ASIC power supplies: what they do and what kills them

Inside every miner sits a power supply unit converting incoming alternating current into the low voltage, very high current direct current the hashboards need. It is unglamorous, it is the single busiest component in the machine, and in most fleets it is the leading cause of a unit going dark.

What it has to do

A machine drawing several kilowatts is pushing enormous current at low voltage into the boards. Conversion is efficient but never perfect, so the PSU itself runs hot, continuously, for years. Unlike a desktop supply that idles most of the day, a mining PSU sits near its rated output every hour it is switched on. That is the whole reason its lifetime is measured differently.

Getting the specification right before you order

• Input voltage range. The single most common ordering mistake. A unit built for 380 to 415 V three phase will not run on a 208 V site. Check it against your actual supply, as set out in three phase power in plain terms.
• Included or not. On some models the PSU is a separate line item. A quote that looks cheaper may simply be missing it, which is one of the checks in reading a quote.
• Headroom. A supply running permanently at the very top of its rating ages faster. Some models leave more margin than others at the same rated output.
• Connector and cable type. Rarely interchangeable between brands, and a mismatch is discovered on installation day.

What actually kills them

Almost never age alone. In order of how often we hear about it:

• Bad power quality. Voltage sag on a weak feed, harmonics from other equipment, poor grounding, or a generator that never quite settles. Failures arrive in batches months after the cause, which is why nobody connects the two.
• Heat. A PSU sitting in the hot aisle breathing its neighbour's exhaust runs far hotter than one in clean intake air. Hot and cold separation is a PSU lifetime decision as much as a chip decision.
• Dust. Conductive dust bridges, insulating dust traps heat. Both kill. Filtration and a cleaning interval are cheaper than replacements.
• Humidity and condensation. Especially on sites that cycle on and off, where a cold machine in warm damp air collects water inside.
• Overclocking. Pushing the machine pushes the supply hardest of all, and usually voids the cover that would have paid for the replacement. See tuning a fleet.

How they fail, and how to spot it early

Sometimes cleanly and completely: the unit is dark and the diagnosis takes a minute. More often they degrade. The machine restarts under load, or drops a hashboard, or reports lower hashrate than it should while everything looks fine on the dashboard. A supply that cannot hold voltage at full draw produces symptoms that look exactly like a failing board, and fleets waste real money replacing the wrong part.

The cheap diagnostic is a swap. Keep a known good spare, put it on the suspect machine, and see which component the fault follows. Doing this before ordering a board saves more than the spare cost.

Spares, and how many

PSUs are the item to hold deepest. They fail more often than boards, they are model specific, and they are the difference between a machine down for an hour and a machine down for a quarter. Build the spare ratio into the purchase rather than treating it as an extra later, and order them with the batch so the part numbers match. The wider spares question is covered in maintenance and spares.

The thing worth doing on day one

Log your incoming voltage for a week before the machines arrive, especially on a rural feed, a generator or anything behind a large industrial neighbour. A supply that dips under load is a problem you can fix for the price of a transformer tap change, and a problem you cannot fix at all once it has quietly cooked forty power supplies over eighteen months.