Six Home Mining Specs That Fail the Arithmetic

None of the six claims below is a lie. Each is a number that was true once, or was copied from somewhere, or was produced by dividing two other numbers and then repeated until it read as measurement. Each collapses under a calculation you can do on a phone. One of them is ours.

The point is not the six examples. It is the three checks that catch them, because new claims appear faster than anyone can audit them.

Key takeaways

• Divide watts by terahash. Nothing on the market today beats roughly 10 J/TH. A published figure far below that is an error, not a breakthrough.
• Compare hashrate against chip count. A single BM1370 does about 1.2 TH/s, so a claimed dual-chip device topping out at 1 TH/s has one chip.
• Date every comparison. Prices in this market fell roughly 38% in six months, and comparison copy does not update itself.
• A “4.8 TH/s at 35 W” claim implies 7.29 J/TH, which nothing achieves. Real stock is about 71 W.
• The universal “15 J/TH” for a Gamma is 18 divided by 1.2 — two nameplate numbers, not a bench result.
• Our own pages used 17 W for a Gamma, inherited from vendor documentation rather than measured. We say so below.

The three checks

Before the examples, the method. All three take seconds and need no equipment.

Check one is the workhorse. For orientation: current BM1370-class hardware sits near 15 J/TH, the 3nm BM1373 generation reaches roughly 10, and the older BM1397 era ran in the 30s. Our ASIC chip evolution guide has the full lineage. If a claim implies single digits, something is wrong with the claim.

1. The 35-watt quad-chip miner

The claim: a NerdQAxe++ delivers 4.8 TH/s at “under 35 W”. It appears on the German and Spanish product reviews of a major European retailer, with a monthly running cost calculated from it.

The arithmetic: 35 ÷ 4.8 = 7.29 J/TH. Nothing on the market reaches that. The most efficient industrial hardware in production is around 9.5 J/TH on 3nm silicon under hydro cooling, and the NerdQAxe++ runs BM1370 chips on air.

What is actually true: at true stock settings the board runs about 4.8 TH/s at roughly 71 W, which is 14.8 J/TH. The 100 W figure sometimes quoted belongs to the factory-overclocked configuration delivering closer to 6 TH/s. Our NerdQAxe++ guide covers the revisions.

What it costs a reader: the running-cost table built on 35 W understates the electricity bill by about half. At the EU average of €0.2896 per kWh that is roughly €15 a month rather than €7.40. Our European cost breakdown has the by-country figures.

2. The 3nm chip that is 5nm

The claim: the BM1370 is a 3nm chip.

The correction: it is a 5nm part from the Antminer S21 Pro generation. Bitmain’s first 3nm SHA-256 chip is the BM1373, which ships in the S23 series.

Why it matters: a process shrink is the single biggest efficiency lever in mining silicon, and this one is worth roughly a third. Believing you are buying 3nm when you are buying 5nm means expecting about 10 J/TH from hardware that delivers about 15. Our BM1373 era piece covers what actually changed.

The check that catches it: the same one. If a device claims a 3nm chip and quotes 15 J/TH, one of those two numbers is wrong.

3. The dual-chip device with one chip

The claim: the current GekkoScience Compac A2 uses dual BM1370 chips. One retailer says two, another says one.

The arithmetic: a single BM1370 in a Bitaxe Gamma produces about 1.2 TH/s. The Compac A2 tops out at 1 TH/s when pushed to its limit at 500 MHz and 300 mV. Two of the same chip would land nearer 2.4 TH/s.

Verdict: one chip. You do not need to open the device or trust either retailer. When sources disagree on a component count, the hashrate settles it, because silicon output per chip is one of the most stable numbers in this industry. Our USB stick comparison works through the rest of that device’s numbers.

4. The best-value claim that expired

The claim: the Bitaxe Duo 650 offers “the best value per terahash in the entire Bitaxe lineup”. It still appears on its seller’s own pages.

When it was written it was true. In March 2026 a Bitaxe Gamma listed at $97.98 for 1.2 TH/s, which is $81.65 per TH/s. The Duo at $104.29 for 1.63 TH/s was $63.98. The Duo won comfortably.

What changed: the Gamma now lists from $61.00, which is $50.83 per TH/s. The Duo’s own price barely moved. The claim did not change because sentences do not update themselves.

This one is the most instructive of the six because nobody did anything wrong. A comparison was accurate, a price moved, and the copy stayed. It is why the third check exists: every comparison has a shelf life, and it is rarely printed on the label. Our Duo 650 review covers what the board still genuinely offers, which is density rather than value.

5. The power supply advice that contradicts its own reviews

This one is not arithmetic, but it is the same shape of problem.

The claim, from a seller’s setup guide: use the provided power supply, because third-party supplies may damage the device or cause hazards.

The counter-evidence, on the same seller’s own product page: six of thirty-two verified reviewers independently recommend replacing the included supply with a Mean Well unit. Two describe the stock adapter running hot. One reports losing two boards to power events while other units on the same surge protector survived.

Both are real. The support desk is warning against the genuine hazard of a mismatched voltage — a 12 V supply on a 5 V board destroys it instantly, and that warning should be taken seriously. The reviewers are reporting sustained use of a correctly specified but marginal adapter. The two are not actually in conflict; they are answering different questions, and the page never reconciles them.

The reading: never substitute a supply of the wrong voltage, and do consider a better one of the right voltage. Our Duo guide covers why 5 V systems are unforgiving of voltage drop.

6. Ours: the 17-watt Gamma

Our own pages have used 17 W for a Bitaxe Gamma, a figure inherited from vendor documentation rather than from a meter. It is not wrong as a conservative planning number. It is also not measured, and we should have said which it was.

The wider problem is worse than one figure. Here is every number available for the same board at stock settings:

Four sources, four answers, spanning 11.92 to 16.64 J/TH for one device. They cannot all be right, and we cannot tell you which one is. The measurements were taken at different points in the circuit, on different silicon, at slightly different frequency and voltage settings, and only one of them was taken at a wall socket.

Note also where the famous number comes from. 15 J/TH is exactly 18 divided by 1.2 — two nameplate figures dividing each other. It has been repeated so widely that it reads as a bench result, and it is not one. It appears in our own articles fifteen times.

Going forward we will say which category any power figure belongs to: nameplate, bench, or measured at the wall. Our power measurement piece explains why the three differ by roughly a fifth at each step.

What to do with this

Three habits, in the order they save you money.

1. Run check one on every listing. Watts divided by terahash. If it lands in single digits, close the tab or ask the seller which number is wrong.
2. Treat comparisons as dated documents. A cost-per-terahash table without a date is a historical artefact. This applies to ours: prices in this market moved 38% in six months, and every table we publish is a snapshot.
3. Measure the thing you actually own. A 20 dollar energy plug settles every argument on this page for your specific device, and your hashrate is already on your dashboard.

A closing observation about why this happens, because it is not incompetence. Home mining is a small market with fast product cycles, most specifications originate from a single manufacturer document, and everyone downstream copies it. One figure entering the ecosystem wrongly gets repeated by every retailer, review site and guide, including ours. There is no adversarial process, so nothing catches it.

Which means the arithmetic is the only defence, and it is available to everyone. Divide the watts by the terahash. It takes three seconds and it is the single most useful thing a buyer in this market can do.