged apart from translation. Where one of them names a knob to turn, we have not turned it.

It is on the memory card, in the small FAT32 partition the boot loader reads — the one Windows and macOS can open without any extra software. On a working card it is one of the first partitions on the device, often a few hundred megabytes against a much larger Linux partition behind it.

Which file is in use is written down in the boot loader's own config: boot.ini at the root of that partition, or extlinux/extlinux.conf. The .dtb named there is the one being loaded — that is the file worth dropping above, because it describes the panel the unit is actually running on.

We wrote the long version of this, with the measured panel timings from four R36S variant files, in the device tree article. If instead your card shows up as “needs formatting” and no boot partition at all, that is a different problem — this article covers it, and the answer is not to click format.

The boards above are grouped the way the database groups them: Clone (克隆机), Original (原版机), SoySauce variant (酱油机). Those are labels a community fingerprint library gives a fingerprint — a name for a group of boards that share a device tree, not a finding of ours and not a claim about your unit. We sell R36S handhelds; we are not going to certify one as “original” or “not a clone” on the strength of a hobbyist table, and the board photograph on our own teardown page describes only the silkscreen that is visible in it.

The same goes for what a board “has”. The database lists features such as an audio amplifier or internal storage, and those entries come from device trees people submitted. This tool prints no memory or battery capacity, deliberately: the memory node inside a device tree is a configuration value, not a measurement of what is soldered onto your board, and reading a capacity out of it would be guessing with extra steps.

Three honest limits, all of them properties of the data rather than of the parser.

One fingerprint names two boards. bd2e29ab7c1fdf9a40066821b6f61549 is listed against both r36s_r36s_2025_05_21 and g80ca_mb_v1.2_2025_04_22. If your file hashes to that value, the file itself cannot settle which of the two you have, and neither can we — this page says so instead of quietly reporting the first row. 3 other hashes also carry two names each, but those pair a specific board with a generic type_… placeholder, which is one fingerprint under two labels rather than a real ambiguity; both names are listed when they come up.

Two entries can never be matched by hash. type_1_with_amplifier and type_1_without_amplifier are listed with no MD5 at all, so they only ever surface through screen parameters. A file that is genuinely one of those boards will not match on its hash, and the result panel will say so.

A screen match is weaker evidence than a fingerprint match, and it can be shared. Two different boards can describe an identical panel — our own sample file, a real .dtb from a working unit, matches two database records on screen parameters alone. The result panel labels which kind of evidence it used for exactly this reason.

The table holds two verdicts its own page does not show. R36S V12 clone has an ArkOS K36 entry flagged “Not supported” whose own text reads “ArkOS K36 is reported to run stably on this board.”; R36S Power has an ArkOS K36 entry flagged “Fully supported” whose own text reads “ArkOS K36 is reported to run stably on this board.”. The board's listed systems do not mention ArkOS K36, so the matrix above and the notes follow the published list: we are not going to promote a stored row into a mark on a grid, least of all one that contradicts itself in the same object. If you want to try that combination, the entry is in the table and it is yours to judge.

A record is listed twice. The database gives R36S original, P4 panel and R36S original, P0 panel (id r36s_v12_2023_08_18) the same id although they are different boards. Our copy keys boards on the record itself and not on that id, so both are reported separately — where the database's own code could only ever reach the first.

And the obvious one: a device tree from another Rockchip handheld, or a modified one, is not in this table and will report no match. That is a gap in the database, not a verdict on your hardware. The parameters it did read are printed under every result so you can compare them with the timings in the device tree article yourself.

The systems named in the matrix are community firmware projects for this hardware family. Two of them publish a project home we have verified; the rest are named without a link because we have no address we can vouch for.

• ROCKNIX ↗The project behind Rocknix, Rocknix MOD, WIP and nightly in the table.
• ArkOS4Clone ↗The port the Arkos4Clone column describes; the database itself came from a tool page this project points at.

Named in the table without a link: ArkOS K36, ArkOS R3XS, UnofficialOS, AmberELEC and PAN4elec.

We do not link to ROM sets, integration images or file-sharing mirrors, whichever project they come from. That is a standing rule here, not a gap in this page.

The parser and the database are a port of a public community tool, not something we wrote from scratch. We adapted it into this site, typed it properly, removed every console call, and — because a table of this kind is only worth reading if it is the table that was actually published — checked our copy against the original field by field on real files before putting it on a page. The four deliberate departures are a record's model string that duplicated its name, a gallery of board photographs we do not hold, one feature bullet about memory capacity that we will not print for the reason given above, and the two stored verdicts the database's own list leaves out.