Compare ASIC miners: efficiency, power draw and running cost

Every working machine we have in stock, in one table, ranked by the
number that decides whether it earns anything: joules per terahash.
For each unit you get what it hashes, what it actually pulls at the wall, its J/TH,
and what the electricity alone costs to run it for a day — then the live price
and the condition, straight from the listing. This is the ASIC miner comparison we
wanted when we started buying and reselling this hardware, and nobody had it: not a
manufacturer’s brochure page, and not a spec sheet for machines that are out
of stock everywhere.

The table is generated from our own catalogue when the page loads. The
price you see is the price in the cart, the stock state is today’s,
and when a machine sells the row goes with it. Nothing here is a screenshot of what
was true in March.

How to read this ASIC miner comparison

Two numbers do almost all the work, and they are not hashrate.

• Efficiency (J/TH) is joules of electricity per terahash of work.
  It is the machine’s entire competitive position in one figure: a unit at
  15 J/TH does the same work as one at 30 J/TH on half the power, so it stays
  profitable at twice the electricity price. Lower is better. When difficulty rises,
  the least efficient machines stop paying first — every time, without
  exception.
• Power at the wall (W) decides what you have to build. It sets
  the circuit, the breaker, the cooling and the noise, and it is the number your
  electrician needs.

  Power at the wall also decides how much heat you have to get rid of, and on the hydro and immersion machines in this table that is a system you have to build rather than a fan you switch on. What each cooling type actually requires sets out the coolant loop, the published input voltage and the heat in BTU/hr for every one of them.

  Nearly everything on this page wants 200–277 V
  and will not run on an ordinary 120 V household outlet.

Hashrate on its own tells you almost nothing about whether a machine is worth
buying. A 580 TH/s hydro unit and a 90 TH/s desktop miner can be the same business
decision or opposite ones, depending entirely on what you pay for power.

The power-cost column is arithmetic, not a forecast. It is
watts × 24 ÷ 1000 × $0.10 — the cost of the
electricity to run the machine for 24 hours at ten cents a kilowatt-hour, and
nothing else. It is not profit, it does not include what the machine earns, and if
your power costs something other than $0.10/kWh, scale it: at $0.06 multiply by 0.6,
at $0.15 multiply by 1.5. We chose $0.10 because it is close to the US residential
average and it is a round number to scale from, not because it is your rate.

Bitcoin miners (SHA-256)

Ranked most efficient first. Every one of these mines the same coin, so J/TH is the whole argument: the lower it is, the higher the electricity price at which the machine still makes money.

Browse all Bitcoin miners in stock →

Litecoin and Dogecoin miners (Scrypt)

Scrypt machines merge-mine LTC and DOGE together, so a single unit earns in two coins at once. Efficiency here is quoted in J/MH or J/GH rather than J/TH - the units are not comparable with the Bitcoin table above, only within this one.

Browse all Scrypt miners in stock →

Everything else: Kaspa, CKB, Dash and friends

These are single-algorithm machines for one network each, so they are listed rather than ranked - comparing J/TH between Kaspa and Dash hardware means nothing. What matters here is the coin, and the fact that if that one network stops paying, the silicon has nowhere else to go.

Browse all altcoin miners in stock →

What this table cannot tell you

Everything above is the manufacturer’s figure for the bin we stock, taken
from that product’s own listing and checked against itself — watts
divided by hashrate has to equal the stated efficiency, and any row where it did not
was pulled out and fixed before this page went up rather than published with a
number that could not be true. What the table still cannot do:

• Bins have tolerance. Most of these are ±3% to ±5%
  on hashrate and ±5% to ±10% on power. A 234T is not guaranteed to hash
  234.00.
• Wall power is a rating, not a promise. Real draw moves with
  ambient temperature, voltage and firmware. A hot room costs you both hashrate and
  watts at the same time.
• Hydro and immersion units are not complete systems. The price on
  a hydro row is the machine. The loop, the dry cooler, the pumps and the plumbing are
  yours, and they are not a rounding error. See what we hold in
  hydro cooling.
• Noise is missing on purpose. Most air-cooled machines here sit
  around 70–80 dB, which is a workshop, a garage or a shed — not a spare
  bedroom, whatever a listing photo implies. The honest version is in
  ASIC miner noise explained, and if the
  room matters more than the hashrate, start with
  the best mining hardware for
  home.
• Firmware tunes move all three numbers at once. An underclocked
  S19 draws less and hashes less, and its J/TH usually improves. Where a listing sells
  a machine on a tune, the table shows the manufacturer’s stock figures and says
  so in the row.

Working out your own break-even

We do not publish a profitability number on this page, and we are not going to.
Any figure of that kind is wrong within a week — difficulty moves, the coin
price moves, and a table that quotes daily earnings is a marketing device rather
than information. What does not change is the arithmetic, so here it is.

Revenue for a Bitcoin miner is hashprice: what the network pays
per petahash per second per day, a figure you can look up in ten seconds and which
we explain in what a miner actually
earns. Cost is your electricity. Put the two together and the machine’s
hashrate cancels out completely, which is the whole reason efficiency is the
specification that decides this:

Break-even power price ($/kWh) = hashprice ($/PH/day) ÷
(24 × J/TH)

Quote hashprice in dollars per petahash per day and the thousand that turns
terahashes into petahashes cancels the thousand that turns watt-hours into
kilowatt-hours, which is why the constant is a plain 24. Put today’s
hashprice in and you get the electricity price at which that machine stops making
money. Do it for the two ends of the table and the shape of the
market falls out: at any given hashprice, the best machine on this page tolerates
roughly three and a half times the power price of the oldest one,
because that is the ratio of their J/TH. That single fact explains most of what
looks irrational about the used market. The longer version, with the second
break-even that matters — the one that pays back the capital, not just the
power bill — is in
mining break-even:
the two numbers, and the running-cost side is broken down line by line in the
ASIC miner buying
guide.

When the least efficient machine on the page is the right buy

The used Antminer S19 near the bottom of the Bitcoin table is around 34 J/TH,
against roughly 9.5 for the newest hydro unit. On efficiency it is not close. It is
still the machine we sell most of, and the reason is that you are buying two
different things: efficiency and capital cost per terahash. A new flagship
costs many times what a used S19 costs for the same hashrate, and it earns the
difference back only through the power bill — which means the cheaper machine
wins outright if your power is cheap, if the heat is doing a second job, or if you
are testing whether you want to do this at all before committing five figures.

It loses badly if your power is expensive, and it is the first thing to switch off
when difficulty climbs. That trade, and how to avoid the units that are cheap for a
reason, is
how to buy a used ASIC miner without
getting burned and
the real advantages of buying
used. What ages a machine out of the market is covered in
physical vs economic lifespan.

What is not in this table

The table is working machines with a full published specification, one row per
distinct bin. Deliberately excluded:

• Lots and pallets. The same machines by the pallet would appear
  as a dozen identical rows. Fleet quantities are quoted, not listed — see
  buying by the pallet or call us.
• Anything sold for parts. A hashrate printed on a machine sold
  as needing parts is a description of
  what it did when it worked, not a specification. Those listings say what is wrong
  with them, which is the point of them.
• Hashboards, power supplies, controllers and fans. They have no
  hashrate of their own. Compatibility is the thing that matters there, and it is in
  Antminer parts compatibility.
• Two listings we are still checking. On one, the model name and
  the specification describe two different machines; on the other, the price is far
  enough below every comparable machine on the same shelf that we want to confirm it
  before a comparison table presents it as the deal of the year. Both are still for
  sale, unchanged, exactly as they were written. We would rather leave a row out than
  publish a number we are not certain of.

Before you commit to one

And if the electricity is going to be self-generated, size the generation before you pick the machine rather than after. What size solar array each miner needs runs every row of this table through the same arithmetic and gives you the array in kW and the battery in kWh — which for the big machines is a far larger number than the one in the price column.

Two things are worth doing before you spend the money, and neither takes long.

Ask what applies to the exact unit. Warranty and return terms on
this store are stated per listing rather than per store, because sensible terms on a
sealed new machine and on a tested used one are not the same terms. Ask us about the
specific machine before you pay — it is a two-minute conversation. The
background is in
ASIC miner warranty explained.

Know how you will test it. The bench procedure we run before a
machine ships is published in full at
bench-testing an ASIC
before it ships, and what you should do in your own first 48 hours is in
acceptance-testing a miner you just
received. A machine that is going to disappoint you usually says so in the first
two days, while everyone’s window to do something about it is still open.

If you already know the pair you are choosing between, we have written several of
them up properly:
S21 Pro vs SealMiner A2 Pro
Air,
Avalon Q vs S19k Pro,
S19 XP vs S21 Pro and
L9 16G vs 17G. For the wider ranking by the
power price each machine breaks even at, see
best ASIC miners 2026. If you are
mining something other than Bitcoin, start with
Scrypt, Kaspa and the
rest. And if the choice is specifically a Kaspa machine against a Scrypt one,
Kaspa miner vs Litecoin miner is the
comparison on this site where the two spec sheets cannot be put on the same axis at
all — 150 J/TH and 210 J/GH are not the same unit, and the break-even power price
is the only figure that survives the crossing.

Or just ask a person who has taken these apart:
+1 (781) 577-0377,
ns@ingmining.com, or the
contact form. Everything in stock is at
all miners.