Efficiency is how much energy you burn for each unit of hashrate. On Rent A Rig product cards you will see J/TH — joules per terahash — and that single number only becomes useful when you read it next to wattage, noise, and the site or bay that has to feed the machine. This lab guide walks the formula, the power envelope, the edge-versus- density trade, how to scan a RAR product card without falling for a cheap sticker, and a worked Unit A vs Unit B comparison you can re-run with your real $/kWh.
J/TH is approximately power in watts divided by hashrate in TH/s. A 3,200 W machine at 140 TH/s is about 22.9 J/TH — the same labeling language HashCore cards use on RAR. Two machines with identical terahash are not equal if one draws more watts; over months the inefficient unit costs more in power whether you self-host or pay Rent A Rig hosting $/kWh. Nameplate efficiency assumes stock firmware and target voltage. Aggressive tunes can move both hashrate and J/TH, which is why telemetry power will often differ from the card — use nameplate for planning envelopes, use live draw for operations.
Same TH/s at 3,600 W would print roughly 25.7 J/TH — more heat and more kilowatt-hours for the same hash headline.
Marketing leads with hashrate because it sounds like revenue. Bills lead with watts. Read both before you compare price or rent/day.
Look at hashrate, power, efficiency, and noise together — then look at price and rent/day. A “cheap” machine with poor J/TH can lose on hosting bills. An “expensive” efficient unit can win when electricity is the scarce resource.
Hashrate is revenue potential on paper — the ceiling of how much useful work the SKU can produce when configured correctly. It is not a payout promise and it is not permission to ignore watts. Use it to size opportunity, then immediately ask what power and noise that opportunity demands.
Power is what you must feed the machine continuously. It sets the bay or breaker conversation before aesthetics. If the nameplate already sits on the ceiling of a 3.5 kW bay, you do not have a clever efficiency story — you have a fit problem.
J/TH is the quality of the conversion from electricity into hashrate. Lower means less energy burned per terahash, which compounds every hour the unit is online. Treat it as the quality score that mediates between the hash headline and the power bill.
Noise decides where you are allowed to place the machine. A perfect J/TH number is worthless in a residential room that cannot legally or socially accept a hall-density fan profile. Edge units trade peak aggression for the right to exist near people.
Your site or bay has a maximum continuous kilowatt rating. Add headroom for PSU inefficiency and startup spikes. If a bay is rated 3.5 kW, a 3.55 kW nameplate unit is already a bad fit — the spreadsheet can look fine until the breaker teaches you physics. On rental pages you do not provision your own breaker because RAR does, but efficiency still decides how much useful hash you extract per dollar of included power inside that hall.
Keep roughly 10–20% thermal and electrical headroom when self-hosting. List the breaker or bay limit, list the SKU watts from the product page, then subtract — if the remainder is a rounding error, pick another SKU or another tier.
Quieter edge units often run less aggressive fans or lower power profiles. That can mean less hashrate or a different efficiency print. You are trading peak TH/s for the ability to run near people or inside noise-constrained sites. If your constraint is residential sound, prefer an R85-class profile over a max-density M160. If your constraint is pure facility hash, take density and let the hall get loud — that is what the building is for.
Neither end of the continuum is morally superior. The wrong end for your constraint is just an expensive way to learn you cannot run the machine where you planned to put it. Write the noise limit before you fall in love with a J/TH number.
Suppose Unit A is 140 TH at 22.9 J/TH and Unit B is 110 TH at 23.6 J/TH but much quieter. In a RAR rack, A usually wins on hash per bay. In a home office, B might be the only legal option. Always re-run with your real $/kWh — hosting rates make A vs B arithmetic, not a vibe.
Unit A usually wins when the constraint is hash per powered slot inside a RAR rack or pod. Better efficiency and higher terahash stack: you extract more work from the same neighborhood of kilowatts, and the quieter alternative’s only advantage — social acceptability — does not apply in a hall designed to be loud. Buy or host A when bay power fits and nobody lives next to the fans.
Unit B can win when the constraint is placement, not peak hash. Slightly worse J/TH and lower terahash still beat a perfect facility SKU you cannot turn on at home. Treat B as the machine that keeps the thesis alive under noise limits — then revisit A when you move the fleet into hosting and the continuum slides toward density.
At very low electricity cost, purchase price and availability can dominate. As $/kWh rises — including RAR hosting rate cards — J/TH becomes the quiet decider that separates two SKUs that looked similar at checkout. Re-run any hardware shortlist at your real rate before you wire capital; the ranking can flip without either machine changing.
01 · PAIR THE SKUS — Write J/TH and watts for two candidate machines from the same decision class so you are not comparing a quiet edge toy to an unrelated hall monster by accident.
02 · WRITE THE CEILING — Record the site or bay kilowatt limit with the headroom you actually intend to keep, not the optimistic number that only works on a cold morning.
03 · WRITE THE NOISE RULE — Facility hall versus edge placement changes the legal set of SKUs before efficiency gets a vote; put the constraint on paper first.
04 · PRICE THE KILOWATT-HOURS — Estimate monthly kWh × your real rate (home or hosting) for each SKU so power cost is visible beside purchase price or rent/day.
05 · THEN COMPARE CAPITAL — Only after the four fields and the power math agree should sticker price or rent/day break a tie — never before.
For power cost, yes — less energy per terahash is the point of the metric. Availability, purchase price, noise, and reliability still matter. The best J/TH machine you cannot rack, cannot afford, or cannot legally run is worthless until those other constraints clear.
Live load varies with tune, temperature, and hash target. Nameplate assumes a stock profile; the hall and your firmware choices can push draw up or down. Use nameplate to plan envelopes and checkout fit. Use telemetry to operate day to day and to notice when something drifted.
You do not pay a separate power bill on rentals the way you do on hosting or self-host, but better efficiency often correlates with healthier thermal margins and more useful hash inside the included power story. It also trains your eye for the buy-or- host decision later, when $/kWh becomes explicit again.
Open hardware and rental product pages and read the four fields as a set. If you are hosting, pair this guide with Hosting Basics so bay ceilings and rate cards sit next to J/TH. If you need a non-catalog build, carry the same discipline into a custom rig brief — efficiency still has to fit a real envelope.