Repair or replace a failing ASIC? Diagnose first, decide second
A miner that ran for a year now shows low hashrate, or a hashboard has vanished from the status page. The instinct is to price a repair or shop for a replacement immediately. Resist it. Almost every bad outcome in this situation comes from deciding before diagnosing, buying a used hashboard for a machine whose real problem was a failing power supply, or paying for bench work on a machine that loses money even when healthy. This guide gives a home operator the full sequence: a cheapest-first diagnostic procedure with explicit stop points, a two-test repair-or-replace decision, and a burn-in verification for after the repair. Nothing here is financial advice; hardware work also carries real electrical risk, and the stop points below are not optional.
A symptom is not a diagnosis
The most useful sentence in ASIC repair is that the symptom and the root cause are different things. "Low hashrate" can be dust-clogged heatsinks, corrupted firmware, pool-side rejection, a dying fan, a weak power supply, or a genuinely failing hashboard. "Hashboard not detected" can be a ribbon cable, an EEPROM, a voltage-domain fault, or a broken chip chain. Each cause has a wildly different price tag, from free to more-than-the-machine-is-worth. Pricing the decision from the symptom alone is how a cable problem becomes a board purchase.
Context helps discipline the search. Repair technicians rank the common failure causes roughly as: power supply failure first, then hashboard chip failure from power surges or thermal stress, fan bearing failure (which causes overheating shutdowns), firmware corruption from interrupted updates, and dust accumulation causing chronic overheating. Note how many of those are not the hashboard at all, and how many are cheap. The procedure below is ordered so the cheap, common causes get eliminated before anyone talks about board-level work.
The home diagnostic procedure: cheapest checks first
Work these steps in order. Do not skip ahead to board swaps because a step "seems unlikely", the order is the value. Power off and unplug the miner before any hardware check; that rule has no exceptions.
1. Read the status page and the kernel log before touching anything. Note exactly what the machine reports: is a whole board missing, or is one board showing a reduced chip count (a log line of the form "chain [X] find N chip" with N below the board's design count)? Is there a "find 0 asic" error on one chain? Save or photograph the log. Rebooting repeatedly before recording the evidence destroys the information you need, and if the log shows the miner repeatedly powering a board off at startup, stop rebooting and diagnose instead.
2. Reboot once, cleanly. A single soft reboot or full power cycle clears transient glitches. If the fault returns identically, it is not a glitch; move on.
3. Check the free environmental causes. Clean the dust from intakes and heatsinks. Confirm every fan spins and reaches plausible speed, and that airflow direction is correct. Overheating shutdowns from a dead fan or a clogged filter imitate board failure convincingly.
4. Update or reflash the firmware. Firmware corruption from an interrupted update is a documented failure cause, and reflashing with the latest official stock firmware for your exact model is free. This is also the last purely software step, everything after this is hardware.
5. Power off, unplug, and reseat the connections. Reseat the data (ribbon) cables and power connectors for the affected board at both ends. Ribbon cables are fragile and a leading suspect whenever a board shows zero chips: on at least one major vendor's troubleshooting sequence for a 0-chip board, reseating the ribbon cable is step one. Inspect for bent pins, torn traces, and heat discoloration while you are there.
6. Swap cables to see what the fault follows. Exchange the data cable between the dead slot and a working slot. If the fault moves with the cable, the cable is the problem and the board is fine. If the fault stays with the board, the cable is cleared. This single test separates a few dollars of cable from a board-level repair, never buy parts before running it.
7. Check the power supply's delivery to that board. Each hashboard needs a stable 12V supply; a disconnected lead or a failing PSU rail can make a healthy board drop out entirely. Confirm connectors are fully seated and, if a known-good PSU is available, test with it. Poor grounding can also cause detection problems.
8. Isolate boards one at a time. Run the miner with a single hashboard connected, rotating through the boards. If initialization fails only when one specific board is connected, that board is the prime suspect. If a board works in a different slot, suspect the control board port it came from; test the control board last, since board and cable causes are more common.
What success looks like at this stage: you can state the fault in one sentence that names a proven root cause, "the data cable for chain 2 is faulty," "the PSU rail drops under load," "board 2 alone prevents startup", not a symptom. If the sentence still starts with "the hashrate is low," the diagnosis isn't finished.
Stop points: when home diagnosis ends
Some findings end the home procedure immediately, regardless of step order:
- Melted or heat-damaged connectors, a burnt smell, or a breaker that trips. Stop. These are the classic stop-and-book-bench-diagnosis signs; a short on a board is a fire and equipment risk, not a troubleshooting puzzle.
- Visible board damage, blackened components, corrosion, or moisture deposits. The cause may extend beyond the visible damage.
- A partial chip loss that grows. A board missing one to three chips but otherwise stable can reasonably keep running at reduced capacity under monitoring. A missing-chip count that increases over days or weeks signals a cascading failure; that board needs professional repair before it takes neighbours down with it.
- Anything inside the board. Chip replacement, trace repair, and BGA rework require specialized equipment and experience most home operators do not have. Home diagnosis ends at proving which board and which class of fault, the repair itself is bench work.
- A machine still under warranty. For current-generation machines, warranty or professional repair comes first; opening and probing can cost more in voided coverage than the repair saves.
Repair or replace: the two-test decision
With a proven root cause and a real quote, the decision still needs two tests, and a repair must pass both.
Test 1: the cost ceiling. A widely used practical rule holds that repair is attractive when the all-in repair cost, parts, labour, shipping, tax, and the value of downtime, lands below roughly 40 to 50% of the cost of an equivalent working replacement. Above that line, you are paying near-replacement money for a machine with a failure history and no fresh warranty. Treat the rule as a ceiling with judgment, not a law: an isolated, well-understood fault (a fan, a cable, one board) on an otherwise stable machine is the profile that passes; multiple questionable boards, scarce parts, or a repeat failure of the same board is the profile that fails even under the ceiling.
Test 2: the machine must still earn its electricity. This is the test most repair decisions skip. A repaired miner goes back to consuming power at its old efficiency. Run the numbers from the electricity-cost guide at your actual rate: if the machine, fully healthy, would not cover its power cost with a margin you accept, then a cheap repair is still money lost, you have paid to resume losing. This test is why generation matters: current-era machines with an isolated fault often pass both tests; older, inefficient machines frequently pass the ceiling test and fail this one, and for them the honest options are the ones in the exit guide, sell, part out, or repurpose, with one genuine exception: a machine whose heat you actually use, where the heat credit is real because it replaces heating you would have bought anyway.
After the repair: the burn-in verification
A repair is a claim, not a result. Before the machine goes back into a corner for another year, verify the claim:
1. Full chip count in the log. Boot the machine and confirm every chain reports its full design chip count, no missing chips, no "find 0 asic" lines.
2. Rated hashrate, sustained. Watch the status page until reported hashrate reaches the model's rated level, then let it run. A board that hashes for ten minutes and sheds chips under sustained thermal load is not repaired.
3. A defined burn-in window. Run the machine under observation for 24 to 48 hours: no new log errors, no growing missing-chip count, temperatures in their normal band, no restarts. Only after a clean window does the machine go back to being ignored.
4. A written record. Log the date, the proven root cause, what was replaced, and the burn-in result. If the same board fails again, that record is what turns "repair it again?" into an evidence-based no, repeat failures of the same board are a replace signal under the decision framework above.
What success looks like: full chip count, rated hashrate held through the burn-in window, a clean log, and a record on file. Anything less means the repair conversation reopens while the evidence is still fresh, and while any repair warranty is still warm.
The honest bottom line
Most ASIC failures are repairable, and many are cheap, fans, cables, power supplies, firmware. The expensive mistakes are procedural, not technical: deciding from a symptom, buying parts before proving the cause, ignoring the stop points, and repairing a machine whose economics were already dead. Diagnose in order, apply both tests without sentimentality, and make any repair prove itself over a burn-in window. The machine doesn't care about your feelings; your records shouldn't either.