Antminer S21 XP Hashboard Repair Service

Antminer S21 XP hashboard repair starts from one number: 91. That is how many chips a healthy board reports. Any other number — a few missing, a whole group gone, or a chain showing nothing at all — comes down to a power fault, a communication fault or a failed chip, and all three are repairable at component level.

Most S21 XP boards arrive here after the same short story. The miner ran normally, then the hashrate fell by roughly a third, the log filled up with errors on one chain, and restarting bought a few hours at best. You do not need to work out which board failed or why before you send it in — that part is our job. What follows is what the miner is actually telling you, and what we do about each of it.

Technical Diagnostics and Common Issues with Antminer S21 XP

Quick Symptom Checklist

These are the lines S21 XP owners send us most often, copied out of the miner's own kernel log. The numbers in brackets change from board to board — the shape of the message is what matters.

• One board reports fewer than 91 chips — anything from a couple missing to the whole chain at zero
• Chain X only find 0 asic, will power off hash board X
• ERROR_ASIC_NUM: asic number is not right
• Repeating !!! reg crc error and !!! nonce crc error, sometimes ending in REG_TYPE = 1 with an asic number
• A run of please check chain[2] bigdomain[12] asic[84] lines, followed by find chain[2] domain[12], nonce[1698] < avg[5013] unbalance and ERROR_UNBALANCE: restart try
• Sweep error string = N:3.
• chain[1] get pll config err, times 4
• Full chip count from cold, then one chain drops out a few minutes into real load

Model-Specific Patterns We See on Antminer S21 XP

• Chips disappear in groups, not one at a time. The 91 chips on an S21 XP board are wired as thirteen groups of seven, and the log says so out loud: a fault usually prints seven consecutive please check ... asic[N] lines, one group's worth. So a count that is short by seven, fourteen or twenty-one is a group problem, while a count short by one or two is a single chip. Same symptom on the surface, different repair underneath.
• Unbalance is a real hardware fault, not the firmware being fussy. When the log reports one group returning a fraction of the work the others return and then throws ERROR_UNBALANCE: restart try, the miner drops into a restart loop. Owners often spend a week reflashing firmware over this. The board is genuinely failing to hold that one group at the right voltage or clock, and it does not fix itself.
• CRC errors mean the chain is talking, but the words come out wrong. A board spraying reg crc error and nonce crc error is a communication failure, not a dead board, and the chip count is often still full. It is one of the more recoverable faults on this platform — and one of the few worth ruling out at your end first, since a poorly seated or damaged ribbon cable produces the same lines. We wrote up that check separately in our guide to reg CRC errors on Antminer S19 and S21 boards.
• A board that passes cold and fails warm is almost always mechanical. Full 91 chips at startup, then a chain drops minutes into load, points at heat rather than electronics — thermal paste past its life, uneven pressure on the heatsink assembly, or residue left behind by an earlier repair. This one is common on boards that have already been opened once.

Hardware Notes

What the Log Is Telling You

Owners usually reach us with a screenshot of the kernel log and no idea which part of it matters. This is how we read the common ones. You do not need to act on any of it — it is here so you know what state your board is in before you decide to ship it.

Diagnostics Focus

• Power to each of the thirteen groups comes first. Most low chip counts on this board turn out to be a group that is not being fed correctly, and that is answered by measurement long before any chip is touched.
• Then the communication path along the chain. CRC errors, unbalance and clock failures all live here, and they are the faults most often mistaken for dead chips by whoever looked at the board first.

Our Professional Repair Process

What to Send Us

Choose one of the two options above before you check out. The repair price is the same either way.

• Hashboard only — you take the board out of the miner yourself and send only the board. This is what most customers choose.
• Complete miner — you send the machine with the faulty board still inside. We take the board out, repair it, put it back and return the miner assembled.

If you are not confident about opening the machine, or you are not certain which of the three boards is the faulty one, choose the complete miner and let us find it. You get back exactly what you send us, and if the option does not match what actually arrives, we contact you before we do anything else.

Gotchas

• We never power an S21 XP board without its heatsink assembly in place. On a 91-chip board, a few seconds of testing bare is how a single-chip repair turns into a five-chip repair — which is also why a board that has already been "tested" by someone else sometimes arrives worse than it left.
• BM1370 chips are sold in mixed lots and shipped at random by the letter grade. Since that letter is a speed grade rather than a model code, we match the replacement to the marking on your board rather than to the name on the miner.

Typical Service Scenario

• Air-cooled S21 XP units started shipping around the turn of 2025 and Bitmain's standard cover runs 365 days, so by now nearly the whole air fleet is outside it. A large share of the boards on our bench arrive after a warranty claim was already declined.
• A board comes back from a do-it-yourself thermal paste job and starts throwing temperature or chip-count errors it never had before.
• A farm with dusty intake air runs the machine hard for months, and one chain starts dropping out under load while the other two keep hashing normally.

What Happens After Intake

Every board takes the same route through our USA-based lab: intake and incoming inspection, measurement to find out what is actually failing, a bench test to establish what the board reports on its own, component-level repair, cleaning and thermal interface replacement, then repeat testing before final validation in a real miner under load for at least one hour. Extended burn-in beyond that hour is available as a separate add-on. Where chips need replacing, the count is confirmed after diagnosis and quoted against our BM1370 chip replacement add-on, because nobody can tell you how many chips are gone until the board is on the bench.

Diagnostics and Validation Equipment

Bench measurement, STASIC and ASIC REPAIR test fixtures for chip count and work verification, and final validation inside a working miner under real load. The fixture proves the chain; only the miner proves the board.

All repairs are performed personally by Alex, who has been servicing ASIC miners since 2019. See credentials →

Contact our repair team today and get your miner back to full power.