J/TH (Joules per Terahash)

J/TH (joules per terahash) is the standard efficiency metric for Bitcoin ASIC miners: how much energy a machine burns to perform one terahash (10¹² hashes) of work. Because a watt is one joule per second, the arithmetic is direct — wall power in watts divided by hashrate in TH/s gives J/TH. A machine pulling 3,250 W at 95 TH/s runs at about 34 J/TH. The lower the number, the more efficient the miner, and the cheaper every unit of hashrate is to operate.

Why it is the number that matters

Hashrate tells you how big a machine is; J/TH tells you whether it can survive. Two miners with identical hashrate can have very different power bills, and in an industry where electricity is the dominant ongoing cost, efficiency — not size — separates profit from loss. The math is unforgiving: at a given electricity price and network difficulty, there is a J/TH threshold above which a machine mines at a loss, and every difficulty increase pushes that threshold down. This is why fleet operators rank machines by efficiency when deciding what runs and what gets shut off, and why inefficient hardware is the first casualty of curtailment and price spikes. Your electricity rate effectively decides which generation of hardware you can afford to run.

The generational ladder

Each hardware generation roughly steps efficiency down. The Antminer S9 era (BM1387 chips) ran near 98 J/TH; the S17 generation brought that to the 40–45 J/TH range; the S19 and S19 Pro landed around 29.5 J/TH; the S19 XP reached about 21.5 J/TH; the S21 family runs near 17.5 J/TH; and the latest BM1370-based flagships push into the roughly 15 J/TH class. Read that ladder end to end and the story is stark: a modern flagship does the same work as an S9 for about one-sixth the energy. That gap is exactly why old miners get curtailed first when power prices rise — and why they find second lives where the "waste" heat is the product, in space-heater and heat-reuse deployments where effective efficiency is judged after the heating value is counted.

Stock is not destiny

A machine's J/TH is not a fixed property of the nameplate; it is a function of its operating point. Efficiency improves substantially at lower frequencies and voltages, so underclocking a miner routinely cuts J/TH by a double-digit percentage in exchange for lower total hashrate — the standard move for home miners on residential rates. Autotuning firmware compounds this by finding each chip's true operating point at runtime rather than driving all silicon at a conservative factory average. The practical upshot: the J/TH figure that matters is the one your machine achieves at your chosen power level, measured at the wall — including PSU losses and fans — not the datasheet's ideal.

Comparing honestly

Two cautions when comparing numbers. First, spec-sheet J/TH is typically quoted at stock settings and nominal conditions; real wall-plug figures vary with temperature, PSU efficiency, and input voltage. Second, hydro and immersion variants of the same chip generation post better numbers partly because superior cooling lets the silicon run in a more efficient regime. Compare like with like, and when in doubt, trust a kill-a-watt meter over a brochure. Also mind the denominator: quoted hashrate is an average over time, and a machine throwing hardware errors or thermal-throttling delivers fewer accepted shares than its dashboard implies — its true economic J/TH is worse than the instrument panel suggests. Efficiency, like everything in mining, is ultimately measured in sats earned per kilowatt-hour paid. Track current models on our live ASIC efficiency frontier, or see practical tuning in the S19 power-efficiency guide.