Etchash algorithm: the Ethereum Classic network, ASIC miners and specs

Etchash is the proof of work of the Ethereum Classic network. Calling it a separate algorithm is more a matter of tradition than of substance: inside sits the same Ethash that Ethereum ran on before it moved to staking, with one constant changed. In November 2020 the Thanos hard fork at block 11,700,000 stretched the epoch from 30,000 blocks to 60,000, and that is the only difference. Everything else, from the structure of the cache to the order of memory accesses, matches byte for byte.

The point of that edit was practical. Ethash is built so that the miner keeps a large file in memory, the dataset, also known as the DAG, and that file grows on its own with every epoch. By 2020 it had passed 4 gigabytes, and graphics cards with four gigabytes on board began dropping out of mining. Doubling the epoch length instantly halved the DAG size, from roughly 3.91 gigabytes to 2.47, and gave those cards several more years of life. It also slowed the further growth of the file by half.

For an ASIC buyer this leads to the main property of the algorithm, one that SHA-256, Scrypt and kHeavyHash do not have. A machine built for Etchash has a shelf life, and it is set not by wear but by the amount of memory soldered on. Once the DAG outgrows that memory, the machine simply stops computing. That has already happened with the Antminer E3 on 8 April 2020: four gigabytes were not enough, and the hardware lost the ability to mine Ethereum. That is why memory size is discussed on this page on equal terms with hashrate and watts.

The miner task is the usual one: pick a nonce so that the block header hash lands below the current target. What is unusual is that computing that hash without a large file in memory is impossible. Ethash was deliberately designed to run into memory bandwidth rather than into arithmetic speed. The idea was that memory is expensive and universal, so a dedicated chip would not pull away from a graphics card the way it does on SHA-256. The idea held for three years and ended in 2018 with the arrival of the first ASICs.

Everything starts with the seed. Once per epoch, that is once every 60,000 blocks, it is recomputed by running through the headers of the preceding blocks. A cache is built from the seed: a pseudorandom array that is filled sequentially and then shuffled against itself several times. The starting cache size is 16 mebibytes. The cache is small enough to be kept by light clients, which do not need to find blocks, only to verify those of others.

The dataset, that same DAG, is expanded out of the cache. Each of its elements is assembled from 256 cache nodes by FNV and Keccak-512 operations. The dataset is large, a gibibyte at the start, today about 4360 megabytes on Ethereum Classic, and the miner has to hold all of it. Hence both the memory requirement for the machine and the gradual dying off of older models: the file grows, the memory does not.

Then comes the search itself. A starting value is taken from the header and the nonce and expanded into a 128 byte mix, after which 64 accesses to random places in the DAG take place, and on each one the mix is blended with the data read. The finished mix is compressed, hashed with Keccak-256 and compared with the target. Below the target and the block is found. Above it, a new nonce and all over again. It is precisely these 64 random reads that make the algorithm dependent on memory: there is not much arithmetic here, but the machine loads the memory subsystem in full.

1. Epoch seedRecomputed once every 60,000 blocks by running through the headers
2. CacheA pseudorandom array from the seed, 16 mebibytes at the start
3. The DAG datasetExpanded out of the cache, every element from 256 nodes
4. Mixing64 random reads from the DAG, a 128 byte mix
5. Comparison with the targetFolding, Keccak-256, comparison. Below the target and the block is found

• 2015Ethereum starts on Ethash. The algorithm is conceived as ASIC resistant: the bet is on expensive memory instead of cheap arithmetic.
• 2016On 20 July Ethereum splits at block 1,920,000 after The DAO hack. Ethereum Classic is the original chain, the one that refused to roll back the transactions of the attacker.
• 2018The Antminer E3 shows that the bet on memory did not work. Dedicated machines arrive on Ethash, and the E3 carries 4 gigabytes on board.
• 2020On 8 April the Antminer E3 stops mining Ethereum: the DAG outgrew its 4 gigabytes of soldered memory. A vivid precedent for how a shortage of memory ends.
• 2020In August come three 51 percent attacks in one month, and before them two more in January 2019. The OKEx exchange put the losses from double spends at 5.6 million dollars and suspended ETC deposits and withdrawals.
• 2020In November the Thanos hard fork at block 11,700,000 doubles the epoch length. The DAG shrinks from roughly 3.91 gigabytes to 2.47. This is where the name Etchash comes from.
• 2022On 15 September Ethereum moves to staking. The Ethereum Classic hashrate reaches 307.1 TH/s that day, and ten days later falls to 160.8. The wave of migrants receded almost at once.
• 2026On 22 July the sixth issuance era begins at block 25,000,001. The block reward drops from 2.048 to 1.6384 ETC. Most calculators had still not reflected this at the time this page was built.

All 4 Etchash models in the catalog, sorted by efficiency with the best on top. Hashrate in MH/s, efficiency in joules per megahash. Click a name to open the profitability calculator for that model.

The short answer: there is no custom firmware for Etchash on any model in the catalog. AsicBoost firmware is built for SHA-256 boards, on Scrypt two models are supported separately, and there are no Ethash machines on that list and none were ever planned. Network installation through HashCore Toolkit belongs to those same supported lines.

The same holds for third party developers in general. We went through the official supported model lists of the well known alternative firmware projects: not one of them declares the Antminer E9, the E9 Pro or the Jasminer line. The largest projects state openly that they work exclusively with machines on SHA-256. For Jasminer the manufacturer distributes only its own builds through its site, and the updates there concern hashrate display and network settings rather than overclocking.

The reason is economic, not technical. The market for Etchash machines is too small to pay back the development and support of custom firmware, and the network itself has not grown for four years. The practical conclusion for an owner: the efficiency figures stated by the manufacturer are your figures for the whole time you own the machine. There will be no overclocking, no undervolting and no reduction of the developer fee on these machines.

Hence the warning as well. Any offer of "firmware for the E9 or Jasminer, plus thirty percent hashrate" is either a scam or malicious code. There is no need to put anything on such machines other than the official manufacturer firmware downloaded from its site.

Let us start with the unpleasant part, because that is the honest way. At the electricity price normal for Spain, around 0.15 euro per kilowatt hour, not one of the four catalog models comes out in the black, and the gap is not marginal, it is a multiple. The best machine, the Jasminer X16-Q, brings in about 20 euro a month against an electricity bill of about 67 euro. The worst, the Antminer E9 Pro, brings in about 37 euro against a bill of about 238. The payback period in a situation like that is not long, there simply is none: every month of operation increases the loss.

The break even threshold on electricity works out like this: the Jasminer X16-Q at about 0.044 euro per kilowatt hour, the X16-P at about 0.043, the Antminer E9 Pro at about 0.023, the Antminer E9 at about 0.018. For comparison, the regulated tariff in Spain in the summer of 2026 holds at around 0.16 euro. The grid price sits three to almost ten times above those thresholds, depending on the model. For the machines to break even at the grid price, the ETC rate would have to rise roughly threefold for Jasminer and six to eight times for Antminer, and that is without counting the fact that network hashrate would follow the rate up.

A separate word about calculators. On 22 July 2026 another issuance cut went through, and the block reward fell from 2.048 to 1.6384 ETC. At the time this page was built some of the large aggregators were still counting at 2.048, and a few even at 2.56, that is at a reward cancelled back in 2024. The error overstates income by a quarter or by half. Check the reward in the network explorer before believing any profitability figure, including ours.

Open the profitability calculator →

Efficiency here is measured in joules per megahash, J/MH. The same unit is used on the Scrypt page, but the figures cannot be compared across algorithms: an Etchash megahash and a Scrypt megahash are different work. Inside Etchash the comparison is valid, because all four machines do exactly the same thing and mine exactly the same coin.

The spread within the catalog is almost threefold. The Jasminer X16-Q needs 0.32 J/MH, the X16-P 0.33, the Antminer E9 Pro 0.6, the Antminer E9 0.8. In money it looks like this: at the Spanish tariff the X16-Q costs about 67 euro a month and the E9 Pro about 238, while the E9 Pro gives only twice as much hashrate as the X16-Q. On an algorithm where income is already negative, a difference like that decides whether the loss will be tolerable or heavy.

The arithmetic goes as usual: multiply efficiency by hashrate in megahashes to get draw in watts, divide by 1000 to get kilowatts, multiply by your tariff and by 720 hours a month. Compare the sum you get with the mining forecast in the calculator. While you are at it, check which block reward the calculator uses: after July 2026 the correct figure is 1.6384 ETC.

Start with the calculator and your real tariff. This is the case where the arithmetic is more honest than any description. At an ordinary electricity price the answer will be negative on all four models, and that is a normal result of the calculation, not an input error.

Ask for the memory size in writing. Within one line there are different variants: some Jasminer machines have five gigabytes, the X16-Q and X16-P have eight. The difference in remaining life reaches ten years. Memory size has to be in the specification before payment, not in a verbal promise from the seller.

Work out the price per megahash, not the price per machine. The efficiency spread in the catalog is almost threefold, while prices on the secondary market live a life of their own. Compare the money per unit of hashrate, and compare J/MH on its own.

Do not count on a fallback coin. EthereumPoW is already out of reach for the Bitmain machines because of its DAG size, and Jasminer has about a year of headroom. The other networks on this algorithm are either switched off or so small that their whole daily issuance runs to hundreds of dollars.

Build the next issuance cut into your numbers. The reward falls by a fifth every five million blocks, which is roughly once every two years and a bit. If you are counting payback over three years, the cut will certainly land inside that period.

Look at the generation of the hardware. The catalog models were released in 2022 and 2023. Machines several times more efficient have appeared since then, sometimes at a comparable price. If a seller offers the old generation at the price of the new one, that is a reason to haggle or to walk away.

Working out profitability on a calculator with the old reward. After 22 July 2026 the correct reward is 1.6384 ETC. At the time this page was built some of the large aggregators were counting at 2.048 and a few at 2.56, that is at a figure cancelled back in 2024. The error overstates income by a quarter or by half.

Thinking that an Ethash machine will mine any Ethash coin. Compatibility is decided by the DAG size of a specific network, not by the name of the algorithm. The Antminer E9 with six gigabytes and the E9 Pro with seven cannot mine EthereumPoW: its file has already passed 7.4 gigabytes. That is a physical limit, not a question of settings.

Not checking the memory size when buying secondhand. One line, different variants, a difference in remaining life of up to ten years. The precedent is well known: the Antminer E3 with four gigabytes stopped mining Ethereum in April 2020 simply because the file grew.

Believing that a falling hashrate will make mining more profitable. The Ethereum Classic hashrate has held for a fourth year at roughly the level of September 2022, while the block reward over the same time has fallen by more than a third. Income per unit of power has been going down, competition has not.

Looking for overclocking firmware for the E9 or Jasminer. It does not exist at any known project. Any offer like that is a risk of losing the machine or of getting malicious code instead of a gain in hashrate.

Expecting income to grow from a network upgrade. Discussions of the upcoming upgrade often bring up the arrival of a base fee and a protocol treasury. The specification says plainly that the block reward and the tips stay with the miner unchanged, and the base fee never belonged to the miner before either. For profitability the upgrade is neutral.