Hashrate, Watts, and J/TH: The Only Three Numbers That Matter

That one row holds every operator decision: what the machine contributes, what it costs to run, whether your wiring can even feed it, and how much of the promise is a rounding convention. The previous lesson opened the box; this one reads its label — the three big numbers first, then the four fine-print traps.

The unit ladder: drawers pulled per second

Hashrate is a speed: drawers pulled per second in lesson 1’s library — one pull, one fingerprint checked (that’s a hash). The ladder climbs by ×1,000 per rung, and mixing up rungs is the most common spec-reading error on the internet:

Two rungs matter to an operator: machines are quoted in TH/s, the network in EH/s — and one EH/s is a million TH/s, because PH/s sits between them. Here is the denominator every machine competes against:

So the 270 TH/s flagship above holds about 1 in 3.59 million of the network’s drawer-pulls (repainted live; the dated snapshot holds otherwise). “My miner is fast” means nothing until it is divided by that number — the network figure’s full story, and why every chart disagrees about it, lives at /btc/hashrate.

The three numbers

Hashrate is your share of the work, and so of the wage. What one TH/s earns per day has a market name — hashprice — with its own live page; this lesson stays on the machine’s side of the ledger. Watts are what you pay: every second, around the clock, block or no block. J/TH (glossary) is the ratio that decides survival — the energy burned per trillion pulls.

The identity behind it is plain dimensional arithmetic. A watt is a joule per second, so dividing watts by terahashes-per-second cancels the seconds and leaves joules per terahash. Which also means W/TH and J/TH are the same number wearing two outfits — sheets write J/TH, J/T and W/T interchangeably — and old sheets write W/GH: the 2016 S9’s 0.098 W/GH is 98 J/TH. Same number, smaller rung.

The worked example: one division

Take the cold-open row and do it by hand: 3,645 W ÷ 270 TH/s = 13.5 J/TH. The efficiency row is not new information — it is the other two rows, divided. Cross-check it in reverse: 13.5 J/TH × 270 TH/s × 0.024 = 87.48 kWh per day, which is exactly 3,645 W × 24 hours. (The 0.024 is nothing exotic: ×24 hours, ÷1,000 to turn watt-hours into kWh.) For scale, that ~87.5 kWh a day is roughly three typical US households of electricity — the EIA puts the residential average near 29 kWh/day.

Why efficiency beats raw hashrate

Receipt one: in the Cambridge CCAF industry survey (2025, 49 mining firms), electricity was over 80% of miners’ cash operating costs. The wage arrives in coin; the bill arrives in watts, forever. Hashrate sets how big your slice is — J/TH sets what the slice costs to earn. Buying hashrate without efficiency is renting revenue from your power company.

Receipt two is the one to remember: the 18 W Bitaxe Gamma on a desk — 1.2 TH/s at 15.0 J/TH, community-measured stock figures, dated on its page — is built on the same BM1370 chip family as Bitmain’s S21-class flagships, and lands within two joules of the flagship-air figure: 15.0 against the S21 XP’s 13.5. A one-chip desk object and a 300+-chip appliance, separated by hashrate and almost nothing else. Efficiency lives in the silicon generation; hashrate is just how many chips you rack. (The desk miner’s full story is the next lesson.)

Two footnotes before the fine print. Underclocking improves a machine’s J/TH but cuts its TH/s — it moves the operating point, it doesn’t create margin from nothing; per-model tuning behavior lives with the rig pages. And when the network’s average J/TH marches past yours, cheap power only delays the arithmetic — the mechanism that retires machines years before they physically die is lesson 5’s story.

Reading the fine print: four traps

Trap 1 — wall vs board power. Your meter bills the wall. Between the wall and the chips sit a power supply and fans, and a chip- or board-level figure quietly leaves them out. The honest example is, of all things, the 2016 S9 sheet: it quotes 1,323 W “at the wall” and names the PSU behind the number — the APW3, 93% efficient — so about 1,230 W reaches the boards, and the ~7% gap is printed right on the sheet. Most listings never say which number they’re quoting. If a sheet doesn’t say “on wall”, assume it’s the friendlier number.

Trap 2 — tolerances. Bitmain’s own spec tables rate hashrate at ±3% and power at ±5% — ±5% and ±10% on older generations (manufacturer support pages). The spec is the center of a band, not a floor: a “270 TH/s” machine delivering 262 is inside spec, and so is one drawing 3,827 W.

Trap 3 — the “at 25°C”. Wall power is rated at 25°C ambient (Bitmain support pages — manufacturer figure). Run the machine in a hotter room and the fans ramp while the silicon leaks more, so community-measured draw runs above the rating — how far above depends on your room, which is why we keep the magnitude qualitative. The direction, though, is one-way: heat never makes the bill smaller.

Trap 4 — “up to”, and the voltage row. A reseller’s “up to 278 TH/s” is the +3% edge of the very band the manufacturer centers at 270 — the same machine, quoted on its best day. And the row almost nobody reads before checkout: input 220–277 V. A 3,645 W machine on a standard North American 120 V outlet would draw about 30 A — past any standard 15 or 20 A branch circuit — which is exactly why the sheet demands 220–277 V. Whether your building has such a circuit is module 1’s at-home lesson.

What counts as good right now

Data as of 2026-09 · every figure a manufacturer rating, ±tolerance, at 25°C · per-model rows with live numbers: /rigs

This table rots by design — that’s why it carries a date instead of pretending to be evergreen. The march from 98 to 13.5 J/TH took about a decade — a ~7× gain — and it doesn’t announce pauses. What that march does to a machine you already own is lesson 5; what it does to undated “most efficient miner” articles, you can now judge yourself. No “best miner” verdict here either — per-model rows, live against today’s network, are /rigs’ job.

FAQ

What does J/TH mean?

Joules per terahash — the energy a miner burns to perform one trillion hashes. Compute it yourself: wall watts ÷ hashrate in TH/s. Lower is better — it is the number that decides whether a machine’s revenue survives its power bill.

What does TH/s mean?

Terahashes per second — one trillion hash attempts per second; in this school’s vocabulary, a trillion drawer-pulls a second. The ladder runs H/s → kH → MH → GH → TH → PH → EH, each rung ×1,000. One modern machine does hundreds of TH/s; the whole network first crossed 1 ZH/s — a sextillion hashes a second — in 2025, and this page renders the live number from our own node.

Is W/TH the same thing as J/TH?

Numerically, yes. A watt is a joule per second, so watts per terahash-per-second equals joules per terahash — the seconds cancel. Spec sheets write J/TH, J/T or W/T interchangeably, and old sheets write W/GH: 0.098 W/GH and 98 J/TH are the same number.

What is a good J/TH right now?

As of September 2026 — this band rots by design, which is why we date it: announced hydro frontier ~9.5 J/TH (a manufacturer claim, not yet community-verified at scale), flagship air ~13.5, current mainstream 15–17.5, previous generation 23–34.5, and the 2016 museum piece sits near 98. Every figure is a manufacturer rating, ±tolerance, at 25°C.

Does higher hashrate mean more profit?

Higher hashrate means a bigger share of the work, and so of the wage. Profit is that wage minus electricity — and electricity was over 80% of miners’ cash operating costs in Cambridge’s 2025 CCAF survey of 49 firms. A big inefficient machine can lose to a smaller efficient one on the same power budget; J/TH is the survival stat.