ASIC Power Supply: PSU and Electrical Considerations

A Bitmain-style ASIC does not use a living-room PC supply. The power supply is usually a server-style PSU: a dense AC/DC converter, forced-air cooled, built for high DC current on a small number of rails, with protections and some talk-back to the control board.

The useful rating is always the one on the model manual and the PSU label. What follows explains how to read those two documents and what breaks when you drift from them.

A server-style brick

A miner PSU looks like server and compute-GPU bricks: metal chassis, AC inlet (often single-phase, site-dependent otherwise), low-voltage DC out to the hashboards. Connectors are generation-specific, and an improvised adapter forcing one rail into another model ends as a hot joint or a dead board.

On the AC side, the order of magnitude is one line of arithmetic. A 3,500 W machine on 230 V draws about 15 A continuously. A domestic 16 A socket is therefore at its absolute limit, on a circuit never designed to run at full load around the clock. In practice that means a dedicated circuit, a properly rated breaker, and C19 / C20 style connectors rather than a household plug. Two machines on the same line means a trip or a hot distribution board.

A unit that is too small collapses under load: resets, vanishing boards, fans that stop. A brick from another family may have a different nominal voltage, pinout, or enable logic, even when the plugs look close. For nominal voltage and current, read the vendor sheet (product pages and manuals on Bitmain). Without that sheet for this miner × PSU pair, no value should be assumed.

Rails: what that means

A rail is a DC voltage path shared by a group of loads. On these machines, hashboards draw most of the current on one or more dedicated rails; the control board and sometimes the fans have smaller supplies, derived or separate.

That separation explains most of the confusing failures. If a power rail collapses, boards on that rail go dark or error while the web interface still answers, because it is fed another way. If the controller rail dies, the machine leaves the network entirely. Telling "no IP" apart from "no hash" saves you from replacing a healthy PSU or a healthy controller.

Cables and lugs are part of the rail. A loose, oxidized, or undersized joint acts like a resistor: it heats locally, sags the voltage, and causes random resets under load while everything looks fine at idle. A torque check on the power screws and a thermal camera pass at start-up catch this class of fault before it costs a board.

Protections: OVP, OCP, OTP

Serious switched-mode supplies include protections whose job is to stop before they destroy the converter or the load.

• OVP (over-voltage protection): if output voltage exceeds an internal threshold, the PSU shuts down or faults, so it never sends too many volts into the ASICs.
• OCP (over-current protection): if output current exceeds a threshold, the brick cuts or folds back. Typical cause: a short, a faulty board, or a demand beyond what the unit accepts.
• OTP (over-temperature protection): if the PSU itself runs too hot (recycled air, clogged filter, dead PSU fan), it protects itself.

Those thresholds live inside the brick and are not tuned like a hash profile. When they trip, the symptom is a stop, a blink, or a silence, rarely an explicit message. Power-cycling in a loop without finding the cause (overload, heat, board fault) wears the hardware and turns a repairable fault into a dead board.

Other trips exist on some units: under-voltage, fan fault, AC out of range. Trade names change, the idea does not: the PSU would rather turn off than keep running off-contract.

Why matching matters

Matching a PSU means checking five points, in this order:

• nominal voltage compatible with the boards;
• current capability sufficient for the firmware profile you actually run, knowing that an aggressive profile asks more of the rail than the factory one;
• exact connector and polarity;
• enable logic, since some bricks wait for a controller signal before delivering power;
• PSU cooling in the same airstream as the miner, or a dedicated path, as the chassis dictates.

A brick can hold idle and collapse the moment the ASICs come up: the fact that it lights up proves nothing. The test that counts is a full ramp to load, with voltage measured at the boards rather than at the PSU output.

Hosting sites often mandate a PSU type, a plug, and a replacement discipline. A mismatched brick is an electrical and thermal risk to the whole rack, not only to the machine it feeds.

Before you plug in

Three checks that prevent most incidents:

• Read the PSU label and compare it line by line with the model manual: output voltage, maximum current, accepted input voltage.
• Confirm the supply circuit carries the calculated current (machine power divided by mains voltage), with the margin your local code requires for a continuous load.
• Verify that the PSU intake sees fresh air, not the hot exhaust from the miner itself or from the machine next to it.

The most common bad reflex is treating the PSU as an interchangeable charger, when it defines the electrical contract for the entire chassis.