Mining Efficiency in J/TH: Four Checks for Energy, Work and Cost

Check one: calculate compatible units

Divide watts by TH/s to obtain J/TH, because one watt is one joule per second and the seconds cancel. A hypothetical 3,000 W divided by 150 TH/s equals twenty J/TH. This measures energy per computational work, not a monthly electricity bill or quantity of coins.

Convert three kilowatts to three thousand watts before using that expression. Dividing three by one hundred fifty and labeling it J/TH introduces a thousandfold error. Likewise, 0.15 PH/s is 150 TH/s. Preserve units beside every input rather than relying on an abbreviated table heading.

Compare the same algorithm. One hash attempt in another algorithm does not automatically represent equivalent economic output. Record network, model and mode, then use a revenue model appropriate to the actual work rather than creating a universal profitability ranking from one energy ratio.

Lower power does not always mean better efficiency

Assume mode one delivers 150 TH/s at 3,000 W, or twenty J/TH. Mode two delivers 120 TH/s at 2,520 W, or twenty-one J/TH. It consumes less total power but more energy per unit of work because the rate fell proportionally more than power.

Mode three delivers the same 120 TH/s at 2,160 W, or eighteen J/TH. It uses less power and less energy per work than mode one while still producing less work per hour. Comparing absolute daily output therefore asks a different question from comparing efficiency.

These are independent arithmetic examples. Actual mode changes need the manufacturer’s instructions, supported limits and warranty conditions. The lowest invented ratio in a table does not establish that real equipment can reach it under another environment.

Check two: define where power is measured

Power can be measured at the wall, after the supply or at component level. Those boundaries include different losses and supporting equipment. A chip-level ratio should not be compared directly with a complete machine’s wall ratio as though their scope matched.

The BITMAIN S19 Pro specification page provides an example with wall-power, reference-temperature and tolerance labels. Its value here is demonstrating the need to read measurement conditions, not quoting an ideal ratio as universal performance.

If a wall measurement already includes the supply and internal fans, do not add those losses again. External pumps, room fans or cooling may be outside that boundary. State whether the analysis concerns a device, room or complete site.

Add a hypothetical dedicated 300 W auxiliary load to the 3,000 W machine at 150 TH/s. Device efficiency remains twenty J/TH, while the expanded boundary gives 3,300 ÷ 150 = twenty-two J/TH. The calculation changed scope rather than changing the machine’s own specification.

Check three: match energy and work over the same period

An instantaneous power reading and a daily hashrate average can be inconsistent if operation stopped or settings changed. A period-level comparison divides energy consumed during that period by total relevant work during it, with compatible units.

Assume three kilowatts for a full twenty-four hours, consuming seventy-two kWh. That equals 259.2 million joules. A constant 150 TH/s over 86,400 seconds produces 12.96 million TH. Dividing energy by work returns twenty J/TH, confirming the instantaneous ratio under the constant assumptions.

If pool-accepted work is lower than local work, energy per accepted work can also be calculated, but label that metric. Do not compare it silently with a local hardware specification. Rejection, connectivity or downtime can change the accepted denominator without proving the chips became less efficient at local computation.

Supporting equipment can also consume energy when hashing stops. Excluding stopped hours produces an operating-only measure, while the accounting period still bears that electricity cost. Preserve both when they answer different questions. The hashrate guide explains period and acceptance boundaries.

Read tolerances and measurement uncertainty

A nominal specification can carry tolerances and reference conditions. Match model revision, firmware, mode and environment before treating a small deviation as a fault. Combining the highest allowed rate with the lowest independent allowed power does not automatically create a promised efficiency for every unit.

For measurements, record ambient conditions, duration, load and meter precision. A small estimated difference is not well supported when instrument uncertainty or normal variation is larger. Repeated comparable periods and accumulated energy can be more useful than one attention-grabbing instantaneous number.

Check four: convert a technical difference into a stated cost scenario

Assume two hypothetical devices each deliver 150 TH/s. At twenty J/TH, one uses 3,000 W. At eighteen J/TH, the other uses 2,700 W. The difference is 300 W or 7.2 kWh over twenty-four operating hours.

At an invented energy tariff of 0.1 monetary units per kWh, daily energy-charge savings are 0.72. If the more efficient device costs an additional one hundred, dividing by 0.72 gives approximately 139 operating days for those electricity savings to equal that price difference under fixed assumptions.

This is not complete investment payback. It excludes installation, supporting equipment, maintenance, resale, downtime and possible revenue differences. A tariff can also include time bands, tiers, demand and fixed charges. Read the relevant agreement rather than applying one advertised unit price to the entire bill.

State which constraint the comparison holds fixed

One study can hold work constant, another total power, device count or purchase budget. A more efficient unit producing less total work can rank differently under those questions. State the constraint before presenting a result as the preferred option.

Build a row with algorithm, mode, local and accepted work, power boundary, period energy, operating time and cost. Distinguish measured values from specifications and assumptions. Use several revenue and tariff scenarios instead of extending a single day into a fixed future year. The ratio is useful when its boundaries and financial implications remain explicit.

Continue reading