• v0.1.17 4aba9cd486

    WarpAxe v0.1.17
    All checks were successful
    CI / Rust checks (push) Successful in 45s
    CI / Regulator scenarios (push) Successful in 47s
    Stable

    dvb-projekt released this 2026-10-06 01:50:49 +00:00 | 470 commits to main since this release

    Free clock values on the Gamma Duo and Gamma Turbo. Their profiles capped the clock at the highest stock step (410 MHz on the Duo, 690 MHz on the Turbo), even with ?oc in the address and over the API. AxeOS has no such cap, and WarpAxe never had one on the plain Gamma. A Turbo that ran 850 MHz / 1250 mV under AXP90 coolers kept those values when it switched from AxeOS but could no longer change them (thanks John11). With ?oc the clock is now free on both boards. Without it you still choose from the stock steps, and the core voltage stays capped at 1250 mV.

    Everything from v0.1.16 applies (regulator floor, NerdAxe Gaia on its own regulator range, the Wi-Fi radio no longer sleeps); see its notes.

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.16 e1e21b175e

    WarpAxe v0.1.16
    All checks were successful
    CI / Rust checks (push) Successful in 46s
    CI / Regulator scenarios (push) Successful in 49s
    Stable

    dvb-projekt released this 2026-10-06 00:48:54 +00:00 | 472 commits to main since this release

    Never below the regulator's clean range. A core voltage is no longer commanded below the range the board's regulator holds cleanly. A TPS546 regulates accurately, with its undervoltage protection, from 0.25 V to 0.75 V at its reference, which is 1–3 V on the 0.25 divider most boards use and 0.5–1.5 V on 0.5. The driver refuses anything under that, PATCH /api/system answers such a voltage with 400, and /api/system/info publishes the floor as regulatorMinMv (1000 mV on the Bitaxe boards and the BitForge Nano, 800 mV on the NerdAxe Gaia and the TPS53647/53667 boards), so a tuner never searches below it.

    NerdAxe Gaia: regulator in the range the board is built for. The Gaia runs its BM1373 at 800–970 mV, but the stock firmware set its regulator to the 0.25 divider, which leaves those voltages below the clean range: no accuracy promised and no undervoltage warning. WarpAxe now uses 0.5 (clean 0.5–1.5 V). After a power cycle the regulator itself comes up on 0.5 with the same loop compensation, so this is the setting the board's straps were chosen for. On our Gaia at 340 MHz / 950 mV over seven hours the regulator read 945 mV in every minute, with 1,828 shares and none rejected; power and the chips' error rate stayed where they were on 0.25. The change puts the regulator inside its specified range; it does not make the Gaia more efficient.

    Honest about what did not happen. A core voltage the regulator turns down no longer counts as set: the old one stays, the request is retried every second, and the API shows coreVoltageRefused with the reason in boardError. Before, a refused value was taken as applied, so the rail could come back after an overheat stop at a voltage it never had. Chips that stop sending nonces are named in silentChips; chain restarts after a silent minute are counted in chainSilentRestarts, and in a row at the current core voltage in chainSilentRestartsSameMv (two or more: the chips do not run at this voltage). The firmware does not raise the voltage on its own; WarpTune reads the counter and steps up.

    Wi-Fi radio stays awake. WarpAxe left the radio in ESP-IDF's default modem sleep, which wakes only at the access point's beacons, so pool jobs and web answers waited up to ~100 ms for the next one. It now stays awake: on our NerdQAxe++ the ping went from 65 ms on average (up to 200 ms) to 8 ms over a night, web answers came about 20 ms sooner, and a network scan no longer misses a miner whose radio was asleep (seen on a dev board). The cost is a few tenths of a watt.

    Also in this release:

    • vrProbe in /api/system/info: the regulator's window (divider, set point, limits, compensation) as found at boot, before the firmware writes it.
    • Share difficulties in the log are rounded to three decimals instead of a long float (thanks John11).

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.15 e7f82106a7

    WarpAxe v0.1.15
    All checks were successful
    CI / Rust checks (push) Successful in 44s
    CI / Regulator scenarios (push) Successful in 48s
    Stable

    dvb-projekt released this 2026-10-03 08:11:00 +00:00 | 478 commits to main since this release

    An update you can follow. An online update now shows a bar under the header on every page: download, check and write with progress, then the restart, then "Update done: the miner runs WarpAxe 0.1.x" once it is back (the page reloads for the new web UI by itself). A failed update, a restart into the old version and a miner that stays silent are named. A manual upload follows the same path instead of reloading blindly after 45 s. (Thanks John11.)

    A full backup before the first install, and a way back. The web installer now saves the miner's whole flash — stock firmware and settings — into a file you choose before it writes anything (on by default; 16 MB in about 2½ minutes, every block checked against the chip). Its new "Restore a backup" section writes such a backup back over USB, settings included. On the miner, System → Back to the stock firmware explains the way and opens the installer there. Reading the flash in Chrome on macOS needed small, acknowledged packets: the serial driver there silently dropped bytes of 0xFF-rich flash data.

    Chips off before every restart. An update, a restart and download mode now first switch the core voltage off and hold the chips in reset, and every boot does so before anything else. The TPS546 regulator (Bitaxe Gamma, Gamma Duo, Gamma Turbo, BitForge Nano, Nerd Gamma/Gaia) keeps its output on through an ESP reset, so the chips sat powered and unwatched meanwhile: a BitForge Nano reached 82 °C, its regulator 95 °C, after three minutes in download mode, and tripped the overheat stop right after the boot. Likely also behind the "overheated" seen during an update on a Gamma Duo/Turbo (thanks John11).

    Also in this release:

    • Room for more web API endpoints (64 instead of 40) and a count against the limit at start; v0.1.14 used 39 of 40, so the next endpoint would have left the web server off.
    • The hash-counter card shows a real 0 % error rate (a Gamma can run minutes without a single error); only the first minute after a start counts as not measured yet.
    • The installer's "no answer" message names download mode in WarpAxe's web UI first: a running WarpAxe holds the USB port, so the installer cannot reset it into the ROM loader itself.
    • The installer's board list and warnings are up to date.

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.14 4da095383d

    WarpAxe v0.1.14
    All checks were successful
    CI / Rust checks (push) Successful in 44s
    CI / Regulator scenarios (push) Successful in 46s
    Stable

    dvb-projekt released this 2026-10-02 07:10:28 +00:00 | 490 commits to main since this release

    Both fans follow the chips, as on NerdOS. On the Nerd boards the second fan output (M1) now runs at the chip fan's speed by default — NerdOS does the same. Before, it followed the voltage regulator's temperature, and when switching from NerdOS that setting was taken over wrongly. On a NerdOctaxe-γ v2.2 at a user's, the regulator stayed cool, so M1 sat at its minimum while the eight chips heated up, and the board stopped on overheat within a minute; with both fans linked it runs. Fresh installs and switches from NerdOS get the linked fans; an existing WarpAxe configuration keeps what it has — if your Nerd board runs hot in automatic mode, check Settings → Fans → Link the fans.

    Fans up while the chips warm up. After the core voltage comes on — at start and after an overheat stop — the fans stay at full until the chips are near their target (or stop warming), then the controller takes over from there instead of first running the fans down on a cold board.

    Also in this release:

    • In manual fan mode the dashboard shows the fan speed again (it read 0 % while the fans ran at full).
    • The fan texts say when the fans are linked; M1 is labelled "chips" then.
    • /api/system/info also carries NerdOS's stratum.usingFallback, next to AxeOS's isUsingFallbackStratum, so scripts written for NerdOS work unchanged (thanks John11).
    • The NerdOctaxe-γ v2.2 with the 4-phase TPS53647 regulator has now run WarpAxe (at a user's, thanks John11).

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.13 b0fe512dd3

    WarpAxe v0.1.13
    All checks were successful
    CI / Rust checks (push) Successful in 44s
    CI / Regulator scenarios (push) Successful in 46s
    Stable

    dvb-projekt released this 2026-10-01 12:06:37 +00:00 | 496 commits to main since this release

    No more lost chip replies. When WarpAxe read several chip registers in a row — the once-a-second hash-counter round, eight registers per chip — the replies came back faster than the receive buffer was drained, and some were cut off. That is where the empty domains came from, and a cut-off burst could also take a nonce with it, more likely on boards with many chips such as the NerdOctaxe. The reads now go out one per pass and the receive buffer is 16 times larger. On a NerdAxe Gaia: 0 dropped bytes, CRC failures and desyncs, and a 64-register scan answered in full (before: about half). Recommended for every board.

    The BM1373 has eight hash domains. Solo Satoshi describes the BM1373 as eight hashing domains; no firmware read them. AxeOS reads four domain counters and doubles them. A register scan on our Gaia found the other four at 0xAC–0xAF, each counting an eighth of the total. WarpAxe now reads all eight; hashrateMonitor lists eight domains for a BM1373 (four for the BM1370), and the dashboard card shows them in two rows of four.

    Naja Duo: overclock up to 600 MHz. The preset list stays AxeOS's (327–410 MHz); with ?oc free values now go up to 600 MHz, the maker's documented overclock (600 MHz / 1120 mV, about 8 TH/s and 92 W). Untested by us — the Naja Duo has not run WarpAxe here.

    Your miner's name in the browser tab. "NerdAxe Gaia · WarpAxe" instead of "WarpAxe", with the hostname in brackets where it differs ("Bitaxe Gamma (bitaxe602) · WarpAxe"), so several open tabs tell their miners apart.

    Also in this release:

    • For diagnosis, read-only: GET /api/v1/chip/regs?chip=0&from=0x80&to=0xbf reads a range of chip registers, GET /api/v1/chip/cores counts the nonces per core.
    • Measured on the Gaia (BM1373) at 380 MHz: the one-minute error rate is 16.9 % at 930 mV, 6.1 % at 950 mV and 2.5 % at 970 mV, while the chip's own hash counter stays the same — real timing errors that fall steeply with voltage, so the error rate is a usable voltage signal on this chip too.

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.12 e76e9df7c1

    WarpAxe v0.1.12
    All checks were successful
    CI / Rust checks (push) Successful in 44s
    CI / Regulator scenarios (push) Successful in 47s
    Stable

    dvb-projekt released this 2026-10-01 06:41:17 +00:00 | 503 commits to main since this release

    Hash counters: a restarted counter no longer spikes the error rate. In v0.1.11 a chip that restarted its counters — on a NerdOctaxe it happened during a core-voltage step — produced one reading with a huge delta, and errorPercentage jumped to 92 % for a second (the shares were fine). A delta larger than any chip can hash in a second now only resets the baseline, so a restart or a garbled reading costs one second instead of a spike. Update from v0.1.11 recommended if a tool or tuner acts on these values.

    Error rate over a minute. New errorPercentage1m in /api/system/info (and miner.error_percent_1m in /api/v1/state): the error rate over the last full minute of counter deltas, 0 until the first minute is complete. errorPercentage stays AxeOS's per-second value, now clamped. Tuners should read the one-minute value.

    Hash counters card and the BM1373. The BM1373 fills only two of its four domain counters (AxeOS scales them by two). The card now hides a domain that stays at 0 on that chip instead of marking it red, and shows its error rate without a colour: on a NerdAxe Gaia it runs at 5–8 %, against 0.4–1.5 % on a BM1370, and there is no known baseline for this chip yet. The card's header shows the one-minute error rate.

    AxeOS's 0/1 switches. isUsingFallbackStratum, useFallbackStratum, useNTP and autofanspeed are now numbers 0/1 as in AxeOS (they were true/false, which some dashboards showed as blank), and a PATCH with 0/1 for an on/off setting is accepted instead of rejected. Reported by John11, who also runs WarpAxe on a Bitaxe Gamma Turbo now — the second profile to run on hardware for the first time at a user's.

    NerdOctaxe v2.2 and the board choice. A user's NerdOctaxe-γ v2.2 stopped within a minute on v0.1.11 with both Nerd profiles. Most likely cause: WarpAxe stopped the chips after 60 s without a chip temperature, and that board's chip sensor may never answer. The 60-s guard now applies only once a chip temperature has been read; a board whose chip sensor never answers runs as on NerdOS — on the regulator's limit with the fans at full — and says so in the log. The v2.2's 4-phase regulator path is still untested; please keep a way back to NerdOS. Also: while a Nerd board waits to be chosen, the display now runs, so the setup hotspot's name and password show (before, the screen stayed dark and the hotspot could not be joined).

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.11 12538d279b

    WarpAxe v0.1.11
    All checks were successful
    CI / Rust checks (push) Successful in 45s
    CI / Regulator scenarios (push) Successful in 48s
    Stable

    dvb-projekt released this 2026-09-30 19:38:45 +00:00 | 509 commits to main since this release

    Hash counters, as AxeOS reads them. Once a second WarpAxe now reads each chip's own hash counters — the total, the four hash domains and the error counter — and reports them in /api/system/info as AxeOS does: hashrateMonitor (per chip total, domains, errorCount) and errorPercentage. Tools that watch a weak domain or the error rate, such as a watchdog that restarts a miner, work unchanged. On a NerdQAxe++ the counters agree with the hashrate from the nonces within 1–2 %. Asked for by John11, who runs WarpAxe on a Bitaxe Gamma Duo — the first run of that profile.

    Hash counters on the dashboard (off by default). Settings → Dashboard → Hash counters on the dashboard shows a card with one row per chip: its hashrate, its four domains as bars and its error counter, with the error rate in the header. A domain more than 10 % below its chip's mean turns yellow, more than 20 % red. The card shows a 30-second mean, because a single reading dips by up to a third while the chip moves work between domains; the API keeps the per-second values. The API reports the counters whether the card is on or not.

    Crash reports you can fetch. After a crash the miner used to log a one-line summary and erase the core dump. It now keeps the dump until you fetch or erase it: GET /api/v1/coredump returns it as an ELF for espcoredump.py, DELETE erases it, and /api/v1/state shows core_dump_bytes. The summary line in the log no longer cuts off the crash address. Background: a NerdAxe Gaia crashed once when its pool restarted, and there was nothing left to analyse.

    Bitaxe wordmark on the Naja Duo and Gamma Hex. The colour-display Bitaxe boards now show the Bitaxe wordmark under the WarpAxe logo on the boot screen, as AxeOS does, with skot's consent. The small-OLED Gamma boards keep the model name.

    Also in this release:

    • README reviewed top to bottom: every board that runs, BM1373, drivers, API fields, first start, and the makers' consent for the boot-screen wordmarks.
    • New texts reviewed by native speakers in all eight languages.

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.10 49951a8514

    WarpAxe v0.1.10
    All checks were successful
    CI / Rust checks (push) Successful in 43s
    CI / Regulator scenarios (push) Successful in 48s
    Stable

    dvb-projekt released this 2026-09-30 04:39:01 +00:00 | 520 commits to main since this release

    Four more Bitaxe boards: Gamma Duo, Gamma Turbo, Naja Duo, Gamma Hex. Profiles for the Bitaxe Gamma Duo (650), Gamma Turbo (801), Naja Duo (1201) and Gamma Hex (1300), matched against AxeOS's source: pins, regulator window and phases, the EMC2103 fan and temperature chip (new driver), the colour display on the Naja and Hex, and the chip enable line. None of them has run on hardware yet: each is offered on the first start and taken only when you pick it. A Gamma Duo on v0.1.9 was taken for a plain Gamma (its board version 650 lies in the Gamma's 600 range) — v0.1.10 tells them apart. If you have one of these boards and a way back to AxeOS over USB, feedback is very welcome.

    BitForge Nano: Max preset. The Nano's regulator warning line is now 45 A (the part still shuts down at 52 A), so tuners can use its Max preset (625 MHz / 1160 mV, about 37 A at the core). At Max the pool counted 2.53 TH/s ± 4.5 % against 2.19 at Standard.

    NerdAxe boards: quieter fan. Their one fan cools chip and regulator; it now holds the regulator at 80 °C instead of 65 °C, which kept the Gaia's fan at 100 % although its regulator runs at 70–75 °C and shuts down at 105 °C. The dashboard shows the regulator target and names the fan "chips · regulator".

    Also in this release:

    • Dashboard: the BitForge Nano shows as its own series with two chip fans, the regulator is on the type plate of the Bitaxe boards too, and the current is marked "input" or "output" as the board measures it (the Gaia said "output" for its input current).
    • gitHash in /api/system/info now always names the commit the image was built from.
    • New texts reviewed by native speakers in all eight languages.

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.9 f97d001451

    WarpAxe v0.1.9
    All checks were successful
    CI / Rust checks (push) Successful in 45s
    CI / Regulator scenarios (push) Successful in 50s
    Stable

    dvb-projekt released this 2026-09-29 18:30:14 +00:00 | 527 commits to main since this release

    Nerd boards hash on par with their stock firmware. Each board now paces its work as its stock firmware does: new work every 500 ms on the NerdQAxe++, NerdOctaxe and NerdQX (NerdOS), and on the NerdAxe Gaia too. Before, the chips got new work only when their nonce space ran out, about every 22 s on the eight-chip Octaxe, and sat idle part of that time. Measured at the pool over hours: the NerdOctaxe went from 10.79 to 11.59 TH/s at 700 MHz (NerdOS: 11.71, ±2.5 %), the Gaia from 87 % to 102 % of its expected rate at 420 MHz, the NerdQAxe++ from 98 % to 100 %. Earlier posts that called the Nerd boards "on par with NerdOS" were wrong by up to 8 %; with this release they are.

    NerdAxe Gaia runs. One BM1373 on a 12 V TPS546, brought up in BitMaker's order; the chip's 11-byte ID reply is read, the clock ramps before the baud switch, and the board reports its input power (VIN × IIN). At 350 MHz / 940 mV the pool counts 2.30 TH/s, as with its stock firmware. At 420 MHz / 960 mV under full load the core draws 31–33 A against the regulator's 33 A over-current limit; tuners should stay a step below (see below).

    BitForge Nano runs. Two BM1370 on a TPS546 with an INA260 on the 12 V input and one EMC2101 per chip behind a PCA9544; profile from WantClue's open hardware files and the regulator's own settings. It is offered on the first start, never taken on its own. At ForgeOS's 525 MHz / 1150 mV: about 2.2 TH/s at 43–45 W.

    For tuners. /api/system/info now reports coreCurrentA (the regulator's output current) with coreCurrentWarnA (its warning limit), faultCause while a regulator fault has stopped the core (overcurrent, overvoltage, undervoltage, overtemp), and gitHash, the commit the image was built from. power is always a number, 0 W with the core off.

    Also in this release:

    • Display turned 180° on all Nerd boards by default, as their stock firmware does. If you had switched "Flip screen" on to make up for it, switch it off once after the update.
    • The wordmarks of the NerdQAxe++, NerdOctaxe, NerdAxe Gamma, NerdAxe Gaia and NerdQX on their boot screen, with the trade mark holder's consent (solomining.de) and the artwork of shufps and BitMaker.
    • When every chip has been silent for a minute, the chain is started over.
    • After a stop (overheat, regulator fault, chips off) the fans start again at full speed; on the NerdAxe boards the one fan follows the chip or the regulator, whichever is hotter.
    • A first start switches off a core rail the stock firmware left on (an install over USB with the 12 V still connected).
    • Settings are also taken from newer NerdOS's json_cfg.
    • The type plate on the dashboard wraps between its parts on a phone; "phases" is translated.
    • New texts reviewed by native speakers in all eight languages.

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

  • v0.1.8 d06739a688

    WarpAxe v0.1.8
    All checks were successful
    CI / Rust checks (push) Successful in 43s
    CI / Regulator scenarios (push) Successful in 50s
    Stable

    dvb-projekt released this 2026-09-28 23:40:04 +00:00 | 558 commits to main since this release

    See who a block would pay. Every job's coinbase is now read the way AxeOS does it: the Pool page shows the share of a found block that goes to your address — green from 95 %, red below — the coinbase value and each output, yours starred; the dashboard's pool card shows the share as one line. A solo pool shows 100 %; a pool that keeps more than it says shows it here. The API serves AxeOS's coinbaseOutputs, coinbaseValueTotalSatoshis and coinbaseValueUserSatoshis (Stratum V1 and V2).

    Dashboard and Pool page in the warp design. The dashboard opens with one hero like the Swarm page — chip grid, the hashrate large with its share of the expected rate, efficiency, power, shares, best difficulty and the history — and three instruments below: temperature, fans with input, core voltage and clock, and the pool with its connection line (protocol, Noise/TLS, fallback in amber, a cut in red). The Pool page gets the same status hero above its settings.

    A calmer history. A 10 s point counts only a dozen chip results on a Gamma, so single points scattered by a third and the temperature axis zoomed into tenths of a degree. The hashrate line is now a two-minute moving average on an axis of at least ±25 % around the mean, and the temperature axis moves in steps of at least 0.5 °C. Real dips still show; counting noise doesn't.

    Also in this release:

    • The Save/Restart bar floats at the bottom while you scroll through a form.
    • The NerdOCTAXE-γ shows as "NerdOctaxe" on the display and in the web UI (the display font has no γ); its board name nerdoctaxe-gamma for the API and tuners is unchanged.
    • New texts reviewed by native speakers in all eight languages.

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).