SHA-256 algorithm: overview, coins, ASIC miners and specifications

SHA-256 is a cryptographic hash function from the SHA-2 family, published by the NSA in 2001 as a federal standard. Satoshi Nakamoto took it off the shelf and built the Bitcoin proof of work around it: to find a block, a miner runs through header variants and looks for a hash below the current target. No magic, just billions of identical calculations per second.

That sameness is exactly what made SHA-256 the perfect target for dedicated silicon. There are no branches, no memory lookups, no random reads, only integer operations in a fixed pattern. A graphics card loses to an ASIC by a factor of thousands on this code, which is why GPU mining of Bitcoin died out by 2013 and is not coming back.

Today SHA-256 is the largest mining algorithm both by capital invested and by the amount of hardware built for it. Our catalog holds 136 models from nine manufacturers, with efficiency ranging from 9.45 to 157 J/TH, so the gap between the best and the worst machine is more than sixteen times. Understanding that gap is the main thing to take away from this page if you are picking equipment.

The miner assembles a block: it pulls transactions from the mempool, builds a Merkle tree over them, adds the previous block hash, a timestamp, the current difficulty target and a nonce field. That gives an 80 byte header. The header is run through SHA-256 twice, producing 256 bits, or 64 hexadecimal characters.

Then the search begins. A hash cannot be predicted: flipping a single input bit changes the whole result, and there is no inverse function. The only way to get a hash with the required number of leading zeros is to change the nonce and compute again. When the four byte nonce runs out, the miner adjusts the timestamp or the transaction set and starts over.

Hashrate is simply how many such attempts a device makes per second. An Antminer S21 XP at 270 TH/s is 270 trillion attempts every second. The network as a whole aims for one block roughly every ten minutes and recalculates difficulty every 2016 blocks, about once every two weeks, to match the total hashrate.

This leads to the point that matters for a farm owner: your income depends not on your absolute hashrate but on your share of the network. If the network grew 10% and your farm did not, your income in coin dropped by roughly the same 10%, even though the hardware works exactly as before.

1. Transactions from the mempoolCollected into a candidate block
2. Block header, 80 bytesMerkle root, previous hash, time, target, nonce
3. Double SHA-256The header hash is computed
4. Compare with the targetHash below target? If not, change the nonce
5. Block foundBroadcast to the network, the miner takes the reward and fees

• 2001The NSA publishes SHA-2 as FIPS 180-2. Nobody was thinking about mining, the function was made for signatures and integrity checks.
• 2009Bitcoin launches. The first blocks are found on ordinary CPUs, reward 50 BTC.
• 2010GPU mining appears and network hashrate grows by two orders of magnitude.
• 2013First ASICs: Avalon and the Bitmain Antminer S1, around 2000 J/TH. Crude by todays standards, yet enough to push GPUs out within months.
• 2016The Antminer S9 on 16 nm, 98 J/TH. It stayed in service for years and still turns up wherever electricity costs almost nothing.
• 2020MicroBT Whatsminer M30 and Antminer S19 arrive, efficiency drops to 30 J/TH and industrial scale mining takes over.
• 2024The halving cuts the reward to 3.125 BTC. Machines worse than 25 J/TH leave the market or move where a kilowatt is nearly free.
• 2026The current bar is 10 J/TH and below for hydro units, while top air cooled machines sit around 13 J/TH. The next halving is expected in 2028.

Below are the twelve highest performing machines in the catalog. The full list of 136 models opens with the button, including the older series that still earn their keep on cheap power.

Stock firmware does exactly what the manufacturer promised: it holds the rated clocks and does not let you go deeper. For hardware under warranty that is the right choice, for everything else it usually is not.

The AsicBoost custom firmware for Antminer opens up what the chip can already do while the factory logic keeps it locked: overclocking and undervolting beyond the stock limits, autotuning of frequency and voltage per chip, power profiles that switch by temperature, a per chip heat map and fleet wide management through HashCore Toolkit. In the catalog this firmware is available for 26 models.

The practical effect depends on your goal. If you want maximum coin, an S21 XP Hydro goes from about 473 to 615 TH/s. If you want a smaller electricity bill, the same chip at lower voltage produces less hashrate but noticeably better J/TH, and over a long run that is often the better trade.

The developer fee is 2.8% of hashrate. Count it honestly against the gain: with a 25% gain and a 2.8% fee you are ahead, with a 3% gain there is no point.

Specific profitability numbers go stale within a week, so there is no sense publishing them on a reference page. It is more useful to understand what they are made of.

Five values drive the result: the coin price, network difficulty, your cost per kilowatt hour, the pool fee and your machine efficiency in J/TH. The first two are out of your hands entirely, the third depends on where the farm sits, and the last two are what you actually control.

A quick sanity check: multiply the power draw in kilowatts by 24 and by your electricity price to get the daily cost. If the daily output in money does not cover it, the machine runs at a loss no matter what it cost to buy.

Open the profitability calculator →

Efficiency is measured in joules per terahash, J/TH, and it is the only figure that compares machines of different generations honestly. Hashrate tells you how much a unit computes, J/TH tells you what that will cost you.

Reference points from the catalog for 2026. Below 12 J/TH sit modern hydro units and the best air cooled models, 12 to 20 J/TH covers current machines that pay off at ordinary industrial tariffs, 20 to 35 J/TH is the previous generation that survives on cheap power, and above 60 J/TH is history such as the S9 at 98 J/TH or the Avalon 7 at 157 J/TH.

The math is simple: a 3000 W machine at 6 cents per kilowatt burns about 4.3 dollars a day. If it gives 200 TH/s that is 15 J/TH. A machine at 30 J/TH spends the same on electricity for half the hashrate. That is exactly why the fleet turns over after every halving: old hardware stops covering the bill.

Electricity comes first. Find out your price per kilowatt and your available capacity before you pick a model. Above 10 cents it only makes sense to look at machines better than 15 J/TH.

Cooling type. Air is simpler and cheaper to start with, hydro gives better J/TH and runs quieter but needs a loop, a heat exchanger and maintenance. Immersion is a separate project with a tank and dielectric fluid.

Power per unit. A top machine drawing 7 to 20 kW needs its own line and a properly sized panel. Count the cable and breaker headroom, not just the miner nameplate.

Noise. An air cooled ASIC is 75 decibels and up, in the same range as a chainsaw. It does not belong in a flat or a residential house.

Warranty and service. Check the warranty period, who repairs units in your country and whether spare parts exist. For a fleet of ten machines that matters more than an extra 5 TH/s on paper.

Payback. Run the numbers conservatively: current difficulty plus room for growth, and a price no higher than today. If a model only pays off when the price doubles, that is a bet, not a calculation.

Buying by hashrate instead of efficiency. Two machines at 200 TH/s can differ twofold in the electricity bill. Look at J/TH, not only at TH/s.

Counting income without electricity. Calculator headline numbers are usually shown before power cost. Put in your real tariff, including losses and cooling.

Ignoring firmware. Factory logic leaves both hashrate and savings on the table. Updating the fleet costs less than buying another machine.

Saving on ventilation. Overheating kills hashboards faster than any overclock. Airflow is calculated for the heat of the whole room, not for one miner.

Buying old hardware ahead of a halving. Models at 40 J/TH and worse almost certainly go negative at any industrial tariff after the next halving.

Running stock firmware for years. Updates close security holes and improve autotuning. A fleet that is never updated slowly loses both stability and money.