Blake2s algorithm: the Kadena network, ASIC miners and specs

Blake2s is a cryptographic hash function from the BLAKE family, presented in December 2012. The authors, Jean-Philippe Aumasson, Samuel Neves, Zooko Wilcox-O’Hearn and Christian Winnerlein, took BLAKE, a finalist of the NIST competition for SHA-3, as their base and released two versions: Blake2b for 64 bit processors and Blake2s for everything else. The specification is set out in RFC 7693 of November 2015.

Of that pair the Kadena network chose the smaller one for its proof of work. Blake2s works in 32 bit words and 64 byte blocks and does ten rounds instead of twelve. For a processor that is a compromise, while for silicon it is closer to a gift: the adder comes out more compact, the pipeline is shorter, and the die area per hash is smaller. No resistance to dedicated chips was ever intended here, and the first ASIC for Blake2s appeared as early as November 2020.

Now comes the unpleasant part, the reason this page is worth reading to the end. The algorithm is alive and computes just fine, but the network all this hardware was built for is in a dire position. On 21 October 2025 the Kadena organization announced that it was ceasing all activity and active support of the blockchain. The network keeps producing blocks only because the community forked the node and maintains it with its own resources. We go through this in detail below, because it bears on a decision to buy hardware far more than any efficiency figure.

The miner task here is as simple as it gets, simpler than on most algorithms in this series. The block header is taken, the nonce field inside it is changed, the whole thing is run once through Blake2s-256, and the result is compared with the target. No large files in memory as on Etchash, no eleven functions in a row as on X11, no second pass as on Blake3. One call to the hash function per attempt.

The Kadena block header occupies 318 bytes, but not all of it is hashed: the last 32 bytes are the field of the hash itself, and they do not go into the work. That leaves 286 bytes. Inside sit the time, the link to the parent block, the target, the chain number, the height, the payload and, most interesting of all, 110 bytes of links to the neighbouring chains. The nonce takes 8 bytes and stands at the very end of the hashed data, at offsets 278 to 285.

This is worth dwelling on, because two algorithms are easy to mix up here. In Alephium, which has a page of its own on our site, the nonce takes 24 bytes and is placed before the header, and the hash is computed twice. In Kadena the nonce takes 8 bytes, stands after the header, and there is a single pass. These are directly opposite constructions, and hardware does not move between them, despite the similar names of the functions.

The finished 32 bytes are compared with the target as a single unsigned number written least significant byte first. Below the target and the block is found. Above it, the next nonce. One more detail that matters in practice: the miner does not choose the chain number, the node sets it and hands the miner a ready work package. There is simply no chain field in the miner settings.

1. Block header318 bytes, of which the first 286 are hashed
2. Links to neighbours110 bytes, three neighbouring chains along the Petersen graph
3. An 8 byte nonceStands at the end of the hashed data, not in front of it
4. A single Blake2s passTen rounds, 64 byte blocks, a 32 byte output
5. Comparison with the targetAs an unsigned number, least significant byte first

• 2012In December Blake2b and Blake2s are presented. The base is taken from BLAKE, a finalist of the NIST competition for SHA-3.
• 2016Kadena is founded. Will Martino and Stuart Popejoy came from the blockchain division of J.P. Morgan.
• 2019In November the mainnet opens for public mining, launching on ten parallel chains.
• 2020On 20 August comes the expansion to twenty chains. A throughput of 480,000 transactions per second is claimed.
• 2021In November the all time high of the KDA rate is set, about 27.6 dollars. A decline lasting years begins after that.
• 2024On 28 January the network reaches its peak hashrate, 666.88 PH/s. This is the last good news in this list.
• 2025On 21 October the Kadena company announces that it is ceasing all activity and active support of the blockchain. The rate falls by roughly half within a day.
• 2025On 8 November the community forks the node and continues the network with its own resources. Large pools suspend payouts, exchanges delist KDA.

All 4 Blake2s models in the catalog, sorted by efficiency with the best on top. Hashrate in TH/s, efficiency in joules per terahash. Click a name to open the profitability calculator for that model.

There is no public custom firmware for this algorithm. AsicBoost firmware is built for SHA-256 boards, on Scrypt two models are supported separately, and there are no Kadena machines on that list. Network installation through HashCore Toolkit belongs to those same supported lines.

Third party projects do not change the picture. We went through the official supported model lists: the largest projects state openly that they work only with Bitmain machines on SHA-256, and the Antminer KA3 is not among them. Goldshell publishes firmware for the KD line itself, but puts out only compiled images with no source code, and the updates there are about stability rather than overclocking.

Goldshell has a story of its own. The stock web interface holds a hidden overclocking menu: press the caption of the miner settings section several times in a row and fields for frequency, voltage and fan speed open up. KD5 owners raised the frequency and got growth from roughly 17.8 to 18.6 TH/s, but the rate of hardware errors and rejected shares rose along with it, to two or three percent, so payouts did not increase. In firmware from version 2.1.6 onwards access to those fields was restricted.

The practical conclusion is simple. The factory efficiency figures are your figures for the whole time you own the machine. And a warning while we are at it: the KD MAX is a separate model with different chips, not a firmware for the KD6, although listings regularly confuse the two. Trying to load an image from one model into another ends with a machine that does not work.

Here we have to start not with the arithmetic but with the state of the network, because the arithmetic without it is meaningless. On 21 October 2025 Kadena announced that it was ceasing all activity and active support of the blockchain. The network produces blocks thanks to a fork that the community launched on 8 November 2025 and maintains with its own resources. Large pools suspended payouts at the same time for the duration of the transition, and F2Pool wrote plainly in its announcement that it does not guarantee the stability of this version, since it is not run by the original team. Binance, its American arm, OKX and Bybit delisted KDA in the autumn of 2025.

The figures follow from that. The KDA rate is currently below one cent against an all time high of about 27.6 dollars in November 2021, and the sources disagree: exchange aggregators give about half a cent, mining calculators roughly four tenths. Daily turnover of the token across every venue in the world is measured in a few thousand dollars, so even what you mine has nowhere to go without a noticeable loss on slippage. Network hashrate has fallen from 666.88 PH/s in January 2024 to roughly 15, that is minus ninety eight percent. Kadena has no halvings, the reward simply melts away smoothly on an exponential curve and currently stands at about 0.914 KDA per block.

At the Spanish tariff of about 0.15 euro per kilowatt hour the monthly loss works out like this: the Antminer KA3 from 250 to 280 euro, the Goldshell KD6 about 270, the KD5 about 235, the KD-BOX about 21. The spread in the first line comes precisely from the uncertainty in the rate, but the sign does not depend on it. Look at the first two machines: the KA3 is almost five times more efficient than the KD6 and loses roughly the same amount. That happens when income is so small next to the electricity bill that the advantage in efficiency simply drowns. The KA3 earns back from seventeen to twenty six percent of its bill, the Goldshell machines about four.

Open the profitability calculator →

Efficiency here is measured in joules per terahash, J/TH. The notation matches SHA-256, kHeavyHash, Eaglesong and Blake3, but that is a coincidence of notation, not of units. A Blake2s terahash over 286 bytes and a SHA-256 terahash over 80 bytes are different work, so putting 19 J/TH on Kadena next to the single digit figures of Bitcoin machines is meaningless. Inside this page the comparison is valid: all four machines do exactly the same thing.

The spread within the catalog is almost sevenfold. The Antminer KA3 needs 19 J/TH, the Goldshell KD6 needs 90.07, the KD5 needs 125, the KD-BOX needs 128.12. Between the leader and the bottom row it is almost sevenfold. In money at the Spanish tariff the bill works out like this: the KA3 about 341 euro a month, the KD6 about 284, the KD5 about 243, the KD-BOX about 22.

The arithmetic goes as usual: multiply efficiency by hashrate in terahashes 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. The break even threshold on electricity lies between 0.026 and 0.038 euro per kilowatt hour for the KA3, and for the Goldshell machines it is below one euro cent. That second figure means the electricity would have to be not merely free but come with a payment on top.

Settle the question about the network first, the hardware second. Efficiency, price and noise are secondary here. The first question to answer honestly to yourself: are you ready to put money into a network whose developer company has ceased operations and whose support is carried by the community. If the answer is no, there is no need to read further.

Do not count on a fallback coin. Blake2s on ASICs means Kadena and nothing else. Verge on the same function is computed on graphics cards and has a header of a different length, Handshake is Blake2b and SHA-3, Alephium is Blake3. Not one of those options fits physically.

Compare against the market, not against a shop price list. Some sites still hold listings from 2022 with four figure and even five figure prices. Real deals on the secondhand market go for several times and even tens of times less. Before you pay, look at live listings.

Count the electricity, not the purchase price. A Goldshell KD5 or KD6 costing a hundred and fifty dollars burns between 240 and 285 euro of electricity a month: the purchase price is matched by the bill in about two weeks, and after that it is pure loss. The small KD-BOX has a bill of about 22 euro, and it is the only machine in the catalog where the loss is measured in tens rather than hundreds.

Check where you are going to sell what you mine. Several large exchanges have delisted KDA, and daily turnover across all the remaining venues runs to thousands of dollars. That is a separate risk which the other algorithms in this series do not have.

Do not mix up the Goldshell models. The KD-BOX, KD-BOX II, KD2, KD5, KD5 Pro, KD6, KD6 SE, KD LITE and KD MAX are different machines with different chips and different firmware. Ask for the exact model name and a photograph of the label before paying.

Looking at 19 J/TH and deciding that the KA3 pays off. Efficiency only makes sense together with the rate of the coin. At a rate below one cent an excellent J/TH figure does not save anything: the loss runs from 250 to 280 euro a month at the Spanish tariff.

Believing that the hardware can be moved to another coin. It cannot. Blake2b has a different machine word, a different block size and a different number of rounds, these are different chips. Verge has a different header and graphics card mining. Alephium has Blake3 with a double pass and the nonce in front.

Mixing up Blake2s, Blake2b and Blake3. Three different algorithms, three different fleets of hardware, zero compatibility. Buyers regularly pick up a Kadena machine after hearing the word Blake and count on moving to Alephium or Handshake later.

Not knowing that the developer company is closed. On 21 October 2025 Kadena announced that it was ceasing all activity and support of the blockchain. The network is held up by a community fork. Many reference sites and product listings still do not reflect this.

Waiting for a halving and a rise in the reward by the cycle. Kadena has no halvings. The reward melts away smoothly on an exponential curve, and no event after which things get better is built into the issuance schedule.

Buying the machine as a heater. The Antminer KA3 is 3154 watts and noticeable noise. The electricity bill works out at about 341 euro a month at the Spanish tariff, and even after subtracting what you mine the heating comes to more than two hundred and fifty euro, which is more expensive than any heat pump. The noise on top of that makes installation in a living room impossible.