Guide

Troubleshooting

No GPUs found, rejected shares, a card that never mines, a dashboard that will not open.

The first forty lines of the log name the build, the config file, the CPU, every device found and the pool being tried. Start there. Then find your symptom below.

It will not start

  • macOS: “cannot be opened because the developer cannot be verified”. macOS has quarantined the download. In the unpacked folder run xattr -dr com.apple.quarantine . once, for the whole folder.
  • macOS: “bad CPU type”. The build is Apple Silicon only. It does not run on Intel Macs.
  • “algorithm ‘X’ is not supported by this build” and it drops into monitoring mode. A typo in -a, or a mining OS wrote a name this binary does not know. bzminer --list-algos prints what it does know; --force_algo names the right one over what the OS wrote.
  • It starts, mines nothing, and says monitoring mode. No pool configured, or "pool": [] in config.txt. The shipped config starts that way on purpose; put your wallet in pools[] and set "pool": 0.

Devices

  • No GPUs found. Install or refresh the GPU driver. --gpu-info shows what bzminer can see; --list-metrics shows every sensor.
  • A GPU is listed at startup but never mines. The log says why on its own line: no kernel for that architecture, a driver too old to load one, or not enough VRAM for the algorithm’s table or DAG. Ergo wants about 2 GB free; Ethash and KawPow want room for a DAG of several GB.
  • A card is skipped with --devices named in the reason. Your own list excluded it. Every excluded card says so by name at startup.
  • Warthog finds nothing. It needs both a GPU and the CPU. Check that neither is disabled — --nvidia on its own switches the CPU off.
  • One identical card is much slower than another. Look at its PCIe link on the device page (d): an x1 riser is a ceiling for algorithms that stream results off the card.
  • An Intel card shows up but is not used for Ethash. Intel is not listed for that algorithm because nobody has mined Ethash on one from this build, so it is not promised.

Shares and pools

  • Pool rejects shares. Almost always the wallet: press u to see exactly what is being sent, and check it is an address for the coin you are mining. On KawPow the five coins’ addresses look alike; on Ethash the wrong -a for the pool’s chain rejects every share while the hashrate looks fine.
  • Connected, hashing, no shares at all. On MoneroOcean, a missing ~algorithm pin in the password: the pool is serving a different algorithm than the miner. See the MoneroOcean guide.
  • Pool hashrate far below the miner’s. Give poolhr an hour; it is computed from accepted shares and swings over minutes. If it stays low, the pool is crediting fewer shares than the rate implies — check for rejects and stales in the table.
  • The worker shows up as rig.rig or all rigs as one. A worker name in the wallet and in --worker, or a dot where Kryptex wants a slash. Put only the address in -w.
  • Certificate error on a TLS pool. --ssl-verify is on and the pool self-signs. Turn it off for that pool.
  • The backup pool is never used. It is a different algorithm from the primary and was skipped with a warning.

Speed

  • RandomX is slow. It wants 2080 MB plus 2 MB per thread and huge pages; the startup line says which it got. Too many threads for the L3 cache makes it slower, not faster. A CPU without AES runs it about four times slower and bzminer says so.
  • XELIS hashrate is lower than an older bzminer showed. The old figure was derived and read high; this build measures the rate itself. Compare accepted shares over a fixed period instead — the same unit in both versions.
  • Pearl’s hashrate is tiny. It is meant to be: every share is a STARK proof, so the number is small by design. Judge it by accepted shares and the tbs column.
  • Warthog is near zero for the first two minutes. The log shows calibrating verus during that time. It is the calibration phase and it ends on its own.
  • The rate reads 0, 0, 0, burst, 0. The hashrate window is too short for a bursty card. Widen it with --hashrate-window 60.
  • A message about MSR tweaks. CPU register tuning is an optional speed-up, not a requirement. On Windows it needs PawnIO (install-pawnio.bat in the archive); on Linux the msr module and access to /dev/cpu/*/msr. Run with --log-level debug to see which register was refused.

Overclocking

  • Overclock does nothing on Linux. Every knob is a privileged driver write, so run the miner as root. The log reports which knob was refused and why.
  • An AMD card refuses a clock or power limit. Only the fan is controllable on AMD from bzminer, so that refusal is a limit rather than a fault.
  • A memory lock lands on a different number. Locks are only honoured at the board’s own states; bzminer tells you which one you got, and the device page lists them.
  • A card still has a fan duty or locked clock after a crash. bzminer --oc-reset on its own puts every card back.

The dashboard

  • Nothing on the web UI. It listens on 127.0.0.1 only. For another machine on your LAN: --set http_address=0.0.0.0, understanding that there is no password on it.
  • The port is already in use. Pick another with --set http_port=8080. The startup log names the port it bound.
  • No Tune button on a card. The miner cannot control that device: on AMD only the fan is settable, on Intel nothing is, and on Linux every control needs root.
  • Hive shows no hashrate. The web server is off or moved from port 4014, which Hive’s stats callback reads.

The console

  • Scrolling the log pane selects text instead, or the other way round. The old Windows console (conhost) cannot do both. Run bzminer under Windows Terminal.
  • The dashboard is a mess of characters in a log file. You forced -o tui on redirected output. Leave -o off and bzminer picks the log screen itself.
  • The table wraps on every row. tui_width is pinned larger than the window. Set it to 0 to follow the terminal.

Two algorithms

  • A second algorithm is not mining. One process mines every algorithm in the -a list, each on its own devices — check the flags carry the right number (--p2, --w2) and that a device was left for it. See mining two at once.
  • I added --cpu 0 to keep the GPU algorithm off the CPU, and the CPU algorithm stopped too. --cpu 0 is rig-wide and takes the CPU away from every algorithm. Put the number of the algorithm you wanted off the CPU on it: --cpu1 0. The same goes for --nvidia 0, --amd 0, --intel 0 and --devices; see which devices mine which algorithm.
  • A ! in --devices did something odd in a terminal. bash expanded it as a history reference. Quote it: --devices '!1'.
  • A second --p1 was meant to start a second algorithm. It adds a failover pool for algorithm 1 instead. Where a flag sits on the line means nothing; only its number does, so the second algorithm wants --p2.

Getting more detail

--log-level debug                  # more about what the miner is deciding
--log-level network                # the raw pool conversation, frame by frame
--log-level info --logfile-level debug   # readable screen, detailed file

The log file sits next to the binary and its name is printed at startup.

Reporting a bug

Please include your OS, GPU models and driver version, the algorithm and pool, and the first forty or so lines of the log. For sensor or overclock problems, the output of --list-metrics and --gpu-info helps. Open an issue on GitHub, or ask on Discord, where there is a channel per algorithm.