ASIC firmware updates, safely: canary, backup, and a tested rollback
A firmware update is a controlled change, not a chore. Done safely it touches one machine first, runs against a baseline recorded before anything changed, and can be reversed by a path that was tested while reversal was still optional. Done casually, fleet wide on a Friday night, it is how a working row of miners becomes a weekend of recovery work. The verdict for a home operator is simple: if you cannot name your rollback image and your canary machine, you are not ready to update anything.
General information only, not financial advice. Mining returns are not guaranteed. Firmware steps reflect the published guidance named in the text, checked on 2026-11-15. Menus and file names differ by model and control board.
Verify the file before it touches a machine
Obtain firmware only from an approved official source. A search result, a file repost, or a familiar looking download page is not a source. Confirm the exact miner model, hardware revision, and control board compatibility before downloading, because an image built for a sibling model is the classic way to strand a machine. When the publisher provides a checksum, signature, or verification procedure, use it. Record the image version, source URL, download date, and integrity result in a change note. That note takes four minutes to write and is the difference between a procedure and a rumour when something misbehaves a month later.
Third party firmware adds a second evaluation on top of the file check: the publisher identity, documented developer fees, update mechanism, and rollback support. The question is never whether custom firmware is good or bad in general. It is whether this specific package can be trusted and safely reversed in your setup.
Three operations that share one sloppy name
| Operation | Firmware after | Settings after | Use it when |
|---|---|---|---|
| Factory reset | Unchanged, whatever is installed stays | Erased: pools, workers, tuning, dashboard password | The configuration is corrupt or the unit changes hands |
| Revert to stock | Replaced with the manufacturer build | Erased, and tuning results are not recoverable from a file | Ending a dev fee, chasing a warranty conversation, or leaving firmware you no longer trust |
| Unbrick or recovery | Rewritten because the current one failed | Erased | The machine will not boot, will not take an address, or a flash was interrupted |
One asymmetry decides rollback planning. Going back to stock is usually the easy direction, because the miner accepts a manufacturer signed image without argument. Coming back to custom firmware afterwards may not be possible at all. Which stock build you restore is a bigger decision than whether to restore, covered again below.
The update procedure
- Build the update record: model, control board, current version, target version, official source, integrity result, maintenance window, and the rollback image name.
- Back up the current state. Export the configuration where the firmware supports it, and write down pool addresses, worker names, and tuning settings somewhere off the miner.
- Confirm power stability for the window. Power must remain stable through an upgrade, because interruption before completion can require repair. No update during a storm warning or on a circuit already misbehaving.
- Update the canary only: one representative machine, ideally the one you can most afford to have down for a day.
- Compare against the baseline for at least one full day: pool side hashrate, power draw, temperatures, and error log entries, before and after.
- Test the rollback on the canary while the stakes are low. A rollback path that has never been run is a hope, not a path.
- Expand in small batches only after the canary passes, keeping the same comparison for each batch.
What success looks like: the canary hashes to your pool at the expected rate for a full day on the new build, temperatures and error counts match the baseline, and the rollback was demonstrated, not assumed.
The stock build trap
On September 18, Braiins advised operators not to update certain recent Bitmain stock firmware, saying the releases may restrict installation of third party firmware, and specifically advised S21 operators already running Braiins OS. Read that warning at its correct weight: it comes from a competing firmware developer, not a Bitmain security bulletin, and no complete matrix of affected models and boards was published with it. The defensible response is not panic and not dismissal. It is the procedure above, plus one extra line in the update record: before restoring the newest stock build, check whether the build you are leaving can still be installed afterwards. If nobody can confirm that, restoring stock is a one way door and should be scheduled like one.
Decision: update, wait, or skip
Update when the release fixes a security issue, a stability fault you actually have, or a documented incompatibility, and your canary passes. Wait when the release is new, your machines are stable, and the changelog offers features you do not use. Skip entirely when the only reason on the table is that a newer number exists. Firmware is the layer everything else stands on. Boredom is not a changelog entry.
— Dex Carter, Tech Writer