Your Miner's Real Power Draw: Four Numbers, One Meter

Ask four sources how much power a Bitaxe Gamma uses at 1.2 TH/s and you get four answers: 12 W, 14.3 W, 17 W and 18 W. Same board, same settings, a 1.5x spread. One of them is what your utility meter records. It is not the one on your dashboard, and it is not the one on the box.

This matters more than it sounds, because almost every calculation a home miner makes — monthly cost, efficiency comparisons between boards, whether a fleet makes sense — starts from a wattage figure that was never measured.

Key takeaways

• Measured wall draw for a stock Gamma is 14.3 W at 1.20 TH/s, from a seven-day Kill-a-Watt average in a 22 °C room.
• The AxeOS dashboard understates by 18 to 19 percent, consistently, at every tuning tier. It measures the chip, not the socket.
• The universal “15 J/TH” figure is arithmetic, not measurement — it is 18 W divided by 1.2 TH/s. Measured at the wall the same board is closer to 11.9 J/TH.
• Overclocking does not improve wall efficiency. Measured J/TH worsened monotonically across four tiers, 11.92 to 13.53, despite the overhead fraction shrinking.
• Air conditioning roughly adds a third to effective running cost in cooling season, because you pay to make the heat and again to remove it.
• Our own pages use 17 W for a Gamma. That sits above the measured figure, and we explain below why we are moving to measured numbers.

The four numbers

All four describe a Bitaxe Gamma at roughly 1.2 TH/s. Measured figures from a seven-day Kill-a-Watt P3 average published 6 May 2026; vendor figures from Solo Satoshi and D-Central listings read 7 August 2026.

Notice what the third row means. Our own pages have been using 17 W, a figure inherited from vendor documentation rather than from a meter. It is not wrong as a conservative planning number, but it is roughly 19 percent above what a meter records, and we would rather say so than quietly keep using it. Where we publish power figures from here on, we will say which of these four categories they come from.

And notice the fourth row’s arithmetic. The “15 J/TH” that appears in nearly every Bitaxe listing, including several of our own pages, is exactly 18 divided by 1.2. It is a spec sheet dividing itself, not a bench result. The measured figure is better — which is the part vendors could advertise and do not, because 15 J/TH is the number the whole market already quotes.

Why the dashboard is lower

AxeOS is not lying. It is answering a different question.

The dashboard’s power reading is calculated as voltage times current at the ASIC input. That is exactly the right number for tuning: it tells you what the chip is consuming, so you can compare two chips or two frequency settings on equal terms. It is the wrong number for your electricity bill, because a board is more than its ASIC.

Add those to a 12 W chip and you land near 14.3 W at the socket. The gap is not an error in either number. It is the difference between the chip and the system.

What measurement across four tuning tiers shows

Seven-day averages per tier, fan on auto, 22 °C ambient, published 6 May 2026.

The understatement is remarkably stable: between 18 and 19.2 percent at every tier. That gives you a usable rule of thumb. Take the dashboard figure and add a fifth. It will not be exact for your brick and your ambient temperature, but it will be much closer than the dashboard alone.

One conclusion in that dataset does not survive the arithmetic

The source that published these measurements draws an appealing inference from them: because fixed overhead becomes a smaller share of total draw as you overclock, wall-perspective efficiency should improve until the chip itself degrades, with a crossover around the third tier.

The overhead part is correct. The conclusion is not, and the same table disproves it.

Wall efficiency across the four tiers runs 11.92, 12.20, 12.59, then 13.53 J/TH. That is monotonically worse at every step, a 14 percent decline from stock to aggressive. There is no crossover, because the chip’s efficiency loss outruns the shrinking overhead fraction from the very first step.

This is worth stating plainly because the wrong version is comforting: it would mean overclocking is free from the wall’s point of view. It is not. Overclocking buys hashrate and costs efficiency, measured at the chip or at the socket. Whether that trade is worth it is a separate question, and for a lottery device the answer is often yes — more hashrate is more tickets, and the electricity difference is small. But it should be made with the right numbers.

What the gap costs

Monthly cost for one stock Gamma running continuously, 30.44-day month.

On one device the gap is small change. It compounds with fleet size: five stock Gammas running a year cost about $16 more than the dashboard suggests at US-average rates, and about $35 more at $0.35 per kWh. If you are planning a stack of ten boards, the error is a meaningful fraction of what you thought the whole exercise cost.

For the same arithmetic applied to specific boards, our Duo 650 guide and Hex 1300 review both note where owner-measured draw ran above the published figure, and the block odds calculator lets you put your own wattage in.

The cost nobody counts

Every watt a miner consumes leaves the board as heat. That is not a metaphor: a 17 W miner is a 17 W heater, putting roughly 58 BTU/h into the room.

In winter that heat is worth something. If you heat with electric resistive heating, it is worth exactly what you paid for it, and the mining is free. Our hashrate heating page covers when that argument holds and when it does not.

In summer, with air conditioning running, the sign flips. You pay to generate the heat and pay again to remove it. At a typical air conditioner coefficient of performance around 3, removing 17 W of heat costs roughly another 5.7 W, so effective running cost rises by about a third during cooling season.

Almost nobody includes this. If you live somewhere with a real cooling season, your annual electricity cost is not twelve times the monthly figure; it is nine normal months plus three at roughly 133 percent. Worth knowing before you size a fleet in a warm climate. For larger deployments, our cooling comparison covers what changes when heat stops being an afterthought.

The brick matters too

A cheap adapter runs 80 to 85 percent efficient at part load. A quality one reaches 92 to 94 percent. That difference sits upstream of everything else, so it multiplies every other number on this page.

On a single small miner the annual difference is a couple of dollars — not a reason to act on its own. Across a fleet it compounds, and for boards running two chips on a 5 V rail the supply is already the component owners most often replace, for reliability rather than efficiency. Both reasons point the same way.

One caveat on the measurements above: a wall meter reads everything downstream of the socket, which includes the brick. So the 14.3 W figure already contains whatever losses that particular adapter had. A worse brick would push the same board higher without the chip doing anything different. When comparing your reading to ours, the supply is part of what you are comparing.

How to get your own number

1. Get a meter. A Kill-a-Watt P3 costs around $25, and energy-monitoring smart plugs from TP-Link Kasa, Shelly or similar work as well and log automatically.
2. Put it between the wall socket and your power supply, not between the supply and the board. You want the brick inside the measurement.
3. Leave it a week. Fan behaviour, ambient temperature and pool difficulty all move draw around; a single spot reading is noise.
4. Read the average, not the instantaneous value. Multiply by 24, by 30, and by the rate printed on your electricity bill.
5. Record the settings alongside it. Frequency, core voltage and ambient temperature, or the number means nothing six months from now when you have tuned the board.

If your measured figure comes in far above the table here, that is diagnostic rather than disappointing: an unusually high reading points at a poor supply, a fan running flat out because of a thermal problem, or a board throttling. Our troubleshooting guide works through those in order.

What to do with this

Three practical adjustments.

• For your electricity budget, use the dashboard figure plus 20 percent, or better, a measured one. Never the vendor nameplate, which is conservative in the other direction.
• For comparing two chips or two tunes, use the dashboard figure. It is the right tool for that job and adding overhead only adds noise.
• For comparing two different boards, insist on wall figures for both or dashboard figures for both. Mixing them is how a board with a big fan and a big display gets to look more efficient than it is.

And a general point that outlives these particular numbers. In home mining, published figures cluster around whatever the first vendor published, and get repeated until they read as measurement. The Gamma’s 15 J/TH is a division problem that became a fact. A $25 meter and a week of patience produces something better, and it is one of the few edges available to anyone with a plug socket.