Repair Guide · Antminer S21 · Burn-In Testing · Hashboard Repair

How We Test a Used Antminer S21 Before It Ships

A used S21 can hash at nameplate for ten minutes and still be a bad buy. What separates a serviceable unit from a liability is what the thermal and per-board data look like after the machine has been hot for hours. Here is the exact test sequence we run before a unit leaves the bench.

By Frontline Hash Technical Team··4 min read
3
Hashboards logged individually per unit
4-6 h
Typical staged burn-in before sign-off
2
Passes required: short ramp and steady-state soak

A used S21 that powers on and reports a hashrate is not a tested miner. The failure modes that cost operators money — a hashboard running on borrowed time, a temp sensor reading low, a PSU sagging under sustained load — all hide behind a clean first ten minutes. Our incoming inspection exists to force those problems into the open before the unit ships.

This is the sequence we run on every used S21 that comes through the bench, in order, with the pass/fail criteria we actually apply.

Step 1: Visual and mechanical inspection

Before the unit is powered, it is opened. Dust loading on the heatsinks and fans tells you how the previous owner ran it. We check for:

  • Corrosion or discoloration on the hashboard connectors and the PSU output terminals
  • Fan bearing play and cable damage on the fan harness
  • Bent or reflowed-looking areas around the BM chips and the voltage regulators
  • Missing or stripped chassis screws, which usually means someone was inside it and did not finish the job

A board that has been reworked before is not automatically rejected, but it gets flagged and inspected at higher magnification. Previous amateur repair is one of the most common reasons a used unit fails later.

Step 2: Power path and control board check

The PSU is tested under load, not just for output voltage at idle. We confirm the rails hold within spec while the unit is drawing real current, and we check the control board boots cleanly, enumerates all three hashboards, and reports a consistent chip count per board.

Chip count is a hard gate

If a board reports fewer chips than it should, the unit does not proceed to burn-in. A missing chip means a dead or degraded domain, and no amount of runtime will fix it.

Step 3: Firmware verification

We confirm the firmware is a known, legitimate build for the model and that the unit is not running a modified image with aggressive voltage or frequency settings baked in. Amateur overclocking is one of the documented killers of second-hand Antminers — the damage shows up as chips that pass at low ambient and fail once the room warms up.

Where the firmware is unknown or modified, we reflash to a stock build and re-run the enumeration check before continuing.

Step 4: Staged burn-in

Burn-in is not one long run. It is two passes with different purposes.

  1. Step 1

    Ramp pass

    Bring the unit up to full load and confirm it reaches its rated hashrate without faults, chip errors or fan excursions. This catches hard failures fast.

  2. Step 2

    Steady-state soak

    Hold the unit at full load for several hours at an ambient that resembles the customer site. This is where marginal boards, weak PSUs and cooling problems surface.

  3. Step 3

    Post-soak restart

    Power cycle and confirm the unit comes back to full hashrate cleanly. A unit that needs a warm restart to behave is a unit that will drop out in the field.

The soak is the part most sellers skip, because it is the part that fails machines. A unit that hashes fine for twenty minutes and then drops a board at hour three is exactly the unit a short test lets through.

Step 5: Hashrate acceptance

We do not accept a single hashrate reading. Each board's hashrate is logged across the soak and compared against its rated share of the unit's total. A healthy board holds steady within a narrow band. A board that oscillates, dips periodically, or drifts downward over the run is flagged even if its average looks acceptable.

Reported chip count

Matches rated count on all 3 boards

Any board short of chips

Hashrate stability

Steady within band across soak

Periodic dips or downward drift

Chip errors

None sustained

Any sustained error stream

Post-soak restart

Returns to full hashrate cleanly

Needs warm restart or drops a board

Fan behaviour

Stable RPM, no hunting

RPM swings or fan faults

Step 6: Per-board thermal data

This is the section buyers ask about least and should ask about most. A single unit temperature reading averages three boards and hides the one that is running hot. We log each hashboard's temperature and each board's individual sensor readings separately, through the full soak.

What we are looking for:

  • Boards that track within a few degrees of each other under identical load
  • No single board climbing steadily while the others plateau
  • No sensor reading that is implausibly low compared with its neighbours — a dead sensor reads cool and masks a hot board
  • Intake and exhaust deltas that make sense for the fan speed the unit is commanding

Step 7: Ambient derating note

We test at an ambient that resembles the destination site, and we record how the unit behaves as the room warms. A unit that holds its hashrate at a cool bench ambient but derates hard at a warm one is not a bad unit — but the buyer needs to know, because it changes how many units that room can carry.

What gets rejected

Board with missing chips

Fails enumeration. Not shipped as a complete unit.

Unstable hashrate after soak

Board drops or drifts under sustained load. Pulled for board-level repair.

Thermal outlier

One board running significantly hotter than its siblings. Inspected and reworked before resale.

PSU sag under load

Rails fall out of spec at full current. PSU replaced, unit re-soaked.

Rejected units are not scrapped by default. Most become repair jobs — a hashboard rework, a PSU swap, a fan and thermal interface refresh — and re-enter the test sequence from step 2.

Why this matters for farm operators

The secondary market is full of machines that were fine when the seller last ran them and are not fine now. The difference between a used S21 that earns for two years and one that fails in month two is almost never visible at power-on. It is visible in the soak data, the per-board thermals, and the post-soak restart — which is exactly what a proper burn-in produces.

If you are buying used S21 units in volume, ask for the per-board thermal log and the soak hashrate trace, not just a screenshot of the dashboard. If a seller cannot produce them, the unit has not been tested.

Frequently asked questions

How long should a used S21 burn in before you trust it?

We run a short ramp to confirm the unit reaches full hashrate without faults, then a steady-state soak of several hours with per-board temperature and hashrate logged. A unit that only looks healthy in the first 30 minutes has not been tested.

What hashrate should a used S21 hold?

Acceptance is based on the unit's own rated configuration, not a single market number. A healthy board should hold steady within a narrow band of its rated share across the soak, with no sustained dropouts or chip-count errors in the log.

Why does per-board thermal data matter more than a single temperature reading?

One hot board drags the whole unit's average and hides the problem. Logging each hashboard separately exposes a single underperforming board, a failing temp sensor, or a cooling path that is only marginal on one side of the chassis.

Can a used S21 pass testing and still fail in the field?

Yes, if it was tested in a cool room at low ambient. We test at an ambient that resembles the customer's site and note the derating behaviour, so the buyer knows how the unit behaves when the room gets warm.

Repair & maintenance

Get Your Miners Back to Full Hashrate

Chip-level diagnostics, component-level repair, and burn-in validation — on-site across the USA, no shipping required.