← All guides

Buying a used ASIC: the acceptance test before money moves

Hardware · 9 min read

Most miners that change hands in 2026 are used miners. New hardware carries a premium and a wait; the secondary market carries everything else, honest retired fleets, tired machines one failure away from a repair bill, and a steady supply of outright fraud. The difference between a bargain and a liability is not the listing photos. It is a short acceptance test, run before money moves, with pass and walk-away bands decided in advance. This guide assembles that test from the checks repair shops and buying guides publish, reconciles their conflicting thresholds, and adds the step most of them skip: the all-in price ceiling. Nothing here is financial advice.

Why the thresholds you will find disagree

Search this topic and the numbers fight each other. One buying guide wants live hashrate within about 3% of the rated figure. A repair shop's used-miner guide wants each of the three hashboards reporting within 5% of rated output. Another guide from the same shop calls a 10% total deficit acceptable and reserves alarm for 20% or more. Test duration is quoted at anything from a short supervised run to 24 hours or more. None of these sources is lying; they are answering different questions. The tight numbers describe a healthy machine. The loose numbers describe the point at which a machine is not worth saving at any price. The test below uses both: a pass band, a negotiate band, and a walk-away band, so a result in the grey zone becomes a price conversation instead of an argument.

Before the test: the evidence bundle

If the machine cannot be tested in person, the purchase stands or falls on evidence gathered while the seller still has every incentive to cooperate. Ask for all of the following, and treat refusals as data:

  1. A running video, not photos. The machine hashing, with sound, fan bearing noise is audible, and a handwritten note or the day's date visible beside the serial plate, so the footage is of this unit, now.
  2. The miner's dashboard and the pool dashboard together. The local dashboard can be flattered by firmware; the pool's independent estimate of submitted work cannot. Buying guides ask for both for exactly this reason.
  3. A close photograph of the serial label, legible enough to run through the manufacturer's warranty lookup before the test is even scheduled.
  4. Interior photographs with the cover removed: hashboard connectors, heatsinks, and the power supply's capacitors, in good light. Corrosion can hide from casual photos, which is why this is supporting evidence, never a substitute for the powered test below.

The fraud pattern is consistent enough to list plainly: a price far below every comparable listing, refusal of a live hashrate test or running video, a missing or unverifiable serial, stock photos instead of the actual unit, and payment demanded in crypto only with no escrow. Any two of those together end the conversation.

The acceptance test: seven checks, three bands

Run the machine on stock settings, on the buyer's own pool account where possible, in this order. The first three checks can be done over a live video call; the full test wants the machine in hand under a written return window (see the payment section).

  1. Serial and warranty. Verify the serial against the manufacturer's lookup and note any warranty remaining, even a few months of coverage has real value against repair costs. A serial that will not verify, or has been removed, is a walk-away, full stop.
  2. All boards present, chip count complete. The status page should show every hashboard detected, three on a typical Antminer, with the expected chip count on each. A missing board or a short chip count is not a discount opportunity for a first-time buyer; it is a repair project. Walk away, or price it as one using the repair-or-replace guide on this site.
  3. Sustained hashrate. Compare average hashrate against the model's rated specification over the run, not the flattering instantaneous figure. Pass: within about 5% in total, with each board within about 5% of its share. Negotiate: a 5 to 10% deficit, priced as the repair risk it is. Walk away: 20% or more below rating, the band the repair literature reserves for serious issues. Duration matters more than precision here: run 24 hours or more before judging, because chips that pass a 30-minute test can fail under sustained load as heat soaks in.
  4. Wall power. Put a plug meter on it. A draw more than about 5% over the rated wattage at stock settings points at degraded boards, and it is a permanent tax, paid every hour the machine runs, at the buyer's electricity rate. Dashboard wattage estimates do not count; measure at the wall.
  5. Temperatures and error rate. Chip temperatures should sit under 80°C in normal operation; sustained readings above 85°C point to degraded thermal paste or airflow problems. Read the hardware error count from the log: a small number is normal, but hardware errors above roughly 1 to 2% of total shares indicate failing chips or poor connections.
  6. Physical inspection, powered off. Heatsinks firmly attached; connectors clean, with no burnt pins or melted plastic; no brown or black burn marks on the boards; no bulging capacitors in the power supply; fans spinning freely without grinding or wobble. Water stains or green corrosion suggest an immersion or hydro history the seller did not disclose, a different and harsher life than air-cooled service. Light dust is normal. Packed dust, moisture marks and corrosion are not.
  7. Firmware provenance. Identify exactly what firmware is installed and shown in the dashboard. Unknown third-party firmware can hide faults, or quietly skim a developer fee to someone else's pool. The safe states are stock firmware or a known, named build from a recognised project; anything unidentifiable gets reflashed to stock before the machine earns a cent, and the reflash is the seller's problem to explain, not the buyer's to discover later.

What success looks like: after 24 hours at stock settings, the machine holds hashrate inside the pass band on the pool's dashboard as well as its own, wall power is within about 5% of rating, temperatures and error rates are in the normal bands above, and the serial verifies. That machine has demonstrated the specification it is being sold on, which is the only version of the specification worth paying for.

The all-in ceiling: what the machine really costs

A used price is never the asking price. Build the ceiling before negotiating:

LineWhat to add
Asking priceThe listing figure, in the listing's currency
DeliveryFreight or fuel and time to collect; imported units add tax, brokerage and import charges to the delivered total, not the sticker
Immediate wear partsFans at $10 to 30 each for any that grind or wobble; a power supply at $80 to 150 if its age or condition is doubtful. PSUs age with heat, and a failing one can damage hashboards
Risk reserveA deliberate allowance for the negotiate-band findings from the test: a board running 5 to 10% down is a future repair, priced now or not at all

Then the decision rule: if the all-in total approaches the price of a newer, more efficient model, new or from a seller offering a tested unit with a return window, the used machine has no discount left to justify its wear, and the correct move is to buy the newer machine or walk away entirely. Canadian buyers have one extra gate the US-focused guides skip: document the unit's Canadian electrical approval and have the intended circuit reviewed against the nameplate load before purchase, and classify the unit honestly, new, refurbished, used or modified, because a cleaned enclosure is not proof of refurbishment, and the evidence owed differs by class.

Payment and returns: sequence the deal around the test

The test above only protects a buyer who can still act on its results, so the deal mechanics matter as much as the checklist. Get the cancellation, arrival-damage, dead-on-arrival, warranty and return terms in writing before payment, and be clear on which warranty is actually on offer, a seller's short return window and a manufacturer's warranty are different instruments. Prefer a payment route with recourse or an escrow step for a remote purchase; crypto-only, no-escrow, no-returns is the fraud profile, not a negotiating style. A seller confident in the unit will agree to a few days' test window. Structure the purchase so the 24-hour run happens inside that window, which means the return terms are agreed before the machine ships, not pleaded for after it fails.

FAQ

The seller will only send a dashboard screenshot. Is that enough? No. A screenshot is stale the moment it is taken and shows only what the machine claims about itself. The minimum credible remote evidence is running video plus the pool-side dashboard, followed by the full test inside a written return window.

A board is missing but the price is half of normal. Worth it? Only as a deliberate repair project, priced with the repair-or-replace framework, replacement boards, repair labour and downtime included, never as a cheap miner. Priced as a miner, a short machine is the most expensive kind.

How much does remaining warranty matter? Enough to check every time. Warranty status is verifiable from the serial in minutes, it transfers real repair risk back to the manufacturer, and a seller who cannot produce a verifiable serial has answered a more important question than the warranty one.

Next: Repair or replace a failing ASIC? →