Plenty of sites run third-party firmware — some pushing frequency for hashrate, some dropping voltage for efficiency. The discussion around tuning usually comes down to one calculation: how much more hashrate, and how much less power.
From the repair bench there is a second calculation. Tuning changes the load on every component, and it changes what failure looks like. This article is about that second one: which parts carry the extra stress under each strategy, which “faults” are really settings problems, and how spares and repair planning should follow.
Tuning changes more than hashrate
Stock settings are specified conservatively, with headroom between components. Change frequency and voltage and that headroom is redistributed:
- the power supply runs closer to, or further from, its rated ceiling
- hashboard and chip operating temperatures shift
- fans hold higher speeds for longer
- cables and connectors carry different current
Two machines of the same model, one at stock and one overclocked, can be in very different condition after a year. When buying spares or judging a repair, what the machine was running matters as much as which machine it is.
Overclocking: where the stress lands first
Power supply. Overclocking raises total draw. Before tuning, confirm the supply’s rated output covers the post-tuning figure with margin left over. A supply held near full load runs hotter and ages sooner. When replacing or upgrading, judge compatibility against the tuned draw, not the factory specification.
Hashboards and chips. Higher frequency means more heat. Where cooling cannot keep up, the hottest region of the board is usually the first to go.
Fans. Holding temperature down means running at high speed for longer, and wear follows. On these machines fans should be treated as routine consumables rather than occasional replacements.
Cables and connectors. More current makes any poor contact run hot. An oxidised or loose connector, or cable undersized for the load, may pass unnoticed at stock and show up once the machine is tuned.
Undervolting: efficient, but easily mistaken for hardware failure
Undervolting generally lowers heat and component load, which is kinder to the hardware. Pushed too far, though, some chips stop working reliably: hashrate drops, chips fall offline, the machine restarts repeatedly, or a hashboard is not detected at all.
These symptoms look like genuine hardware faults. Judging by the error alone, it is easy to replace a board that was never broken.
Suggested order. Return the machine to stock, or to a profile known to be stable, and run a test. If the problem disappears, it was the settings. If it persists, then investigate the hashboard, the power supply and the control board.
Confirm firmware and settings before sending a unit in
Whether sending a unit for repair or sourcing a replacement part, include:
- stock firmware or third-party firmware
- the running profile: stock, overclocked or undervolted
- how long after tuning the problem appeared, or whether it appeared immediately
- the fault behaviour and any logs
For the repair side this determines the test method: a repaired board should be tested at the settings it will actually run, not only at stock. Note also that flashing third-party firmware can affect the manufacturer warranty, so it is worth confirming warranty status before tuning.
Planning spares around the running strategy
Fleets run for hashrate. Hold deeper stock of power supplies and fans. Check the supply’s rated output against the tuned draw before tuning, not after.
Fleets run for efficiency. Component load is lower and spares pressure is lighter. But when a hashboard appears to have failed, rule out the voltage setting before replacing anything.
Mixed fleets. Record firmware and profile per group of machines. When something fails, that record cuts diagnosis time substantially.