What Is an ASIC Miner?
Open the box: four parts
Undo a dozen screws and the mystery mostly evaporates — there are four things in there.
Hashboards do all of the mining. Usually three per machine — some designs use four — each a dense circuit board carrying on the order of a hundred ASIC chips wired in chains: a current flagship-class board carries roughly 108 chips, so a whole machine holds 300+ (community teardown figure, not a manufacturer page). In library terms: the chips pull the drawers, and every other part of the box exists to keep them pulling. One reading habit worth installing now: chip count is a layout choice, not a score — more chips is not more speed, and comparing chip counts across different algorithms compares nothing at all.
The control board is the only real computer inside: a router-class ARM board running Linux (the exact chip varies by generation; community repair documentation). It fetches work from your pool over Ethernet, spreads it across the chips, drives the fans, and serves the web dashboard you open from any browser on your network — and it hashes nothing. There is no screen and no PC attached: wall power, an Ethernet cable and pool credentials, and the machine runs standalone.
The power supply is built in and sized in kilowatts. The row to find on any spec sheet is input voltage: modern flagships list 220–277 V (manufacturer spec, via secondary listings) — an appliance circuit, not a bedroom outlet. Whether your building can feed one is M1's home-mining question; this lesson only teaches you to read the row. One anatomical note to bank for next lesson: sheets sometimes quote power at the wall and sometimes at the boards, and the two differ because the PSU's conversion losses and the fans sit between them — catching which number the marketing chose is lesson 2's job.
The fans — typically four high-static-pressure units at machine-class level (per-model counts vary between sources, so we don't quote one) — are there because of physics you'll meet in the afterlife section: effectively every watt the box draws comes back out as heat, and the chips sit millimeters from their own furnace. They are also the machine's only moving parts and its first wear item — the hour ratings, and what they mean in a real room, live in the lifespan lesson. How loud four of them get is a lesson of its own — miner noise, really — so this page prints no decibel numbers, deliberately.
The same anatomy scales in both directions. A Bitaxe is one of these chips on a desk-sized board with one small fan — its own lesson — while a flagship appliance is 300+ chips in a steel wind tunnel. Same parts list, different multiplier.
You open a three-board machine and one hashboard has died. What's the honest diagnosis?
Not a computer: why the trick can never change
A CPU has an instruction set — hand it any program and it runs it. A mining ASIC has no instruction set to hand anything to. The hashing pipeline is not software running on the chip; it is the chip — the algorithm is laid down as physical wiring at the fabrication plant, the way a key's teeth are cut into the metal. No screen, no usable operating system, no graphics output, nothing to install anything on. It cannot be flashed into a gaming rig, an AI box, or a different-algorithm miner, because there is no software layer where the algorithm lives.
This is the part most write-ups skip: they say “specialized” and leave you believing the specialization is a settings problem some sufficiently determined person could undo. It is not — it is physically irreversible. Firmware on a miner steers voltage, frequency, and how work is spread across chips that can each do exactly one computation. Custom firmware can make a machine faster, quieter, or more efficient at its one job; no firmware can give it a different one. When a listing says a machine “supports many coins,” it means coins sharing the same algorithm — never a new trick.
Mining stops paying and the machine still runs perfectly. Which second career is actually open to it?
The algorithm lock, live
Here is the sharp version of the claim, and it cuts in both directions: mining machines are not coin-locked, they are algorithm-locked. SHA-256 silicon can search Bitcoin's library — and any other chain that pays for SHA-256 work. Scrypt silicon searches Litecoin's, and because Dogecoin has accepted Litecoin's proof-of-work since September 2014 (merged mining), one Scrypt machine's pulls pay in both coins at once. Retailers name machines after coins because coins sell; the honest unit is the algorithm. An Antminer L7 — Bitmain's own sheet says Scrypt, full stop — contributes not “very little” to Bitcoin but exactly zero drawers per second: wrong library. Every other L7 number belongs on its rig page, not here.
Watch the three libraries being searched right now, from our own nodes:
The two Scrypt tiles sit about 3% apart and move together — one Scrypt machine searches both libraries at once, which is merged mining showing up in our own node data rather than being asserted. Two estimators reading one fleet through two chains' block times will never agree exactly; the merged-mining lesson has the mechanics. Fallback figures dated 2026-09-27.
And the unit gap does the rest of the arguing: Bitcoin's library is being searched about 368,000× faster than Litecoin's (recomputed live; fallback dated 2026-09-27) — EH/s against PH/s, a thousand-fold apart on units alone. That is not one race with Bitcoin in front; it is two different sports in two different stadiums, and no machine holds a ticket to both. The whole BTC library's pull rate, charted with its history, lives at /btc/hashrate.
A friend owns “a Litecoin miner” and wants to point it at Bitcoin, the bigger prize. What does the machine actually do there?
The bet you buy with the box
Now read the property at the top of this page as money instead of engineering. Because the chip can do nothing else, the machine's entire value is a bet on exactly one thing: that its algorithm's library keeps paying enough to cover the watts. A GPU that stops paying in one market gets resold into gaming or rendering; an ASIC that stops paying has no second market for its computation. Nothing breaks when that happens — the machine hashes exactly as well as the day it shipped. The bet underneath it expired. That cliff is structural — a property you buy with the box, not a defect you might dodge — and this lesson only establishes that it exists. When and how the bet dies — difficulty creep, the efficiency march, halvings — is the lifespan lesson, where this module's “scrap overnight” teaser resolves.
The afterlife: three exits, all real
“Scrap overnight” is how the bet reads on paper. What physically happens to a machine that stopped paying is more interesting — and not something a retailer page will walk you through.
Exit one: parts. A real repair trade works at the chip level — public repair manuals, donor boards, component-by-component diagnostics — because a “dead” machine is usually one dead hashboard, and one dead board is two live ones plus a donor. The depth of that public repair literature is the receipt for the claim.
Exit two: migration. A machine that stopped paying at your power price may still pay at somebody else's, so uneconomic iron tends to move toward cheaper electricity rather than die. One sentence is all this gets here; the framework for when a machine crosses that line belongs to the lifespan lesson.
Exit three: heat. First-law physics: the computation stores nothing meaningful, so effectively every watt a miner draws leaves the box as heat — a 3.5 kW machine warms a room like a 3.5 kW space heater. Heat-reuse projects claim 90–96% of it is recoverable, air-cooled to immersion (industry-measured claims; the physics is only responsible for the heat existing). The honest accounting cuts both ways: the heat is real, and it is not free — it costs exactly the electricity on your meter, and any hash income is the only new line in the ledger. Heater-miner marketing oversells the first half; “miners waste heat” critics undercount the second.
And the e-waste receipts, both sides labeled. The number that travels is ~30.7 kilotonnes of mining e-waste per year, from a peer-reviewed estimate (de Vries & Stoll, Resources, Conservation and Recycling, 2021, as of May 2021) — built on an assumed average device life of 1.29 years. The repair and hosting trade disputes exactly that assumption, claiming 4–7-year physical lifespans — an industry figure from commercially interested sources, so it carries a label too. Both can be honest at once, because they measure different deaths: a machine dies economically when the bet expires and physically when boards or silicon fail, often years apart — the three exits above live in that gap. The split is the opening question of the lifespan lesson.
A retired miner runs as a space heater all winter. What's the honest accounting?
FAQ
- Why can't an ASIC miner mine other coins?
- It can — but only coins that share its algorithm. The algorithm is etched into the chips at the factory, so a SHA-256 machine can mine Bitcoin or any other SHA-256 chain, and a Scrypt machine earns Litecoin and Dogecoin at the same time through merged mining. What it can never do is cross the line: an Antminer L7's Scrypt silicon contributes exactly zero hashes to Bitcoin. Machines are algorithm-locked, not coin-locked.
- Is an ASIC miner just a computer? Can I use it for anything else when it stops mining?
- No. Its chips have no instruction set — the hashing pipeline is the physical wiring, so there is no software layer where a new job could be installed. The only computer inside is a router-class control board that schedules work. A retired miner's remaining uses are as parts (hashboard and chip repair is an established trade), as a machine for someone with cheaper electricity, or as heat — physics turns effectively every watt it draws into warmth.
- What is actually inside an ASIC miner?
- Four things: hashboards (usually three, each carrying on the order of a hundred ASIC chips — a current flagship holds 300+ in total, per community teardowns), a router-class ARM Linux control board that talks to your pool and hashes nothing, an integrated kilowatt-class power supply, and high-pressure fans. No PC attached: wall power, an Ethernet cable and pool credentials, and it runs standalone. The chips pull the drawers; everything else keeps them pulling.
- What happens to old or dead ASIC miners?
- Three honest exits: parts (a public chip-level repair trade treats dead units as donors), migration (machines that stop paying at one power price keep running somewhere cheaper), and heat reuse. The scary e-waste headline (~30.7 kt/yr) comes from a 2021 peer-reviewed estimate that assumed a 1.29-year average life — an assumption the repair industry disputes with 4–7-year physical-life claims. Both numbers carry labels here; the economic-vs-physical split is the lifespan lesson's opening question.
- Why do miners become worthless overnight if they still work?
- Because the chip can do exactly one job, the machine's entire value rides on that job still paying more than the electricity it eats. Nothing broke — the bet expired. How that happens — difficulty creep, the efficiency march, halvings — is its own lesson: miner lifespan.