Skip to content
Small team, full backlog, zero orders dropped. Support replies are slower than we’d like. Read our status update → Zero orders dropped. Status → 📬 Check your spam folder — most of our replies land there. We do answer. Status update → 📬 Check your spam folder. Status →

Guided Firmware Selector

Quick answer

The Guided Firmware Selector is a five-question wizard that recommends ASIC mining firmware for your hardware and goals. It matches Bitaxe boards to open AxeOS, and S19/S21-class Antminers to per-chip autotuning firmware such as BraiinsOS+ — the open path to Stratum V2 — or commercial VNish and LuxOS. Dev fees are ranges; flashing carries brick risk.

Answer five short questions and this guide will point you toward a firmware path that fits your goals — with an honest rationale. Custom Antminer firmware builds on the work of projects that opened this space; this tool simply helps you find a sensible starting point. It is educational, not a guarantee — always confirm support for your exact control board before flashing, and remember flashing custom firmware is done at your own risk.

1. Which miner are you running?
2. What is your primary goal?
3. Do you need native Stratum V2?
4. How much do you value open source?
5. What’s your risk tolerance?

Dev-fee figures are ranges quoted as hashrate and can change — check each project’s current terms. Feature notes are grounded in D-Central’s Mining Bible research and reflect publicly documented firmware behaviour as of June 2026.

Where DCENT_OS stands, in plain numbers

Four firmwares get named above. Three of them are closed where it counts: BraiinsOS+ opened its build system, its gRPC API and the BCB100 control board, and it still ships a closed mining binary; VNish and LuxOS are closed. DCENT_OS is the only Antminer firmware in this comparison you can read all the way down — the Rust daemon, the dashboard and the Buildroot image are GPL-3.0, and that is a licence file, not a positioning statement.

It mines. DCENT_OS has produced accepted pool shares on three Antminer models: sustained cold-boot mining on the S9, standalone accepted shares on the S19j Pro, and ~66 TH/s sustained on the S21. Signed images are published today for the S9 and S19j Pro Zynq/XIL lanes. The S21 route is bench-proven and not yet packaged for an in-place install over stock. On the S19 Pro there is a cold-boot bring-up across three chains with zero hardware errors, reachable through a build-and-configure route rather than a stock image, and the S19 rides the same driver path. Everything else in the range — T17, T19, S19 XP, S19k Pro and the Bitaxe-class boards — is driver and bring-up work with no customer write route in this release. The S17 in particular has no route: the daemon refuses that target before it touches the hardware, on purpose, and we would rather tell you that here than have you find out at the flash.

On the fee: there is not one. DCENT_OS charges no dev fee. What it ships is an optional donation, on at 2% by default, with a one-click 0% preset and a 0–5% slider in the first-boot wizard or in Settings any time after. At 2% that is 72 seconds of each hour on D-Central’s pool and the other 3,528 on yours, with a DONATING badge up the whole time. It is a plain configuration value, it is visible on the dashboard while it is running, and the line that sets it is in the public source. Compare like for like, and give the honest ones their due: stock Bitmain firmware carries no fee at all and on pure hashrate cost it beats our default-on 2%; ePIC UMC OS cut its fee to 1.5% in January 2026, which is cheaper than our default for an operator who never touches the setting. BraiinsOS+ publishes 2–2.5% of hashrate, model-dependent; LuxOS publishes 2.8% with no operator-side setting to zero it — mining Luxor’s pool buys you free pool usage, a separate charge, not a dev-fee waiver; VNish publishes 1.8–2.8% of total hashrate, with volume discounts negotiated rather than a switch you can flip. ePIC and Braiins will invoice a flat licence instead, which we do not offer and which some institutional operators genuinely prefer. So we are ahead on this only after you use the control. The difference is not the size of the number. It is that with a closed mining binary the number is a claim, and with ours it is a file you can open.

What it does that nothing else here does: one image auto-detects across five ASIC families by ChipID, so mixing hash-board generations on one control board is a supported idea rather than a hack. It boots management-only — dashboard and SSH up, hash power off — so a fresh flash never surprise-starts a loud miner. Fans come up low-PWM under PID thermal control with a policy that cuts hash before it raises noise, and every dashboard mode carries a BTU/h readout, because a lot of these machines are heating a room. The dashboard is local-first: no cloud, no telemetry, no CDN calls. There is a CGMiner-compatible API on port 4028, so pyasic and the tooling you already run work unchanged.

It is a public beta. Two things are true at once and we publish both: it mines, and the witnessed install capstone is not finished on every lane. Keep a known-good recovery path, read the runbook for your exact control board, and run dcent doctor and dcent install --dry-run before any write. Model-by-model evidence, install routes and the release record →

The four firmwares this wizard chooses between

The wizard asks five questions and then explains itself. Here is the same reasoning in one place, so you can read it without answering anything — and so a search engine can too.

BraiinsOS+

Braiins opened this space. BraiinsOS+ pioneered per-chip autotuning on Antminers, its build system and gRPC API are open, and the BCB100 control board is genuine open hardware under CERN-OHL-S-2.0 — nobody else in this comparison has made an open-hardware contribution of that kind. Two boundaries worth knowing: the mining binary at the centre of it is closed, so there is a floor on how far you can audit it, and it carries a fee of roughly 2–2.5% of hashrate depending on hardware model, collected by directing that share to Braiins’ pool. It is also the only production Antminer firmware shipping native Stratum V2, with V1 fallback — and Braiins authored Stratum V2, which is worth saying plainly.

VNish and LuxOS

The mature closed-source commercial firmwares pioneered much of modern tuning: fine-grained frequency targeting, watt-anchored power limits and home-friendly 110/120V operation. VNish has the broadest model range and the deepest tuning controls; it publishes no licence-terms page and no public source repository on any domain we checked, which is an absence rather than a published closed-source statement. LuxOS has the best-documented API of the three — uninstallluxos is a first-class API command rather than a footnote — and leads on curtailment speed. Its published fee is 2.8%, with no operator-side setting to zero it: mining Luxor’s own pool buys you free pool usage, which is a separate charge, not a dev-fee waiver. Both are Stratum V1 only. Both are capable, well-supported tools and the right answer for a fleet that cannot go down; we default to open options where one fits your goal, and we say so rather than pretending the choice is obvious.

AxeOS (Bitaxe and open hardware)

AxeOS is the open-source firmware for the open-source Bitaxe single-board miner: the cheapest, lowest-power place to learn the whole stack end to end. Since ESP-Miner v2.14.0 (June 2026) it ships a native Stratum V2 client — on the BM1397 the extended channel is the default — which makes the Bitaxe the second family in this list you can run SV2 on today. It is a different class of hardware from an industrial Antminer and we grade it separately rather than flattening the two.

Stock (Bitmain)

Stock firmware is a perfectly valid, zero-dev-fee, lowest-risk choice if you do not need tuning, Stratum V2 or open source. It is also the baseline every custom firmware here is measured against, and below a certain electricity price it simply wins: our cost-of-ownership calculator puts the break-even for a 2.25% fee at roughly 1.5¢/kWh at a 10% efficiency gain. Run the number for your own rate before you flash anything.

One thing that is true of all of them

Autotuner values are calculated at runtime for your specific chips — they are not fixed tables shipped with the build — so allow a tuning pass to settle before you judge a result, and push higher only within your PSU and cooling headroom. And note the distinction that gets garbled everywhere: frequency is set per chip, but voltage is set per power domain, never per chip.

Where to go next

The guide above narrows the field. These are the pages that take a decision and turn it into an install — in the order the work actually happens.