A first ASIC is the easiest piece of mining to get wrong, because the metric everyone fixates on — efficiency, watts per terahash — is genuinely the right headline number, which lulls you into thinking it’s the only number. It isn’t. The decisions that actually determine whether you keep the machine or sell it at a loss in six months are the ones nobody puts on the box: firmware lock-in, resale value, your electricity arrangement, and how loud the thing is in the room you have to put it in.
This is a capability-and-cost-of-running guide. I’m not going to tell you what you’ll make — that depends on difficulty, your power price, and a dozen variables outside this article, and frankly telling you would be guesswork dressed as advice. I’ll tell you what each machine can do, what it costs to run, and where the spec sheet quietly lies.
01 · Efficiency is the headline, not the whole story
Watts per terahash (W/TH) is the number that matters most, so start there. It tells you how much power the machine burns to produce a given amount of work, and since power is your largest ongoing cost, a more efficient machine costs less to run for the same hashrate. Two machines at the same hashrate but different efficiency are not close to equivalent — the less efficient one is quietly more expensive every hour it’s on.
But efficiency is a ratio, and a ratio hides two absolute numbers you also need: total power draw and total hashrate. A 3,400W machine and a 1,500W machine can have identical efficiency and wildly different demands on your wiring, your cooling, and your patience. Read the ratio for the headline; read the absolutes for whether you can actually run it.
02 · Spec-sheet efficiency vs the room you’re in
Here is the uncomfortable part. Spec-sheet efficiency is measured at the wall under ideal conditions — a cool ambient, a clean power supply, firmware fresh. The machine in your garage in August, drawing through a long extension lead, with eight months of dust on the heatsinks, does not hit the spec sheet. The gap is real and it’s always in the same direction: worse.
The numbers below are illustrative, current-generation spec-sheet figures for broad classes of machine — not a live price list, not a recommendation, and not what any specific unit will deliver in your room. Treat them as the shape of the trade-off, not gospel.
| Class (illustrative) | Spec-sheet efficiency | Total power | Realistic in a warm room |
|---|---|---|---|
| Older flagship (S19-class) | ~29 W/TH | ~3,250 W | noticeably worse once warm + dusty |
| Current flagship (S21-class) | ~17 W/TH | ~3,500 W | closer to spec if well-cooled |
| Mid-tier (WhatsMiner M50-class) | ~26 W/TH | ~3,300 W | sensitive to ambient temp |
| Newer mid (M60-class) | ~18 W/TH | ~3,400 W | strong if airflow is good |
The pattern across all four: the newer the silicon, the better the efficiency — and the more that efficiency depends on you actually cooling it properly. An older machine in a cool, clean environment can out-perform a newer one you’ve stuck in a hot cupboard. Ventilation isn’t an accessory; it’s part of the efficiency figure.
03 · Firmware lock-in — the trap with no spec line
Stock firmware is what ships on the machine. Custom firmware (third-party builds) can unlock tuning — undervolting, per-chip frequency control, better fan curves — that meaningfully changes a machine’s efficiency and noise. This sounds like a pure win, and sometimes it is. The trap is lock-in: some manufacturers sign their firmware, lock the bootloader, or void the warranty the moment you flash anything that isn’t theirs. Some custom firmware takes a developer fee off the top. Some bricks the machine if you get the flash wrong.
None of this is on the spec sheet. Before you buy, find out: can this model run custom firmware at all? Does flashing it void the warranty? Is the stock firmware tunable enough that you don’t need to? A machine that’s locked to mediocre stock firmware is a worse buy than its efficiency number suggests, and a machine with a healthy custom-firmware ecosystem is a better one.
04 · The electricity arrangement is the real constraint
Most first-timers think about the machine and forget about the supply feeding it. That’s backwards. A 3.4kW ASIC pulls serious current, and the questions that decide whether you can run it have nothing to do with hashrate:
- Single-phase vs three-phase. Domestic supply is usually single-phase; a single high-draw machine is fine, but several quickly approach the limit of a domestic circuit. Three-phase (common in commercial units) spreads the load and is what serious deployments use. Know which you have before you buy the machine, not after.
- Circuit headroom. A machine’s continuous draw should sit well under the circuit’s rating, with margin. Continuous loads near a breaker’s limit trip it — and ASICs run continuously by definition.
- Your tariff. Power is the dominant running cost, so the price per unit you pay swamps almost every other factor. A fixed-rate contract, a time-of-use tariff, a separately-metered supply — these change the cost of running the machine far more than a few percent of efficiency does.
You can optimise the machine all you like; if the supply can’t feed it or the unit price is high, none of it matters. Sort the electricity arrangement first, then choose a machine that fits it.
05 · Noise, heat, and where it physically goes
These machines are loud — flat-out, a flagship ASIC is in the region of a loud vacuum cleaner that never switches off, dominated by fans screaming to shift the heat. People consistently underestimate this. A machine that’s tolerable in a detached garage is intolerable in a spare bedroom, and there is no firmware setting that makes 70-plus decibels of continuous fan noise pleasant. If the only place you have is inside your living space, the noise floor is a hard constraint — factor it before efficiency, because an efficient machine you can’t bear to be near is a machine you’ll switch off.
Heat is the same problem from the other side. Every watt a machine draws comes out as heat into the room. A 3.4kW machine is a 3.4kW heater. In winter that’s a bonus; in summer it’s a ventilation problem that, left unsolved, drags the machine’s real-world efficiency below its spec sheet (see section 02 — hot machines run worse). Plan the airflow before the machine arrives.
06 · Resale, warranty, and availability
The last cluster is about the machine as an asset you might sell on. Resale value varies enormously by manufacturer and model — a unit from a maker with strong secondary-market demand holds value far better than an obscure one, which matters because hardware moves on and you may want to recover some of the cost. Warranty length and whether the manufacturer honours it (and from where — import duties and RMA shipping across borders are real) belong in the buying decision. And availability: a machine you can actually buy from a reputable seller, today, at a sane markup, beats a marginally better one that only exists on a six-week pre-order from a vendor you can’t vet.
Get those six right and you’ve made a competent first purchase. Fixate on the headline efficiency number alone and you’ll learn the other five the expensive way — which is, unfortunately, how most people learn them.