-
WarpAxe v0.1.47
Stablereleased this
2026-11-19 13:00:23 +00:00 | 1 commits to main since this releaseClock per chip on the dashboard, Bitaxes WarpAxe cannot run are named before anything gets power, and a silent pool no longer costs minutes. On boards where WarpTune or the owner slows a single chip, the dashboard now shows each chip's clock, and who set the offsets and when. A Bitaxe on another chip (Max, Ultra, Hex, Supra, Supra Hex) or a TinyChipHub Zyber is recognised from the settings its stock firmware left behind, by the firmware and by the installer page, and nothing is switched on or written. A pool that accepts the connection and then sends nothing now costs 30 seconds instead of three minutes per attempt. The image is about 34 KB smaller than v0.1.46. A miner that runs without trouble mines as before.
New.
- Clock per chip on the dashboard. Since v0.1.44 the chips of a BM1370 or BM1373 board with several chips can run at their own clocks (WarpTune uses this to slow down a chip that makes errors before the whole chain has to). Until now that was only visible in the folded "Clock per chip" fields on the settings page. While at least one chip has an offset, the dashboard's power card now reads "Clock (base)" and lists each chip with its clock, the offset in yellow and, where the board measures one per chip (the Naja Duo and the Gamma Turbo), its temperature, together with who set the offsets (by hand, WarpTune or the API) and how long ago. The hash counters show each chip's clock under its name, and "Clock per chip" in the settings opens by itself. With every chip at the chain's clock the pages look as before. API:
chipFreqOffsetsBy,chipFreqOffsetsAt. A change by hand or through the API also goes into the event log. Details in docs/18.
Changed.
- Bitaxes on other chips are named, not started. A Bitaxe Max (BM1397), Ultra or Hex (BM1366), Supra or Supra Hex (BM1368) cannot run WarpAxe. Until now a Supra Hex was offered the Gamma Turbo on its first start, and a Supra or an Ultra was taken for a Gamma, with its chip switched on until the chip's ID refused it. WarpAxe now reads the board model the stock firmware (AxeOS) left behind, before any chip gets power, and the web UI shows "Recognised: Bitaxe Supra Hex (6× BM1368)" with "WarpAxe cannot run this board yet, so its ASICs stay off." If the recognition is wrong, a board can still be chosen by hand. The chip name alone is enough, so a board version AxeOS added later (a Supra Hex 703) is caught as well and shown as "Bitaxe (BM1368)". TinyChipHub's Zyber boards (Zyber8S, 8G, 8G+, 12G), which run an AxeOS fork and look like a Supra Hex on the bus, are named and stopped the same way, by the model their factory settings store.
- The installer page checks the board first. Before the backup and before anything is written, the browser installer reads the same AxeOS settings from the miner. On one of these boards it stops with "This miner is a Bitaxe Supra Hex (6× BM1368), which WarpAxe cannot run. Nothing was written; its firmware stays as it was." and restarts the miner into its own firmware. A miner without such settings installs as before. If reading the flash fails, the install stops before anything is written.
What is fixed.
- A pool that accepts the connection and then says nothing no longer holds the miner. Until now such a pool (a proxy whose backend is gone, or the IPv6 side of a pool that only works over IPv4) kept each attempt for three minutes, three times before the fallback took over. Until the first job arrives, a connection now ends after 30 seconds without data from the pool; once work arrives, quiet periods of up to three minutes are fine as before.
- IPv4 first after a silent IPv6 connection. A pool that sent no work over IPv6 is tried over IPv4 first for the next hour, with IPv6 still in the race behind it in case IPv4 does not answer. Changing the pool settings resets this.
- The fallback is kept while the main pool sends no work. While the miner is on the fallback, a main pool that accepts the connection but sends no work no longer pulls it away every five minutes. It gets one try (a second one at once over IPv4 if IPv6 stayed silent), then the miner goes back to the fallback and waits twice as long before the next check: 10, 20, then 40 minutes. Work from the main pool sets the wait back to five minutes.
- A readable timeout in the log. When a pool goes quiet, the log now says "no data from the pool for 30 s" (or 180 s) instead of "No more processes (os error 11)".
For developers.
- Smaller formatting code. The firmware now formats text through newlib's nano printf (
CONFIG_NEWLIB_NANO_FORMAT), 44.8 KB less flash; the stacks that ESP-IDF and the Wi-Fi library shrink with that option are kept at their old sizes. A few ESP-IDF and Wi-Fi log lines with 64-bit or size_t values (%llu,%zu) now print letters such asluinstead of the number; floating-point numbers still print. WarpAxe's own log lines and every value in the API are unaffected. The list is in docs/16. - Installer:
?backuponly(backup only, nothing written) now says so when it finishes, and the browser console logs the board check with what it read.
Image. About 34 KB smaller than v0.1.46 (34,368 bytes): 44.8 KB less for nano printf; the board recognition, the clock per chip and the silent-pool handling take about 10 KB of that back.
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Source: git.warppool.org/dvb-projekt/dvb-WarpAxe · Mirror: github.com/dvb-projekt/dvb-WarpAxe
Downloads
- Clock per chip on the dashboard. Since v0.1.44 the chips of a BM1370 or BM1373 board with several chips can run at their own clocks (WarpTune uses this to slow down a chip that makes errors before the whole chain has to). Until now that was only visible in the folded "Clock per chip" fields on the settings page. While at least one chip has an offset, the dashboard's power card now reads "Clock (base)" and lists each chip with its clock, the offset in yellow and, where the board measures one per chip (the Naja Duo and the Gamma Turbo), its temperature, together with who set the offsets (by hand, WarpTune or the API) and how long ago. The hash counters show each chip's clock under its name, and "Clock per chip" in the settings opens by itself. With every chip at the chain's clock the pages look as before. API:
-
WarpAxe v0.1.46
Stablereleased this
2026-11-18 13:42:56 +00:00 | 19 commits to main since this releaseIPv6 for pool connections, drivers you can reuse, and a fan PWM fix on the Nerd boards. A miner on a network with IPv6 now reaches pools that offer it over IPv6, with IPv4 at most a quarter second behind; on a network without IPv6 the only change is that the first pool connection after joining Wi-Fi waits up to 4 seconds for an IPv6 address. Several of WarpAxe's hardware drivers now live in their own crate under MIT/Apache, so other firmware can use them, and all drivers use embedded-hal 1.0. The EMC2101 (Bitaxe Gamma and Gamma Duo, NerdAxe Gamma and Gaia, BitForge Nano) now runs on the community crate emc2101-rs. Rewriting the EMC2302 driver from its datasheet turned up a wrong bit position inherited from NerdOS: on the Nerd boards the main fan now gets the 19.53 kHz PWM that was always intended, instead of 4.88 kHz. A miner that runs without trouble mines as before.
New.
- IPv6 for pool connections. The miner takes an IPv6 address from the router, and a pool that has both an IPv6 and an IPv4 address is tried over IPv6 first, with IPv4 starting a quarter second later if IPv6 has not answered by then, or at once if it was refused (Happy Eyeballs, RFC 8305); whichever answers first is used. A pool whose IPv6 does not answer costs a quarter second, not a timeout. A pool address may be an IPv6 literal, with or without brackets. A switch on the Network page turns it off (on by default); under it the miner's IPv6 addresses, and the pool cards show whether the connection runs over IPv4 or IPv6. Over IPv6 the web UI and API answer only the local network (the link, unique-local addresses and the miner's own /64): a global IPv6 address has no NAT in front of it. Tested on a Bitaxe Gamma on a native dual-stack line with dual-stack and IPv4-only pools, a pool whose IPv6 refuses connections, TLS over both, Stratum V2 over IPv6, and IPv6 literals and local names; a route that drops IPv6 silently is not tested on hardware. API:
ipv6,ipv6Addresses,poolAddress,poolAddressFamily,ipv6Refused. Details in docs/17. - Note on IPv6 and privacy. The miner's IPv6 address is built from its MAC (lwIP has no privacy addresses), so IPv6 pools see the MAC. Turn the switch off if you mind.
Changed.
- Nerd boards: main fan PWM at 19.53 kHz. On the NerdQAxe++, NerdOctaxe and the other boards with the EMC2302 fan controller, the fan on the ASIC cooler ran at 4.88 kHz; the setting meant 19.53 kHz for both fan outputs, but put the second output's bits one place too far (EMC2302 datasheet, register 6-10). Measured on a NerdQAxe++ before and after: the same speed at 100 % (2820 / 2818 rpm) and at 25 % (790 / 791 rpm), and the fan control holds the chip at its target as before. Reported to NerdOS (shufps/ESP-Miner-NerdQAxePlus#767).
- EMC2101 through emc2101-rs. On the Bitaxe Gamma and Gamma Duo, the NerdAxe Gamma and Gaia and the BitForge Nano (two EMC2101), the fan and temperature chip is now driven by the emc2101 crate 4.2 (GPTechinno). The board's diode settings stay as they were (Gamma and Gaia 0x24, Nano 0x37), as does the PWM frequency (about 7.8 kHz). Before the switch we reported a division by zero on a zero tach count and two settings it could not make (diode ideality/beta, PWM clock); all three were fixed and released the same day. A later suggestion of ours is in 4.2: the fan speed is written without reading the fan configuration back each time, so a misread byte can no longer change the fan's polarity or PWM clock. Checked on a Gamma, a Gaia and a Nano: the same temperatures and fan speeds as before.
- Night mode, as the settings now describe it. The settings texts said night mode shows only the clock. It shows the dimmed clock screen instead of the pages: on the colour displays the time with date, hashrate and temperature, on the OLED its clock slide. Corrected in all eight languages.
- Network page: which changes need a restart. The note above the network settings said every change takes effect after a restart. It now says that Wi-Fi, hostname, USB address and NTP need a restart, while the IPv6 switch and the time zone apply at once. Changed in all eight languages.
What is fixed.
- IPv6 after the setup hotspot. When the setup hotspot opened or closed, the next Wi-Fi connection skipped the IPv6 start, and the miner stayed on IPv4 until the link dropped for another reason. It now starts IPv6 again with every new connection.
- No extra second with IPv6 off. With the IPv6 switch off, every Wi-Fi connection waited one second for an address that never came.
- TLS pools: one 10-second budget. Choosing between a pool's IPv6 and IPv4 address and the TLS handshake now share one 10-second budget, with at least 5 seconds always kept for the handshake. Before, each had its own 10 seconds.
- IPv6 address in the pool card. The "Server" row writes an IPv6 address in brackets, so the port stays readable.
- EMC2101 boards: the chip is identified first. On the Bitaxe Gamma and Gamma Duo, the NerdAxe Gamma and Gaia and the BitForge Nano, the fan and temperature chip is identified again before anything is written to it, as in v0.1.45 and earlier. With a wrong board choice, nothing is written to another chip at the same address.
- Nerd boards: no doubled fan speed reading. If setting up the EMC2302 fan chip failed after a warm restart, the displayed fan speed could read twice the real value. Only the reading was wrong; the fans themselves were unaffected. The driver now uses the right range from the start.
For developers.
warpaxe-drivers, MIT OR Apache-2.0. The drivers that follow no other firmware's code now live incrates/driversunder MIT OR Apache-2.0: EMC2302 (rewritten from the datasheet), INA260, TMP1075, PCA9544, SSD1306 with its 5×7 font, the PMBus number formats and the NTC thermistor. The rest of WarpAxe stays GPL-3.0-or-later, including the drivers that partly follow ESP-Miner or NerdOS (TPS546, TPS53647, EMC2103, FXL6408), and the EMC2101 board adapter, whose register code is the emc2101 crate's (MIT OR Apache-2.0).- embedded-hal 1.0. All drivers use embedded-hal 1.0's I²C trait instead of WarpAxe's own. The bytes on the bus are unchanged.
- GitHub mirror. The repository is mirrored at github.com/dvb-projekt/dvb-WarpAxe; issues and pull requests are welcome there. Releases and the devices' online updates still come from git.warppool.org.
- Reported upstream: ESP-Miner #2106 (BM1373 single-chip address), tps546 #1 and #2, bm13xx-rs #10 (BM1373 notes), emc2101-rs #5, #6 and #7, NerdOS #767.
Image. About 20 KB larger than v0.1.45 (19,952 bytes): about 12 KB for IPv6, about 5 KB for the emc2101 crate and embedded-hal, the rest for the fixes.
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Source: git.warppool.org/dvb-projekt/dvb-WarpAxe · Mirror: github.com/dvb-projekt/dvb-WarpAxe
Downloads
- IPv6 for pool connections. The miner takes an IPv6 address from the router, and a pool that has both an IPv6 and an IPv4 address is tried over IPv6 first, with IPv4 starting a quarter second later if IPv6 has not answered by then, or at once if it was refused (Happy Eyeballs, RFC 8305); whichever answers first is used. A pool whose IPv6 does not answer costs a quarter second, not a timeout. A pool address may be an IPv6 literal, with or without brackets. A switch on the Network page turns it off (on by default); under it the miner's IPv6 addresses, and the pool cards show whether the connection runs over IPv4 or IPv6. Over IPv6 the web UI and API answer only the local network (the link, unique-local addresses and the miner's own /64): a global IPv6 address has no NAT in front of it. Tested on a Bitaxe Gamma on a native dual-stack line with dual-stack and IPv4-only pools, a pool whose IPv6 refuses connections, TLS over both, Stratum V2 over IPv6, and IPv6 literals and local names; a route that drops IPv6 silently is not tested on hardware. API:
-
WarpAxe v0.1.45
Stablereleased this
2026-11-16 05:47:52 +00:00 | 49 commits to main since this releaseA clock and a night mode on the display, the Naja Duo at its full power, and a power-supply sag watch. The display can show the time, and dims itself at night. The OLED moves its picture a pixel every ten minutes, so nothing burns in. On the Naja Duo the boot key now steps through the display pages, its current and power limits follow the maker's overclock, and WarpTune can take it up to 550 MHz. A miner that runs without trouble mines as before; the clock and the night mode are off until you switch them on.
New.
- A clock page and night mode. The colour displays get a page "clock": the time large, with weekday, date, week and hashrate beside it, and below it the block height with the price and fee when market data is on, else the pool and the uptime. The OLED gets a clock slide with time, short date and hashrate. Neither is among the factory pages; you add them on the Settings page under Pages (on the OLED: OLED slides), and night mode under Clock and night mode. In night mode (off by default, 22:00 to 07:00, brightness 50 %) the display shows a dimmed clock screen instead of the pages: on the colour displays the time, date, hashrate and temperature, on the OLED its clock slide. A key press brings the full display back for 30 seconds, and an alert (a found block, an overheat stop, an update) lights it at full brightness. Time format: 24 hours, or 12 hours with AM/PM in time zones that use it (
clockFormatauto/24h/12h). API:nightMode,nightStart,nightEnd,nightBrightness,nightActive,clockFormat,clock12h. - Night brightness in even steps. On the OLED (Bitaxe Gamma and others with the 128×32 screen), contrast, pre-charge and VCOMH go down together, so 100 %, 75 % and 50 % look clearly different and a low night setting is really dark. On the NerdQAxe++, NerdOctaxe and the other Nerd boards, the backlight driver (AW9364) does not dim by PWM: it counts pulses. The firmware sends the pulses, and the brightness picks one of its 16 levels. The Naja Duo and Gamma Hex dim by PWM.
- OLED: protection against burn-in. The picture moves by one pixel every ten minutes, through twelve positions (at most one pixel left or right, one up or two down). All OLED screens leave room for it, so nothing is cut off; the first ten minutes after a start stay unshifted.
- Naja Duo and Gamma Hex: the boot key steps the display pages. These boards have no page key. A short press now goes to the next page, holding it for a second goes to the home page. The press that wakes the display does nothing else.
- Power supply sag watch. The firmware compares the input voltage with the chips off against the voltage while hashing. If it drops by 8 % of the nominal input or more for a full minute, the System page shows a warning with the drop in volts (gone again below 6 %). It only reports; nothing is switched off. API:
vinIdle,vinLoad,vinSag,psuSagWarning. - Hint for a block that is not Bitcoin's. If the pool reports a found block below height 900,000 while the coin setting says BTC (a pool for another coin, with the ticker left on BTC), the block banner shows the coin as "?", leaves out the mempool link and suggests checking the Coin setting.
Naja Duo.
- Current limits 61/68 A (warning/fault), as on ESP-Miner's master for the maker's 600 MHz overclock, and
maxPower105 W instead of 60 W. The firmware only reportsmaxPower; WarpTune takes 90 % of it (94.5 W) as its power limit, and at 60 W it skipped the Max step before measuring anything. - WarpTune up to 550 MHz at 1120 mV. The clock options now go up to 550 MHz and the core voltage up to 1120 mV, the voltage of the maker's own overclock; WarpTune's Max mode stops there. On our bench, 550 MHz needed about 1100 mV on the weaker chip, and at 600 MHz the regulator on the back reached 84 °C without a fan there. A core voltage above 1120 mV is no longer accepted; a higher value stored by an earlier version is set to 1120 mV at start. With the factory target temperature of 50 °C, the fan runs close to full speed at 550 MHz. In OC mode the clock can still be set up to 600 MHz by hand.
- Expected hashrate. Both chips' counters read about 6,900 small cores, as on the Gaia; the expected hashrate is now based on the chip's 6,860 instead of ESP-Miner's 6,725, which put it 2.5 % low (reported as ESP-Miner issue #2102).
- VOUT up to 2 V per voltage domain. On boards with two domains in series (Naja Duo, Gamma Hex, NerdQAxe++ Rev 7) a core voltage above 1000 mV read over 2 V and was dropped, so VOUT, the core voltage and the power were missing.
What is fixed.
- Settings form on a Bitaxe with an old chip stop. v0.1.32 lowered the Bitaxe chip stop ceiling to 75 °C. A Bitaxe that still had a higher chip stop stored from before could not save its settings: the browser refused the whole form. The stored value is now brought down to the ceiling at start-up, and the form saves again. The miner already stopped at 75 °C before.
- A pool session that cannot send shares ends. The thread that sends shares now runs on a PSRAM stack. Before, with little internal memory left, it could fail to start, and the session took work and hashed without ever sending a share. If it still cannot start, the session ends, so the retry and the fallback pool take over.
- Error rate colours on the BM1373. The Gaia and the Naja Duo showed their error rate without a colour. Green is now below 5 %, yellow from 5 %, red from 9 %; the BM1370 keeps 1 % and 5 %.
- NerdNOS: the core voltage as it is. On the NerdNOS, whose core rail is fixed, the API and the self-test reported the stored but unused setting (1150 mV); they now report the board's 1400 mV.
- One update progress bar. The System page showed two bars during an update; now it shows one.
Smaller image. About 185 KB smaller than v0.1.44: the size work took out 232 KB (start-up pictures packed, API documents built in one call each, debug lines left out), and the clock and night mode added some of it back. This leaves room for future updates.
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Downloads
- A clock page and night mode. The colour displays get a page "clock": the time large, with weekday, date, week and hashrate beside it, and below it the block height with the price and fee when market data is on, else the pool and the uptime. The OLED gets a clock slide with time, short date and hashrate. Neither is among the factory pages; you add them on the Settings page under Pages (on the OLED: OLED slides), and night mode under Clock and night mode. In night mode (off by default, 22:00 to 07:00, brightness 50 %) the display shows a dimmed clock screen instead of the pages: on the colour displays the time, date, hashrate and temperature, on the OLED its clock slide. A key press brings the full display back for 30 seconds, and an alert (a found block, an overheat stop, an update) lights it at full brightness. Time format: 24 hours, or 12 hours with AM/PM in time zones that use it (
-
WarpAxe v0.1.44
Stablereleased this
2026-11-12 21:04:45 +00:00 | 95 commits to main since this releaseA clock of its own per chip, the Bitaxe Naja Duo on our bench, and more internal memory. On boards with more than one chip, each chip can now run at its own clock, so the weakest chip no longer sets the voltage for all of them. The Naja Duo now runs on our bench, and its second chip now gets a setting that its init never delivered. A single-chip miner with no trouble mines as before.
New.
- A clock of its own per chip (
chipFreqOffsets, MHz added to the frequency per chip). The firmware writes each chip's clock to that chip alone, without restarting the chain, and no chip passes its own target clock on the way. The expected hashrate, the self-test and the watchdogs follow each chip's clock. Limits: no chip above the board's maximum or below half the base clock. On boards whose chips sit in series (Naja Duo, Gamma Hex, NerdQAxe++ Rev 7) the clocks may lie at most 25 MHz apart (or 8 % of the base clock, where that is less), because a slowed chip takes voltage from the other. The web interface has a field per chip under frequency and voltage. For now the offsets are set by hand; an automatic search in WarpTune is planned for a later WarpPool release. Tested on a NerdOctaxe at 625 MHz: one chip set 25 MHz lower hashed about 4 % less, the other seven were unchanged. - New API fields for tuners:
asicFrequencies(each chip's clock),chipFreqOffsetsSupported,voltageDomains,chipFreqOffsetSpread, and per chiphashCount(a cumulative hash counter, so a tuner gets each chip's hashrate from two readings) andcounterRestarts(a chip that reset itself).minerStackFreereports the mining loop's spare stack. - Bitaxe Naja Duo on our bench. It has run on our bench since 12 November; the web interface still marks it untested until it has run for some days. It uses ESP-Miner's BM1373 init by default, with a slower variant and NerdOS's init as fallbacks (
chipInit). Its 320×170 colour screen shows the Bitaxe logo as printed on the board, with "1201" at boot, and the self-test reads both phases of its stacked regulator. Found on the way: the init sent the second chip's domain-end IO strength (register 0x58) to an address no chip has (0x10 shifted right by two, as in ESP-Miner), so the second chip never got it. Single-chip writes on the BM1373 now go to the chip's full address; reported to ESP-Miner as issue #2094.
What is fixed.
- Two-chip boards: no false overheat stop before the chips run. On a user's Gamma Turbo, chip 1's sensor read 72.75 °C eleven seconds after power-on, before the chips were running. That is over the 70 °C stop line, so the start stopped and the chips stayed off for a minute. Until the chips have power and the chain has answered after a reset or power-on, the chip sensors of the Gamma Turbo and the Naja Duo now count as no reading. The regulator's own sensor keeps watching throughout, and once the chips run the stop line applies as before.
- Back on after a regulator fault. On boards whose chip sensor gives nothing while the chips are off (Bitaxe Gamma, the two-chip boards), the chips stayed off after a regulator fault until someone switched them on. They now come back after the minute off with the regulator cool, as after an overheat stop; a third stop within the hour still latches.
- Self-test phase check. A phase reading taken while the chips draw no load (all phases together under 5 A) is skipped instead of switching the chips off.
- Display orientation on import. A Naja Duo or Gamma Hex coming from AxeOS 2.15 showed its screen upside down; the AxeOS rotation is now taken per display type.
- More internal memory. The mining loop's large state moved to PSRAM and its stack shrank, so more internal RAM stays free than under v0.1.43 (measured on a NerdAxe Gaia on a TLS pool: about 30 KB instead of 25 KB).
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Downloads
- A clock of its own per chip (
-
WarpAxe v0.1.43
Stablereleased this
2026-11-12 06:31:04 +00:00 | 123 commits to main since this releaseThe Gamma Turbo starts without a false overheat stop, and the fallback pool behaves. A fix release for what a user's Gamma Turbo showed after v0.1.42, plus two fallback-pool fixes that apply to every board. A miner on a single-chip board that has no trouble with its pools mines as before.
What is fixed.
- Two-chip boards: no overheat stop right after the chips start. Since v0.1.42 WarpAxe reads both chips' temperatures on the Gamma Turbo and the Naja Duo. On a user's Gamma Turbo this stopped the chips on every boot. While the chips are off or in reset, the second chip's diode reads the sensor's ceiling (127.875 °C), and the EMC2103 averages each diode over four conversions. So for about a second after the start, its reading still includes those ceiling values: 108 °C, then 80 °C, while the chip was really at about 50 °C. WarpAxe stopped on the first of these readings. AxeOS never switches the rail off at boot, so it never sees them. Now a second diode over the stop line in the first three readings after a chip start (rail on, reset, chain init) is held: the fans run at full and the chips keep running. If it is still over the line after that, the chips stop as before. The first chip's diode keeps its stop on the first reading. Tested on that user's Gamma Turbo: the second chip read 98.75 °C right after the start and about two seconds later both chips read 40.5 and 41.25 °C, with no stop.
- A diode stuck at the ceiling. A diode that reads 127.875 °C, the sensor's highest raw value, now counts as no reading, never as a temperature. As in v0.1.42 for a diode that gives no reading, the fans go to full and the chip with the working diode keeps the stop line.
- Shares on the fallback pool name the fallback's worker. The Stratum V1
mining.submitalways sent the main pool's worker name, also while the miner was on the fallback. A pool that checks the name (ckpool answers "Worker mismatch") rejected every share whenever the two worker names differed. Seen on a user's miner: 12 of 12 shares rejected on the fallback. This bug is older than v0.1.42. - No switch to the fallback before Wi-Fi is up. After a cold boot with a slow access point, the miner's first pool attempts failed for lack of Wi-Fi, counted as three failures, and the miner went to the fallback pool. Attempts made while Wi-Fi is not connected no longer count toward that switch. The check that switches the ASICs off when no pool answers is unchanged.
- Diagnostics file. MAC addresses (the device's own and the access point's BSSID) are now masked down to the maker's prefix. Before, they were in the file unmasked.
- Web interface. While the chain has not counted its chips yet (for example after an overheat stop before the chain start), the name plate and the chip tile show the board's chip count instead of "× 1". API:
boardAsicCountin/api/system/info.
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Downloads
-
WarpAxe v0.1.42
Stablereleased this
2026-11-11 08:34:56 +00:00 | 129 commits to main since this releaseASICs off while no pool answers, a lower voltage waits for the clock, and a found block shows its status. A healthy miner mines as before. When neither the main pool nor the fallback answers, WarpAxe now switches its ASICs off and switches them back on with the first job. A lower core voltage now waits until the clock has come down. A found block keeps its hash and shows whether the chain built on it. The Bitaxe Naja Duo and Gamma Turbo now read both chips' temperatures. A hash domain that is dead right after the chain starts is healed after about three minutes.
What is new.
- ASICs off while no pool answers. Until now a miner without a pool hashed for nothing at full power. When the main pool and the fallback pool (where one is set) have both failed, with at least two attempts on each and at least 60 seconds since the first failure, the core voltage goes off, the ASICs are held in reset and the fans go to their lowest duty. A short Wi-Fi or pool blip switches nothing. Both pools are then probed in turn every 30 seconds, and the first job from either brings the ASICs back with the same clock and voltage. The same applies when no pool or no worker is set up. API:
poolsUnavailable(the name AxeOS and ForgeOS use),chainStopped,chainPhaseNOPOOL; the dashboard shows a notice, the display shows "NO POOL", and the event log has one entry for going off and one for coming back. Nothing is written to flash per attempt. This does not apply while a self-test runs or during the first start of a newly chosen board. Tested on a Bitaxe Gamma: with both pools gone the ASICs went off after about 110 seconds and the miner drew 5 W; when a pool came back, it was mining again on its own within about 10 seconds. The README comparison table now lists this row as "yes" for WarpAxe, next to AxeOS and ForgeOS. - Changed pool behaviour. A pool session that never sends a job now counts as a failed attempt, however long it stays open; this also counts toward the switch to the fallback. After two failures on the fallback, the miner now tries both pools in turn instead of staying on the fallback.
- A lower voltage waits for the clock. A
PATCH /api/systemwith a lower clock and a lower core voltage used to set the voltage within a tenth of a second while the clock came down in steps (625 to 425 MHz takes 3.2 s), so the chips ran at the old clock on the new voltage for those seconds. The voltage drop now waits until the chips run at the new clock (30 s at most). A higher voltage is set at once, as before. The API now refuses a clock below 50 MHz and one the PLL has no setting for; a stored clock of that kind is replaced by the board's default at boot. Seen on a NerdAxe Gaia going from 380 MHz / 960 mV to 350 MHz / 950 mV: the voltage stayed up (955 mV read) until the chips were at 350 MHz, then went down, as the log lines show. WarpTune v1.48.0 already sends the clock first and the voltage after it. - Found-block status. A found block is now stored with its hash. Its status is
pendingat the find,confirmedonce the pool's next work builds on it,orphanedwhen the work for the next height builds on another block, andunknownwhen the miner was offline or without a pool for more than one block. The banner shows the status and the shortened hash, for Bitcoin with a mempool.space link, and shows again when the status changes; the event log has an entry for each status. API:lastBlock.hash,lastBlock.status(null for a block found before v0.1.42). The find itself is now checked against the exact target of the block header. The block count is kept in the form older versions read, with hash and status in a second record, so a downgrade keeps the count. On a test chain at minimum difficulty, where every nonce meets the target, a block is stored once instead of hundreds of times a second. Not tested live, since that needs a found block; covered by unit tests with real block headers. - Both chips' temperatures on the Naja Duo and the Gamma Turbo. These boards have a diode per chip on the EMC2103, and only the first was read, so the second chip had neither a fan control nor an overheat stop. Fans and the overheat stop now follow the hotter chip, as ESP-Miner does. API:
tempis the hotter chip,temp2the second chip's own (-1 without a reading, as on AxeOS),asicTempsboth. A diode that read before and fails counts as unknown, never as cool: fans at full and the lost-sensor stop after a minute. A diode that never reads after boot sets the fans to 100 % after a minute and shows a notice; the chips keep running. We have not run this on such a board yet. If you have a Gamma Turbo, please send ustemp2andasicTempsfrom/api/system/info. - Faster heal for a domain that is dead from the start. v0.1.41 heals a hash domain that stays under 25 % of its sibling domains for ten minutes. A domain under 5 % of its siblings in the chain's first window is now healed after about three minutes. A domain between 5 and 25 %, or one that dies later, still has the ten minutes. The other rules stay: at most one intervention an hour, given up after three resets that do not hold. On a user's Bitaxe Gamma the ten-minute heal of v0.1.41 has worked twice: a domain at 0 % right after an update was healed, and all four domains were fine after. In both cases the clock was well above stock. The three-minute case is tested in the chip simulator.
- Self-test without pool work. Without work from a pool the self-test now marks the hash checks (hashrate, domains, error rate) as skipped with the note "no work from the pool", not failed, and checks the rest: chip count, voltages, regulator, fan, temperatures. The self-test may switch the ASICs on while no pool answers, so a bench test works before a pool is set up.
- The board in the Stratum V1 user agent.
mining.subscribenow sendswarpaxe/0.1.42 (nerdaxe-gaia): the board's profile name in brackets, the same name a Stratum V2 connection already sends, so a pool can tell your boards apart. Until a board is settled (first start not yet confirmed) the agent stayswarpaxe/0.1.42. A pool that sorts miners by user agent may now place a WarpAxe board in its Bitaxe or NerdNOS class. - Diagnostics file. A worker's bare address is now masked too; it showed in the coinbase outputs. A numeric pool password such as "1000" no longer drops the whole info section.
- Fixed during testing. With the ASICs off on purpose (no pool, or switched off by hand), a temperature diode that gives no reading no longer trips the overheat stop. After a real overheat stop the miner restarts once the regulator has cooled and a minute has passed, as AxeOS does.
- Web interface.
- The footer under the swarm list now says what your browser asks for and what your WarpAxe reads itself (an Avalon over the cgminer API, port 4028), and that only addresses are stored.
- A device found in the network search counts as a miner only when it reports a numeric
hashRate. - "Chips off" now reads "ASICs off" wherever the hash chips as a group are meant; the rest of the board keeps running.
- Event log entries added since v0.1.40 are now translated in all languages; they showed in English before.
- WarpTune. WarpTune v1.48.0 reads
poolsUnavailableand neither heals nor steps down while no pool answers.
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Downloads
- ASICs off while no pool answers. Until now a miner without a pool hashed for nothing at full power. When the main pool and the fallback pool (where one is set) have both failed, with at least two attempts on each and at least 60 seconds since the first failure, the core voltage goes off, the ASICs are held in reset and the fans go to their lowest duty. A short Wi-Fi or pool blip switches nothing. Both pools are then probed in turn every 30 seconds, and the first job from either brings the ASICs back with the same clock and voltage. The same applies when no pool or no worker is set up. API:
-
WarpAxe v0.1.41
Stablereleased this
2026-11-10 06:56:39 +00:00 | 155 commits to main since this releaseTLS to a pool on your own node, a stuck hash domain is healed, and long pool messages are handled. A healthy miner mines as before. This release lets WarpAxe connect over TLS to a pool with a self-signed certificate, such as WarpPool on Umbrel, once you trust its key. The firmware now heals a hash domain that stops counting, and WarpTune knows what the firmware handles. Pools that pay many miners in the coinbase no longer fill the miner's internal memory. Certificate dates are now checked once NTP has set the clock, and the last eMudhra root is gone from the trust store.
What is new.
- TLS to a pool with a self-signed certificate. WarpPool on Umbrel creates its own certificate, and until now WarpAxe could only reach it without TLS. When a pool's certificate is not from a public authority, the Pool page now shows the fingerprint of the pool's key (SHA-256 of its public key,
AB:CD:…) and asks whether to trust it. WarpPool v1.48.0 shows the same value in its Connect window to compare. After the click the miner connects with that key only, also by IP address. If the key changes later, the miner does not connect and says so, and a new key needs a new click. "Forget" removes the trust. A pool with a certificate from a public authority keeps going through the certificate store as before. Main and fallback pool each keep their own key, and a saved pool takes its key along. API:stratumTLSPin,stratumTLSPending,stratumTLSKeyChanged,stratumTLSPinned(and thefallbackStratum…ones);PATCH /api/system {"stratumTLSPin": "…"}trusts a key,""forgets it. Tested on a NerdAxe Gaia against WarpPool on Umbrel, by name and by IP address, plus a key change against a local test server; that a saved pool keeps its key is covered by unit tests only. - A stuck hash domain is healed. A BM1370 or BM1373 counts its work in four or eight domains. One domain can stop counting while the rest of the chip hashes. If a domain stays below a quarter of the average of its sibling domains for ten minutes while work goes out, the miner switches its chips off and on. This takes about six seconds; clock and voltage stay the same. At most one such restart an hour, shared with the restart for a silent chip. A domain that sticks again within an hour after each of three restarts is left alone until the next boot or a new clock or voltage. API:
domainLatchResets,lastDomainLatchReset {chip, domain, percent, uptime_s},domainsGivenUp; the event log shows each one. Tested only in the chip simulator: a domain at 3 % is found, one at 50 % is left alone. We have not seen a real case yet. - WarpTune knows what the firmware handles.
/api/system/infolistsselfHeals(chain_silent,chip_silent,domain_latch), and WarpTune v1.48.0 waits for the firmware in those cases instead of acting itself. WarpTune also tells the miner that it is regulating it (POST /api/v1/regulator {"ttl_s": 600}, shown asexternalRegulator {active, expires_in_s}). This lives in RAM only, writes nothing to flash and does not wake the display. - What WarpAxe heals by itself, now in three cases. If the whole chain goes silent for a minute, the miner restarts it. If one chip stays silent for ten minutes while work goes out, the miner restarts its chain (since v0.1.40; on our NerdOCTAXE-γ one of eight chips had stayed dark for almost two days before that). New in v0.1.41: if one hash domain stays far below the others for ten minutes, the miner switches its chips off and on (tested in the simulator only so far). The two newer cases share their rules: at most one restart an hour; after three tries that do not hold, the miner gives up until the next boot or a new clock or voltage; and the timers stop while the pool is away. WarpTune v1.48.0 reads
selfHealsand leaves these cases to the firmware, so it no longer turns the clock down for a silent chip or a stuck domain. For miners on their original firmware, WarpTune handles the silent chip itself, but not domains. - Pools that pay many miners in the coinbase. Some pools pay their miners directly in the coinbase, which gives a pool message of tens or hundreds of KB (ESP-Miner issue #2081). WarpAxe now decodes the coinbase once per job instead of copying it into every piece of work for the chips, so the scarce internal RAM is no longer filled with copies of it. The payout panel lists at most 32 outputs: yours, the SegWit checksum, then the largest, with the rest summed in one row. Until now a coinbase with more than 64 outputs showed no payout at all. A pool line over 512 KB ends the connection and the miner reconnects (
stratumLinesTooLong), instead of reading until the memory is gone. Stratum V2 accepts messages up to 64 KB instead of 8 KB. Tested against a test pool sending 110 to 3,000 outputs (lines up to 186 KB): shares kept flowing and 25 to 45 KB of internal RAM stayed free. An endless line led to a reconnect in under a second. - Certificates: the last eMudhra root removed, dates checked. WarpAxe now also leaves out
emSign Root CA - G1, the one eMudhra root still in Mozilla's store after the October 2026 incident, so the firmware trusts 117 roots (curl's extract of the Mozilla store of 20 October 2026, minus this one). No pool, update or price server the firmware talks to chains to it. The firmware now also checks each certificate's dates: once NTP has set the clock after a boot, an expired certificate, or one not valid yet, is refused. Until then the dates are not checked, so a miner whose clock is not set yet still connects. In the first five minutes after a boot, a TLS pool connection waits up to 15 seconds for NTP. A pool trusted by its key does not depend on dates. Tested on an ESP32-S3 devkit: a server with an expired certificate was refused; valid servers, the update check and the price fetch kept working. On a NerdAxe Gaia the clock was set 13 seconds after the boot and the pinned TLS connection to WarpPool on Umbrel was up at 17 seconds. - Web interface.
- During an online update the card on the System page shows the same progress as the bar at the top; it stayed at 0 % before (thanks to John11).
- On the Pool page the connection card updates by itself after a save, without a reload.
- Saving pool settings no longer says a restart is needed: pool changes apply at once.
- The witness-commitment row in the payout panel now reads "SegWit checksum (no amount)".
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Downloads
- TLS to a pool with a self-signed certificate. WarpPool on Umbrel creates its own certificate, and until now WarpAxe could only reach it without TLS. When a pool's certificate is not from a public authority, the Pool page now shows the fingerprint of the pool's key (SHA-256 of its public key,
-
WarpAxe v0.1.40
Stablereleased this
2026-11-04 12:16:33 +00:00 | 181 commits to main since this releaseOne silent chip no longer stays dark until the next restart, and TLS trusts a pinned certificate store. This release changes nothing in how a healthy miner mines. It brings a chip back that stops answering mid-run, and it checks every TLS connection against a fresh certificate store after the eMudhra incident.
What is new.
- A chip that stops answering is brought back. Until now the miner restarted its chips only when all of them had gone silent for a minute. One chip that dropped out while the others kept hashing stayed dark until the next restart: on our NerdOCTAXE-γ one of eight chips stopped after five days at a steady clock, and the miner ran at 7/8 for almost two days. Now a chip silent for ten minutes while work goes out makes the miner start its chips over, which takes a few seconds. At most one such restart per hour. A chip that goes silent again within an hour of each of three restarts is left alone until the next boot or until the clock or voltage changes. A pool outage stops both timers, so it never counts. The event log shows each restart, and the API reports
chipSilentRestarts,lastChipSilentRestart(chip from 0, mV, uptime) andchipsGivenUpin/api/system/info;chainSilentRestartskeeps counting only the all-chips case. WarpTune v1.47.0 reads these: it judges the hashrate on the chips that run instead of turning the clock down, and settles afresh after such a restart. - A pinned certificate store for TLS. After the eMudhra incident (intermediate CA keys compromised, October 2026), the Mozilla root store dropped the eMudhra root
emSign ECC Root CA - G3and no longer includesAAA Certificate Services; the lists built into ESP-IDF v5.3.3 still carry both. WarpAxe now checks every TLS connection — Stratum V1 over TLS, the online update, the BTC price — against curl's extract of the Mozilla store of 20 October 2026 (118 roots) and nothing else. A root of that store that a server sends along, cross-signed by an older root the store no longer has, counts as that root, so such chains keep working (ESP-IDF's verifier alone rejects them; api.coingecko.com is one). If you mine over TLS: a pool whose only certificate chain runs to one of the removed roots no longer connects, and its operator has to serve a chain to a remaining root. Stratum V2 is not affected (it checks the pool's authority key, not a CA).
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Downloads
- A chip that stops answering is brought back. Until now the miner restarted its chips only when all of them had gone silent for a minute. One chip that dropped out while the others kept hashing stayed dark until the next restart: on our NerdOCTAXE-γ one of eight chips stopped after five days at a steady clock, and the miner ran at 7/8 for almost two days. Now a chip silent for ten minutes while work goes out makes the miner start its chips over, which takes a few seconds. At most one such restart per hour. A chip that goes silent again within an hour of each of three restarts is left alone until the next boot or until the clock or voltage changes. A pool outage stops both timers, so it never counts. The event log shows each restart, and the API reports
-
WarpAxe v0.1.39
Stablereleased this
2026-10-28 07:25:30 +00:00 | 189 commits to main since this releaseA found block in the web UI, and the right hashrate on pools that start low. This release changes nothing in how your miner mines. It shows a found block in the web UI, not only on the display, and fixes a hashrate that read far too high for a few minutes on pools that start at a very low difficulty.
What is new.
- Block found, in the web UI too. Until now a found block showed only on the miner's display; the web UI had nothing, and after a restart it was gone. Now every page of the web UI shows a "Block found!" banner with the height, the coin and the time, in gold with confetti (without animation or confetti if your system asks for reduced motion), until you close it in that browser. The miner keeps the all-time count and the last block across restarts, the best-difficulty tile shows how many blocks it has found, and the event log has a "Block" entry. In the API:
blocksFoundTotalandlastBlock(height, time, coin) in/api/system/info, next toblockFound. Thanks to Rockem, whose Bitaxe Gamma found a block while the dashboard showed nothing. - The hashrate on pools that start at a very low difficulty. When the pool raises its difficulty, the miner raises the chips' reporting threshold with it. Results the chips had found under the old, lower threshold were still arriving for a moment, and each was counted at the new one: 256 times too much on a BM1370 going from 1 to 256. On a pool that starts at 0.001 and steps up several times in the first minutes, a Bitaxe Gamma showed 3.2 TH/s against 1.42 TH/s from the chip's own counter. For two seconds after a raise, such results now count at the old threshold. Shares and what the pool credits you were never affected, only the hashrate the miner showed and reported. Thanks to Rockem for the diagnostics file that showed it.
- The installer page (warppool.org/warpaxe) now has the web UI's starfield behind it.
One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Downloads
- Block found, in the web UI too. Until now a found block showed only on the miner's display; the web UI had nothing, and after a restart it was gone. Now every page of the web UI shows a "Block found!" banner with the height, the coin and the time, in gold with confetti (without animation or confetti if your system asks for reduced motion), until you close it in that browser. The miner keeps the all-time count and the last block across restarts, the best-difficulty tile shows how many blocks it has found, and the event log has a "Block" entry. In the API:
-
WarpAxe v0.1.38
Stablereleased this
2026-10-24 07:59:27 +00:00 | 194 commits to main since this releaseDownload mode without a crash, and no passwords in a downloaded core dump. This release changes nothing in how your miner mines. It fixes a crash on the way into download mode, keeps passwords out of a downloaded core dump, and smooths a few rough edges around updates and the pool connection.
What is new.
- Download mode no longer crashes. Up to v0.1.37, System → Download mode crashed on the way into download mode, every time: the USB driver was taken down while log lines still went to its console port, and the next line, often from the Wi-Fi shutting down, hit a USB lock that no longer existed. The miner still ended up in download mode, so flashing worked, but the next start showed a crash. The console now moves back to the chip's own USB port before the driver goes down. Found while switching our miners to the larger core dump partition; with that partition, the crash left a 70 KB dump that showed exactly this chain.
- The crash banner after a USB session tells the truth. The reset with which a flashing tool leaves download mode no longer counts as a crash, even when the chip reports it as a watchdog reset. A real crash on the way into download mode still does. The raw reset cause from the chip now appears next to the reset reason in the log, in
/api/system/info(resetCause) and in the diagnostics file. - Passwords stay out of a downloaded core dump. A core dump holds what the firmware's tasks had in hand at the crash, and that can include a password. Its download (
GET /api/v1/coredump) now replaces the Wi-Fi password, the pool passwords, the hotspot password, the Wi-Fi name and the worker names and addresses withxand corrects the dump's checksum, so the usual tools still read it. Other details, such as the pool addresses or the miner's IP address, stay in the dump. The copy in the miner's flash stays as it was written. - One online install at a time. A second click on Install, or a click from another browser tab, used to queue a second install that started as soon as a failed first attempt ended. The button now turns off as soon as you confirm the install, and the miner refuses a second install request while one is queued or running. Thanks to John11 for the report.
- The pool's certificate waits for the clock. After a USB session or a restart, the miner's clock is carried over from before the reset and can be wrong until the network time arrives. An Sv2 pool's certificate checked against that clock looked invalid, and three such failures sent the miner to the fallback pool for up to five minutes. In the first five minutes after a start, the miner now waits up to 20 seconds for the network time before it connects, and a certificate that fails against an unconfirmed clock no longer counts towards the fallback.
- 'B' on the USB console goes the clean way. Typing
Bon the miner's USB console port puts it into download mode the same way as the web UI does: the chips are switched off first, and a restart marker lets the next start tell the flashing tool's reset from a crash. If the firmware does not respond, the miner still enters download mode on its own after 10 seconds.
Details:
docs/12-Panic-without-a-core-dump-in-the-restart-after-an-update.md.One image for every supported board:
warpaxe.bin,warpaxe.bin.sha256,warpaxe.bin.sig; for the first installwarpaxe-factory.binandweb-flasher.tar.gz(checksums ininstall.sha256).Downloads