• v0.1.37 3e0188ba64

    WarpAxe v0.1.37
    All checks were successful
    CI / Rust checks (push) Successful in 56s
    CI / Regulator scenarios (push) Successful in 55s
    Stable

    dvb-projekt released this 2026-10-23 19:32:56 +00:00 | 202 commits to main since this release

    A crash now leaves a trace. This release changes nothing in how your miner mines. It leaves more to look at if WarpAxe ever crashes, and it removes a likely cause of the crash that started this.

    Why. A user's miner came back from the restart at the end of an online update with a crash (reset reason panic) instead of a normal restart, and left no core dump. Looking into it showed that this firmware's crash dumps are 60–64 KB, while the partition that holds them is 64 KB; a dump that does not fit is not written at all. So whether a crash left a dump was largely luck.

    What is new.

    • Last words. A Rust panic now writes the task, the file and line and the message into memory that survives the reset; a stack overflow writes the name of its task. The next boot logs them, and the diagnostics file carries them as lastWords, whether or not a core dump was written. Other crashes, such as a hardware exception or a watchdog reset, leave no last words; for those there is the core dump.
    • A crash during a restart is named. If the miner crashes while restarting (after an update or an upload, or a restart from the web UI), the next boot records an event that says so and names the step: while switching the chips off, or in the system reset itself.
    • A likely cause, fixed. The interrupt watchdog now allows 1000 ms instead of 300 ms. Erasing a flash sector can take up to 600 ms, and while it runs, the watchdog is not fed, so the old limit could run out in the middle of an erase. On our Bitaxe Gamma, an image uploaded after 39 minutes of uptime came back with an interrupt-watchdog reset. The restart event named the step "in the system reset", the moment the restart writes its record to flash, and there was no core dump even though the partition was 256 KB: a crash during a flash operation never leaves one, because the dump cannot be written while the flash is busy. The user's miner, on v0.1.34, reported a panic rather than a watchdog reset, and it too wrote to flash (its install record) shortly before the reset; so this is a probable cause, not a proven one. If a flash write during a restart takes 100 ms or longer, the log now says so.
    • Updates report themselves on every bootloader. A finished update or upload now shows up as "installed" in the event log even on miners whose bootloader predates v0.1.22, but not when the miner rolled back to the previous version. Like the restart event, this works for restarts begun by v0.1.37 or later: the update from an older version to v0.1.37 is not yet covered.
    • No crash banner after an install over USB. The browser installer, espflash and esptool -a watchdog-reset leave download mode with an RTC-watchdog reset, and since v0.1.34 WarpAxe took that reset for a crash: the browser installer brought up the crash banner after every install. Now that reset no longer counts as a crash on the first boot of a newly flashed image, or after a restart into download mode; it still does after a rollback to the previous version. A panic, an interrupt-watchdog reset or a task-watchdog reset always counts as a crash. The banner still shows when you reflash the image that ran last over USB without starting at System → Download mode. The price: a real watchdog hang in the very first moments of a new image goes unreported.
    • A 256 KB core dump partition. The partition grows from 64 to 256 KB, so a full dump fits. It comes only with an install over USB, because an online update never rewrites the partition table. The browser installer keeps your settings; warpaxe-factory.bin flashed with esptool starts with empty settings. A miner first installed before v0.1.37 and since updated only online keeps the 64 KB partition: last words and the restart event work there too.

    Details and the investigation: 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, tested-on.txt; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256).

    Downloads
  • v0.1.36 bad4a85a56

    WarpAxe v0.1.36
    All checks were successful
    CI / Rust checks (push) Successful in 58s
    CI / Regulator scenarios (push) Successful in 52s
    Stable

    dvb-projekt released this 2026-10-23 12:16:13 +00:00 | 208 commits to main since this release

    Only the BitForge Nano gets new limits and a new temperature scale; every other board keeps its temperatures and limits. The Nano's stock firmware, ForgeOS, is open source, and this release brings WarpAxe in line with what it does on the board it was made for (ForgeOS v1.7, read from its source; details in docs/08).

    The Nano's temperatures on the maker's scale. The chip temperature sensor reads the BM1370's diode through an ideality setting. ESP-Miner tunes it to 0x24 on the Bitaxe Gamma boards; ForgeOS has tuned the Nano to 0x37 since its first release. Measured back to back on the same chips, 0x24 reads 8.2 °C higher. WarpAxe now uses 0x37 on the Nano, so its 75 °C chip stop applies where ForgeOS applies it (the 105 °C regulator stop reads the regulator's own sensor and is unchanged). After the update a Nano shows about 8 °C less at the same point. The fan target moves with the scale to 52 °C, so the chips stay at the same real temperature as under v0.1.35 and the fans run as before (a first candidate kept 60 °C on the new scale, and the Nano's error rate rose from about 1.5 % to 6 % at 425 MHz / 1000 mV). If the miner has the old default of 60 °C stored (an earlier save of the settings page stores it even if you never changed it), it is reset once to 52 °C on the first start; a different value you chose stays and now counts on the new scale (about 8 °C lower gives the chip temperature it held before). The same goes for a chip stop you set yourself below 75 °C: it is not moved, so lower it by about 8 °C if you want it to stop at the same real temperature. If you tune the Nano with WarpTune, the WarpTune card shows a yellow hint after the update that the fan target has changed since the calibration. That is expected and needs no action; it goes away after the next calibration. A new calibration is optional and only worth it to get the Max preset back to a full clock under WarpTune's limits.

    The Nano's regulator limits as ForgeOS sets them. The regulator now warns at 38 A and shuts down at 40 A (before: 45 / 52 A; the part is rated for 40 A), and the input over-voltage limit is 13.5 V (before: 14.8 V). The 625 MHz / 1160 mV choice stays: measured on our Nano at 37.1 A on average and 37.6 A at most, it runs close to the warning line; ForgeOS offers its own 636 MHz / 1160 mV preset under the same 40 A limit. If it ever trips the limit, WarpAxe handles it like an overheat stop: chips off for at least a minute, then back on; the third stop within an hour keeps them off and shows why.

    A pool's extranonce size is checked. A pool that asks for an extranonce2 size outside 1–16 bytes (Stratum V2: 1–32) is refused instead of trusted. Before, a broken or hostile pool could make the miner reserve as much memory as it named.

    Moving from the stock firmware.

    • A pool the previous firmware ran on TLS with your own certificate now stays on TLS, with the built-in certificate bundle, instead of being switched to plain text as before. WarpAxe has no place for your own certificate, so a pool on a private certificate fails visibly; switch TLS off for it on purpose if you trust the network.
    • If the previous firmware was in its overheat mode (ForgeOS and older AxeOS versions leave 1000 mV, 50 MHz and fans at 100 % behind), WarpAxe does not take over clock, voltage and fans; the board's own defaults apply.

    Tested on 2026-10-23 on the BitForge Nano: both diode settings read back to back on each chip (106 pairs, warm at 525 MHz / 1150 mV), error rate unchanged, regulator limits in place.

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

  • v0.1.35 29ff8dbeb9

    WarpAxe v0.1.35
    All checks were successful
    CI / Rust checks (push) Successful in 54s
    CI / Regulator scenarios (push) Successful in 56s
    Stable

    dvb-projekt released this 2026-10-22 09:07:07 +00:00 | 223 commits to main since this release

    Nothing changes for miners that already run WarpAxe. A board set up by an earlier version is taken as verified on the update and starts as before. What follows applies to a first start, a board chosen anew, or a check you ask for.

    A board is checked before its chips get power. On a first start the web UI lists only the boards that fit what answered on the miner's I²C bus; Show all boards reveals the rest, with a warning. Whatever is chosen, the firmware checks it against the bus at the next start before it drives any of the board's pins: the voltage regulator, the sensors and the other parts that board must have, and the ones it must not have. A board that does not fit is refused, its chips stay off, and the web UI says why (for example which part did not answer) and lists the boards that do fit. A missing sensor on a newly chosen board is refused as well.

    A cautious first start, then a count. A board that passes the check starts its chips at a cautious voltage, the lowest of the boards that look alike on the bus, and counts them. If their number and type match the board, it counts as verified: the configured voltage applies, and the self-test runs once by itself about 20 seconds later. Another board's chip count, a chip of another type or no chips at all switch the chips off again and show the reason, with Restart and try again or the choice of another board. Fewer chips than the board has, where no look-alike board has that count, only bring a warning (a dead chip, not a wrong board). This applies to boards the detection found as well as to boards chosen by hand. To put a verified board through the check again on purpose: PATCH /api/system with {"boardModel": "<its name>", "reverify": true}, then restart. Details: docs/11-Board-choice-guard.md.

    The BitForge Nano and the NerdAxe Gaia are found without asking. The Nano is taken when its PCA9544 mux, INA260 and TPS546 answer on the bus; the Gaia when its TPS546 and EMC2101 answer and the regulator reads 12 V at its input. The same parts at 5 V are a NerdAxe Gamma, which is still offered for one click. The Nano, the Gaia and the NerdNOS, which have been on our bench for weeks, are no longer marked "not yet tested with WarpAxe".

    The event log in your language. The details of each event (restart reasons, pool switches, board checks, overheat stops, regulator faults, self-test results) are now shown in the web UI's language. The API and the diagnostics file keep the English text in d word for word, so tools reading it see no change; new optional fields c and p carry a short code and the values.

    Smaller things. When a first start or a check reads the bus, the regulator's input voltage gets a log line of its own, in volts.

    Tested on 2026-10-22 on a NerdAxe Gaia and a BitForge Nano: first start, chips counted and verified, self-test PASS with 9 of 9 checks. New strings in eight languages, translated and then read by independent editors.

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

  • v0.1.34 18cbd6a614

    WarpAxe v0.1.34
    All checks were successful
    CI / Rust checks (push) Successful in 1m1s
    CI / Regulator scenarios (push) Successful in 56s
    Stable

    dvb-projekt released this 2026-10-21 13:45:39 +00:00 | 239 commits to main since this release

    WarpTune users: update to v1.42.1 first. WarpTune v1.42.0 can no longer read a miner once a self-test has run on it, and stops tuning it. Versions up to v1.41 are not affected.

    Self-test. System → Run self-test checks the miner while it mines, in about 80 seconds: the number of chips, the hashrate of every chip and every domain, the error rate, the core voltage against its setting, the regulator, the fans (ten seconds at full speed, then the automatic fan control takes over again), the temperatures against their stop lines and the input voltage. Each check gets a ✓ or ✗ with its measured value (a dash if the board lacks the part), in the web UI, on the device display and in the API (/api/v1/selftest). The test ends by itself if the clock or voltage changes during the run, and WarpTune (from v1.42.1) holds back while it runs. Tested on a Bitaxe Gamma, a NerdQAxe++, a BitForge Nano and a NerdAxe Gaia. We recommend running it once after installing WarpAxe.

    A diagnostics file for support. System → Download diagnostics file gives you one file with everything we need to help: firmware and board, the last self-test, hash counters, pool status, regulator and temperatures, memory, the event log, recent settings changes and the last log lines. Passwords are never in it, and wallet addresses and the Wi-Fi name are masked. If something seems wrong, run the self-test, download the file and send it to us on Discord.

    An event log that survives restarts. The last 32 important events (restarts with their reason, overheat stops, regulator faults, pool switches, updates and rollbacks, silent chips, self-test results) are kept across restarts and shown under System. They are written only when something happens, never periodically.

    A notice after a crash. If the miner restarted because of a panic or the watchdog, the web UI shows a banner with the reason and the time, a button for the diagnostics file and, if this crash wrote one, a button for its core dump. If the crash wrote none, the banner says so, and the Diagnostics card marks a dump still in flash from an earlier crash or another firmware version as such instead of passing it off as this crash's. The banner stays until you dismiss it.

    A smaller image. The firmware no longer carries Rust's backtrace symbolizer, which could only print addresses on the miner anyway: about 125 KB less, so the new features fit with room to spare. Panic messages and the core dump stay.

    New strings in eight languages, translated by native speakers and read by independent editors. Details: docs/10-Self-test-diagnostics-and-event-log.md.

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

  • v0.1.33 0fab581c61

    WarpAxe v0.1.33
    All checks were successful
    CI / Rust checks (push) Successful in 55s
    CI / Regulator scenarios (push) Successful in 54s
    Stable

    dvb-projekt released this 2026-10-21 07:47:20 +00:00 | 258 commits to main since this release

    Texts that read like the language they are in. Every string of the web UI and the device display was proofread in all eight languages, first in a full proofreading pass and then once more by independent editors who read each language as a whole. Literal translations, broken sentences, inconsistent terms and typography are fixed. Two messages were wrong in substance and are now right: in Japanese the job-interval help described fetching work from the pool instead of handing work to the chips, and the BTC-price hint in the cockpit settings pointed down to a switch that sits above it. In the web UI the percent sign now follows each language (25 % in German, French, Spanish and Italian, 25% in English, Portuguese, Japanese and Chinese). Swarm is now called Swarm in every language; Japanese and Chinese had translated the name.

    Faster updates. Since v0.1.32 the firmware is written by the main task, so an update always finds the memory it needs; but that task runs at a low priority, below the mining and network threads, and on a NerdQAxe++ an update took about 144 s from upload to restart. The write now runs at the priority the old writer thread had and then drops back: about 60 s.

    No change to mining or the regulators.

    On the T-Display-S3 (experimental): the update to v0.1.33 still goes through v0.1.32's writer, the low-priority main task, which shares its core with the CPU miner at the same priority. The final image check then takes about 5 minutes, and the page shows "verifying" the whole time. Leave it alone; the miner restarts by itself. From v0.1.33 on, updates there take seconds too.

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

  • v0.1.32 821e8d2890

    WarpAxe v0.1.32
    All checks were successful
    CI / Rust checks (push) Successful in 55s
    CI / Regulator scenarios (push) Successful in 1m0s
    Stable

    dvb-projekt released this 2026-10-20 08:47:51 +00:00 | 268 commits to main since this release

    Overheat stops at the stock firmware's lines. Up to v0.1.31 a Bitaxe and the BitForge Nano stopped at 90 °C on the chips and 145 °C on the regulator: values we had taken from two ESP-Miner constants that ESP-Miner itself never uses. AxeOS stops at 75 °C and 105 °C, and so does WarpAxe now. Every board profile was checked against its stock firmware's source (the details are in docs/08): the NerdQAxe++, NerdOctaxe and NerdQX stay at NerdOS's 70 / 70, the NerdQAxe++ Rev 7's regulator now stops at 70 °C as on NerdOS, and the NerdAxe Gamma and Gaia keep a 105 °C regulator stop that NerdOS does not have on these one-fan boards. You can lower the chip limit in the settings; on the Nerd boards you can also raise it up to 90 °C, as NerdOS allows, on a Bitaxe not above its default (a higher value stored earlier now counts as 75 °C). Thanks to the user on our Discord who pushed on the 145 °C.

    Saved pools. Pool → Saved pools keeps up to eight pools, each with a name and the coin it mines. "Save current pool" stores the main pool as it is set now, "Use" makes a saved one the main pool, and the miner reconnects at once, no restart. Passwords stay on the miner. Coming from AxeOS 2.15, the pools it had saved are taken over on the first start.

    Solo mining another SHA-256 coin. "Payout if you find a block" decodes the address in your worker name (base58, bech32/bech32m and CashAddr) and matches it against the coinbase outputs, so solo mining Bitcoin Cash, eCash, DigiByte, FixedCoin or another fork shows your real share instead of 0 %. The other outputs are written the way that coin writes addresses. A new Coin field in the pool settings sets the unit of the amounts (default BTC). A worker name that is a pool account and not an address now reads "paid by the pool".

    Error rate on the dashboard. The chips' error rate over the last minute sits next to the expected hashrate, as AxeOS shows it (green below 1 %, yellow from 1 %, red from 5 %). A rising error rate is the first sign of a core voltage too low for the clock. The per-chip hash counters stay optional.

    Fixes

    • The Avalon card on the swarm page showed "not reachable" on a miner whose internal memory was fragmented: the read now runs on a PSRAM stack, as the network search does.
    • A pool change made while the miner was still connecting, or waiting after failed attempts, applied only minutes later; it now applies at once.

    Updates always find room. After a long uptime the internal memory of a miner can be too fragmented for the 16 KB the update writer needs; up to v0.1.31 the update then said "not enough internal memory" and nothing happened until the miner was restarted. From v0.1.32 the writer runs on the firmware's main task, whose stack is set aside at boot and was idle after start-up, so neither the click update nor an uploaded image depends on free memory any more (tested on a NerdQAxe++ with the internal heap cut into 8 KB pieces). Restart and download mode go the same way, so the chips are always switched off before the reset. Details: docs/09.

    Updating to v0.1.32 itself still runs the old code. If the update says "not enough internal memory", restart the miner (System → Restart) and start the update right after.

    New strings in eight languages, translated by native speakers. Details: docs/08-Temperature-stop-limits-against-the-stock-firmware.md, docs/09-Update-with-a-fragmented-heap.md.

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

  • v0.1.31 1b4322b608

    WarpAxe v0.1.31
    All checks were successful
    CI / Rust checks (push) Successful in 53s
    CI / Regulator scenarios (push) Successful in 54s
    Stable

    dvb-projekt released this 2026-10-18 12:36:11 +00:00 | 285 commits to main since this release

    Your display, your choice. Settings → Display now lets you pick which LCD pages the key steps through, their order and the start page, and what the cockpit shows: the large number (hashrate, efficiency or power, with the curve under it following), its three places (chip or regulator temperature, power, J/TH, fan, shares, best difficulty, block height, BTC price, pool difficulty, uptime, input voltage) and the footer line. On the OLED you choose which slides rotate and for how long. A live view of the screen sits next to the settings. The factory layout is the one you know; a value your board does not have shows as a dash.

    An Avalon on the swarm page. An Avalon speaks the cgminer API, which a browser cannot reach, so the miner reads it for the page. "Search network" now finds it by itself: while your browser looks for AxeOS/NerdOS miners, the miner walks its own /24 for port 4028 and asks only version (private address ranges only, about 15 s for a /24, one search at a time). The hit shows as "Avalon · cgminer" and adds like any other; you can also add it by hand as "Avalon (cgminer API)". It shows as a card with its work mode (Eco / Standard / Super), chip temperatures, fans, power, shares and best share, and counts into the fleet totals. Read-only by design: only summary, estats, pools and version, only for hosts on your list, one at a time, cached for 10 s, with a time and size bound.

    Questions on the page. The confirmations (board choice, restart, chips off, install update, download mode, reset best difficulty, network restart) open a box on the page instead of the browser's own dialog, which the sign-in window on the setup hotspot dropped silently — "Use this board and restart" did nothing there.

    A smaller image. The web UI is stored gzip-compressed (292 KB of HTML become 90 KB) and served with Content-Encoding: gzip; every browser unpacks it. All eight languages stay; the image is about 200 KB smaller than before these features. The network stack now allows 16 sockets instead of 10, so the search has room beside the web server, the pool and mDNS.

    Fixes

    • The shutdown-temperature help says what it is: the chip limit. The regulator has its own fixed limit per board, shown on the dashboard; both must be 10 °C below their limits before the core voltage comes back. Thanks to the user who asked about the 145 °C on a Gamma.
    • Display: the block-found line and the log page title are translated; French says "Actif depuis" instead of "Uptime". Spanish uses "voltaje" throughout.
    • README screenshots of the display are current again.

    New strings in eight languages, reviewed by native speakers. No change to mining or the regulators. Details: docs/07-Display-choices-and-Avalon-swarm.md.

    One image for every supported board: warpaxe.bin, warpaxe.bin.sha256, warpaxe.bin.sig, tested-on.txt; for the first install warpaxe-factory.bin and web-flasher.tar.gz (checksums in install.sha256). The installer page now opens with an overview of WarpAxe and a link to our Discord.

  • v0.1.30 7e0126bcc0

    WarpAxe v0.1.30
    All checks were successful
    CI / Rust checks (push) Successful in 51s
    CI / Regulator scenarios (push) Successful in 50s
    Stable

    dvb-projekt released this 2026-10-16 11:13:25 +00:00 | 309 commits to main since this release

    A user manual. WarpAxe now has an illustrated manual for beginners, in English, German, French and Spanish: the browser install with its backup, the first start and the setup hotspot, every page of the web UI, the device display, updates and rollback, and the way back to the stock firmware. It is at docs.warppool.org/warpaxe-guide.html, and the installer page and the README link to it.

    Text fixes in the web UI, on the display and on the installer page, found while writing and translating the manual:

    • The installer page lists the NerdNOS among the boards that run, and the BM1397 among the chips.
    • English: the maintenance note names the overheat button as it reads ("Clear overheat shutdown").
    • German: the minimum-fan hint says "about 25 %", as the other languages do.
    • French and Spanish: clearer wording for the overheat button, the Wi-Fi setup screen, "block found" and "block height" on the display, the restart and update notes, the swarm header and the installer's backup box. New strings reviewed by native speakers; the display strings were rendered to check that they fit.

    No change to mining, the regulators or the API.

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

  • v0.1.29 52cbb9d1e4

    WarpAxe v0.1.29
    All checks were successful
    CI / Rust checks (push) Successful in 58s
    CI / Regulator scenarios (push) Successful in 56s
    Stable

    dvb-projekt released this 2026-10-15 09:53:45 +00:00 | 314 commits to main since this release

    The T-Display-S3 hashes about 16 % faster. On the NerdMiner hardware without a mining chip, WarpAxe now writes the next block into the ESP32-S3's SHA peripheral while the engine still runs on the current one, and the loop is in C. Over 30 minutes on our devkit: 337–340 kH/s instead of 292, with no wrong hash and no rejected share, and the chip temperature unchanged at about 59 °C. Every batch re-checks one nonce in software; a single mismatch switches the board back to the previous loop for good (hwPipelined and hwSpotBad in the cpuMiner part of /api/system/info). docs/06 has the measurements, register by register; the bench that took them is in the source and is built with --features shabench, not in the release.

    The boards with mining chips are unchanged in this release.

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

  • v0.1.28 72b20ddf43

    WarpAxe v0.1.28
    All checks were successful
    CI / Rust checks (push) Successful in 1m2s
    CI / Regulator scenarios (push) Successful in 55s
    Stable

    dvb-projekt released this 2026-10-14 05:06:52 +00:00 | 322 commits to main since this release

    Found by name. The miner now announces itself over mDNS the way AxeOS and NerdOS do: open http://<hostname>.local instead of looking up its address, and tools that discover Bitaxes by DNS-SD find it as _http._tcp with the _axeos subtype and the same TXT items (board, family, asic, asic_count, fw_version). The name is the hostname's letters, digits and dashes (NerdQaxe++ → nerdqaxe.local); a name already taken on the network gets -2. /api/system/info reports it as mdnsHostname and fullHostname, as AxeOS does, and the router's DHCP now sees the miner under its hostname instead of espressif. Thanks to John11 for pointing it out.

    Also in this release:

    • README: mDNS in the comparison with the stock firmware (both have it); comparison checked again against AxeOS v2.15.3 and NerdOS v1.1.0.1.

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