J/TH (joules per terahash) is the energy a miner uses to compute one trillion hashes. You get it by dividing the machine's power draw in watts by its hashrate in TH/s. A lower number means less electricity for the same mining work, so J/TH is the figure that decides how much of a machine's revenue goes to the power bill.
Key points
- J/TH = power (W) ÷ hashrate (TH/s). A 200 TH/s machine drawing 3,500 W is 17.5 J/TH.
- kWh per TH/s per day = J/TH × 0.024. Multiply by your price per kWh for the electricity cost of each terahash per day.
- Only efficiency measured at the wall, including the power supply and fans, matches your electricity bill.
- Published efficiency is a typical value with a tolerance, commonly ±5%, at a stated inlet temperature.
- Efficiency is half the picture. Read it together with price per TH.
What a joule per terahash is
A watt is one joule per second. Hashrate in TH/s is terahashes per second (1 TH = 10^12 hashes). Divide one by the other and the seconds cancel:
W ÷ (TH/s) = (J/s) ÷ (TH/s) = J/TH
So 17.5 J/TH means the machine spends 17.5 joules of electrical energy on every 10^12 hashes, at whatever speed it is running. You will also see the same quantity written as W/TH or J/T. The number is identical.
How to calculate it
- Take the machine's power at the wall in watts.
- Take its hashrate in TH/s.
- Divide power by hashrate.
| Illustrative machine | Power | Hashrate | Efficiency |
|---|---|---|---|
| Older-generation air-cooled | 3,000 W | 100 TH/s | 30.0 J/TH |
| Air-cooled | 3,500 W | 200 TH/s | 17.5 J/TH |
| Air-cooled | 3,500 W | 250 TH/s | 14.0 J/TH |
| Hydro-cooled | 5,500 W | 500 TH/s | 11.0 J/TH |
Total power draw says nothing about efficiency. The hydro unit in the last row draws the most power and is the most efficient, because it produces far more hashrate per watt.
Real specifications work the same way. Bitmain publishes 200 TH/s and 3,500 W for the Antminer S21, which is 17.5 J/TH, and 580 TH/s and 5,510 W for the Antminer S23 Hyd., which is 9.5 J/TH. Both are manufacturer-published figures.
From J/TH to kWh and cost per terahash
Because 1 J/TH is the same as 1 W per TH/s, a machine at 1 J/TH uses 1 W × 24 h = 24 Wh = 0.024 kWh per TH/s per day. That gives two formulas:
kWh per TH/s per day = J/TH × 0.024
cost per TH/s per day = J/TH × 0.024 × price per kWh
| Efficiency | kWh per TH/s per day | At $0.04/kWh | At $0.06/kWh | At $0.08/kWh | At $0.10/kWh |
|---|---|---|---|---|---|
| 10 J/TH | 0.240 | $0.0096 | $0.0144 | $0.0192 | $0.0240 |
| 12 J/TH | 0.288 | $0.0115 | $0.0173 | $0.0230 | $0.0288 |
| 15 J/TH | 0.360 | $0.0144 | $0.0216 | $0.0288 | $0.0360 |
| 20 J/TH | 0.480 | $0.0192 | $0.0288 | $0.0384 | $0.0480 |
| 25 J/TH | 0.600 | $0.0240 | $0.0360 | $0.0480 | $0.0600 |
| 30 J/TH | 0.720 | $0.0288 | $0.0432 | $0.0576 | $0.0720 |
The dollar columns are electricity cost per TH/s per day. Multiply by the hashrate to get a machine or a fleet: 200 TH/s at 17.5 J/TH uses 17.5 × 0.024 × 200 = 84 kWh per day, the same answer as 3.5 kW × 24 h.
The table also prices an efficiency difference. At $0.06 per kWh, 15 J/TH costs $0.0216 per TH/s per day and 20 J/TH costs $0.0288, a gap of $0.0072. Across 200 TH/s that is $1.44 per day, or $525.60 over a year of continuous running. The gap scales with your power price: at $0.03 it is half as large, which is why cheap-power sites can run less efficient hardware.
Revenue per TH/s is the same for every machine on the network, so the cost per TH/s per day in this table is what you compare against it. Use the calculator to work out revenue at current network values.
Stated versus derived efficiency
A specification sheet usually gives three numbers: hashrate, power and efficiency. The efficiency the manufacturer prints is the stated figure. The one you compute from the printed power and hashrate is the derived figure.
They normally match. When they do not, the cause is usually rounding or a table that covers several hashrate versions. For example, Bitmain's specification for the 105 TH/s version of the Antminer S19 Pro publishes 3,250 W and 30.5 J/TH, while 3,250 ÷ 105 = 30.95 J/TH. Neither number is yours to correct. Note the discrepancy and plan on the less favourable one.
On this site, manufacturer-published specifications are shown with their source. If you are comparing a listing elsewhere that gives only hashrate and power, derive the efficiency yourself before comparing it with a stated one.
Wall power versus chip-level claims
Efficiency can be quoted at three points, and they are not interchangeable:
- Chip level: the energy used by the ASIC chips alone. Chip announcements use this figure, and it is always the lowest.
- Board level: DC power into the hashboards, excluding power supply losses and fans.
- At the wall: AC power drawn from the outlet, including the power supply's conversion loss, the control board and the fans. Bitmain and MicroBT label their published figure as power on wall.
Only the wall figure matches what your meter records. As an illustration, hashboards that consume 3,290 W of DC power behind a power supply that is 94% efficient draw 3,290 ÷ 0.94 = 3,500 W from the wall. Quoting the DC figure would make the same machine look 6% more efficient than it is.
One further step applies to sites. Hydro and immersion machines have no on-board fans, so the energy used by pumps, dry coolers and other cooling plant is outside the machine's J/TH. To compare cooling methods fairly, use site-level efficiency:
site J/TH = total site power (W) ÷ total hashrate (TH/s)
Tolerance
Published hashrate, power and efficiency are typical values. Bitmain's listing for the Antminer S21 states ±3% on hashrate and ±5% on power and efficiency. MicroBT's datasheet for the WhatsMiner M60S+ gives its power ratio as 17 J/T ±5% at 25 °C.
A ±5% tolerance on 17.5 J/TH is a range of 16.6 to 18.4 J/TH. Two consequences follow:
- Two models half a joule apart on paper overlap in practice. Do not pay a large premium for a difference smaller than the tolerance.
- For budgeting electricity and sizing circuits, use the upper end of the range, not the typical value.
How efficiency changes in operation
The published figure applies at the manufacturer's reference conditions. Bitmain states the air-cooled Antminer S21 figures at 25 °C inlet air and the Antminer S23 Hyd. figures at 35 °C inlet coolant. Away from those conditions, expect movement:
- Temperature. Hotter inlet air makes fans run faster and chips run warmer, and both raise the energy used per hash. Cooler, cleaner air does the opposite. In general terms, a machine uses the least energy per hash when it is kept comfortably inside its published operating range.
- Operating mode. Many machines offer low-power, normal and high-performance modes. Running chips at lower voltage and frequency improves J/TH and reduces hashrate. Pushing for more hashrate worsens J/TH. MicroBT's WhatsMiner M73 datasheet, for example, lists separate normal and high-performance modes with different power levels.
- Firmware and tuning. Firmware that tunes each chip can shift the trade-off in either direction. Changing firmware can affect warranty, so check the manufacturer's terms first.
- Condition. Dust-clogged heatsinks, worn fans and degraded thermal interfaces raise chip temperature and cost efficiency over time. See ASIC Miner Maintenance.
The only way to know what your machine achieves is to measure wall power and compare it with the hashrate reported by your pool over a day or more.
Browsing by efficiency band
The catalog groups machines into efficiency bands so you can shortlist by power cost:
The lower bands hold the newest generations and many hydro-cooled models. The higher bands hold older generations that draw more power per terahash and suit sites with low electricity prices. Use the table above to turn the difference between two bands into a daily cost at your own rate, then weigh it against the difference in purchase price.