Port thesis
The driver was written and reviewed against documentation; it had never been run against silicon. This site tracks the bring-up on an RDS 2.0 control board: what we tried, what the harness got wrong, what the daemon got wrong, and which cases still gate hashing.
It is a lab notebook for a firmware port — not a throughput leaderboard, not a storage benchmark, and not a product launch page.
Current phase
Upstream enablement
The enablement goes upstream as three pull requests, split by what would change each one: general Mujina, the chip, and the hashboard. The lab harness stays here. Branch heads, and what each branch carries, are read from git when the page is built. The branches live in a private repository, so they are named here, not linked. Every hardware run cited on this page used a Mujina build from before the two newest changes these branches carry, the in-flight result fix and the supply client. The first run of the branches' own code is under way.
The chip and board facts the port rests on, written from our own captures, are in the BZM2 hardware reference.
Proven on hardware
Grouped by the pull request whose branch carries each capability's code, which the generator checks against the branches; a capability not yet built sits where its code is expected to go. A proven row folds its evidence away; a row short of proven keeps its runs open and names the roadmap item that closes it.
RDS characterization
Power-on, logging, calibration, thermal response under the vendor stack — the baseline Mujina must match or beat safely.
Daemon bring-up
Chain link, observation, takeover of a running chain, then mining on one, two and all three boards.
RDS 2.0 topology (schematic)
Stylized control-board schematic: PSU, fans, hashboard chain seats. Not a photo, not a vendor drawing. Status affordances are wired for later; this page does not invent temperatures or hashrates.
What we publish / do not
Publish: case identity, objectives, procedures, pass criteria, per-run status and timings, defects in our own code.
Do not publish: measurements, temperatures, voltages or hashrates unless that specific figure was cleared by explicit decision (each page carrying one says which), serials, private addresses, absolute paths, NDA material. The rule is in PUBLISHING.md and enforced by tools/check-public.sh.
Read next
- Mujina's first pool-accepted share on BZM2 silicon — 2026-09-23, chain tip 968,216
- First hardware week — five defects, four of them ours
- Roadmap — what is required to go upstream and to run on its own, what is nice to have, and enhancements, each naming its epic and issues
- Validation matrix — live phase / case status
- RDS schematic — control board, PSU, fans, hashboards
- Hardware reference — BZM2 ASIC docs (bring-up, board design, UART/TDM, pinout); its repository is public too.