AsicBoost Explained — 13% Free Efficiency

AsicBoost is a SHA-256 optimization that lets a mining chip skip part of the hashing work, cutting power consumption by up to about 13% for the same hashrate. It exploits a quirk in how SHA-256 processes the Bitcoin block header, and in 2017 it nearly fractured Bitcoin’s mining community — because the first version of it, used covertly by one dominant manufacturer, quietly distorted the network’s incentives. By 2026, every modern ASIC ships with the “good” version enabled by default, and the “bad” version is a historical curiosity.

Key takeaways

• AsicBoost saves power, it doesn’t add hashes: the same hashrate runs at up to ~13% less energy by reusing SHA-256 “midstate” computations.
• Covert vs overt is the whole story: covert AsicBoost (Merkle grinding) incentivized empty blocks and fought SegWit; overt AsicBoost (version-rolling) does the same math harmlessly.
• The 2017 fight reshaped Bitcoin: it added urgency to SegWit, produced a defensive patent license, and helped motivate Stratum V2.
• It’s universal now: standardized as BIP320 version-rolling, every modern chip from S9 onward and every Bitaxe supports it.
• Your pool has to speak version-rolling for the gain to apply — point an AsicBoost-capable miner at a pool that doesn’t, and you burn ~10-13% more power for nothing.

The SHA-256 trick at the heart of AsicBoost

To understand AsicBoost you need one fact about SHA-256: it processes data in 64-byte chunks. A Bitcoin block header is 80 bytes, so SHA-256 splits it into two chunks — chunk 1 (bytes 0-63) and chunk 2 (bytes 64-79, padded to 64). Computing chunk 1 is the expensive part; chunk 2 is cheap. The reason is that SHA-256 carries a running “state” between chunks: once chunk 1 is processed, that state — the midstate — is fixed, and any header sharing the same chunk 1 can reuse it and skip straight to chunk 2.

The header is laid out so that chunk 1 holds the version (4 bytes), the previous block hash (32 bytes), and the first 28 bytes of the Merkle root, while chunk 2 holds the last 4 bytes of the Merkle root, the time, the bits, the nonce, and padding. A normal miner iterates the nonce (in chunk 2), reusing the one chunk 1 midstate for free. AsicBoost asks a sharper question: what if you could generate several different chunk 1 inputs, each producing its own midstate, and reuse each across many nonce iterations? With four midstates you do roughly 4× the chunk 2 work for the same chunk 1 effort — and the expensive computation runs far less often.

The result: roughly 13-20% less power per hash depending on how aggressively it’s implemented in silicon. Same hashrate, less heat, lower power bill.

Covert AsicBoost — the controversial version

The first publicly known implementation was covert AsicBoost, or “Merkle grinding.” It generates new chunk 1 midstates by changing the Merkle-root portion of chunk 1 — shuffling transactions in and out of the block, recomputing the Merkle tree, and hunting for roots that share the right bytes. It was controversial on four counts:

1. It incentivized smaller blocks. Finding viable Merkle collisions is faster with a smaller tree, so a covert-AsicBoost miner had economic motivation to mine smaller or empty blocks even with transactions waiting — directly at odds with users paying fees to be included.
2. It interfered with SegWit. SegWit’s Merkle structure made covert AsicBoost harder to deploy, pitting the optimization against a protocol upgrade.
3. It was undetectable. A covert-AsicBoost block looks identical to a normal one, so the network couldn’t tell who was using it. Suspicion ran high; proof was scarce.
4. It was patented. The technique was patented by Timo Hanke (former CoinTerra CTO) and Sergio Lerner, meaning a single entity could hold a legal monopoly on a 13-20% efficiency edge — the worst case for a decentralized network.

In April 2017, Bitcoin Core developer Greg Maxwell published a now-famous email arguing — with strong technical evidence — that Bitmain was using covert AsicBoost in production and leveraging that advantage to oppose SegWit. Bitmain denied using it on mainnet while holding patent applications describing exactly that capability; the denials convinced few. The community response was swift: Maxwell, Adam Back and others opposed it publicly, and SegWit activated in August 2017 partly through the user-activated soft fork (UASF) movement. Covert AsicBoost was effectively neutralized.

Overt AsicBoost — the protocol-friendly version

While covert AsicBoost was tearing up the community, a clean alternative emerged: overt AsicBoost, or “version-rolling.” It achieves the same SHA-256 gain by varying the version field (the first 4 bytes of chunk 1) instead of the Merkle root, using bits the consensus rules don’t otherwise need. The advantages over covert are decisive:

• No incentive to skip transactions. Version bits are independent of block content, so the fee market stays intact.
• Compatible with SegWit. Version-rolling doesn’t touch the Merkle structure.
• Transparent. Anyone watching the chain can see overt-AsicBoost blocks by their version-bit patterns.
• Actually more efficient. No Merkle-tree shuffling means less overhead than covert grinding.

The political breakthrough came in March 2018, when the patent holder, Little Dragon Technology LLC, opened the patent under the Blockchain Defensive Patent License (BDPL): any manufacturer could use version-rolling AsicBoost, provided they joined the BDPL and pledged not to wield patents aggressively. Halong Mining’s DragonMint was the first hardware to ship it, Slush Pool the first major pool to support the version-rolling extension, and Bitmain released overt-AsicBoost firmware for the Antminer S9 — which had quietly supported it in hardware all along — later that year. By 2019 it was the de facto standard; by 2026 you can’t buy a SHA-256 ASIC without it.

BIP320 — the technical specification

Overt AsicBoost is standardized in BIP320. It designates 16 bits of the 32-bit version field (bits 13-28 inclusive, mask 0x1fffe000) as rollable — free for the miner to modify without affecting consensus, and removed from soft-fork signalling so they can’t collide with it. That’s 65,536 version values per header, far more variation than AsicBoost needs (it requires as few as two bits for a 4-way collision). The miner iterates these bits to generate fresh chunk 1 midstates and runs many nonces against each.

How the pool and miner agree on it

For the gain to apply, pool and miner must agree on which version bits the miner may modify, via a Stratum extension called version-rolling (negotiated through mining.configure). The handshake is simple:

Miner -> pool:  mining.configure
                version-rolling.mask = 1fffe000
Pool  -> miner: version-rolling = true
                version-rolling.mask = 1fffe000
Then the pool sends jobs without pinning the full version,
and the miner rolls the masked bits freely.

If the pool doesn’t support version-rolling, the miner falls back to standard mining and runs at lower efficiency. That’s why pool support matters — a pool without it leaves ~13% of its miners’ efficiency on the table. SoloFury implements version-rolling on every stratum endpoint across all five SHA-256 chains, so any AsicBoost-capable miner gets the gain automatically. (Version-rolling is also a building block of the broader protocol overhaul covered in our Stratum V2 vs V1 guide.)

Which chips support AsicBoost (2026)

Every modern Bitmain mining chip supports overt AsicBoost natively in silicon:

MicroBT (Whatsminer) chips have supported overt AsicBoost since the M30 series (2020), and Canaan’s Avalon line since around 2022 — it’s universal among modern hardware. Bitaxe, NerdQAxe and NerdOCTAxe devices running open-source AxeOS enable it by default and negotiate version-rolling during the stratum handshake. (For the full chip lineage, see our ASIC chip evolution guide.)

The actual efficiency gain — measured, not estimated

The theoretical maximum on SHA-256 is around 20%; real-world implementations land at 5-15% depending on how aggressively the silicon is optimized. Braiins (the team behind Slush Pool and Braiins OS) verified the Antminer S9’s AsicBoost in 2018 and measured roughly 13% energy savings in production. Modern chips (BM1370, BM1373) integrate it more deeply and achieve similar or slightly better results. What 13% means in practice:

For a single Bitaxe at home it’s a couple of dollars a year; for an industrial farm of 1,000 S21+ rigs it’s roughly $394,000 per year in saved electricity. Either way, the whole network runs more efficiently — all from one clever observation about how SHA-256 handles its inputs.

The legacy: how AsicBoost reshaped Bitcoin

The saga left lasting marks. The push to neutralize covert AsicBoost added urgency to SegWit, which in turn enabled the Lightning Network and later second-layer work. The BDPL set a precedent for opening mining patents while guarding against litigation, reducing the risk of future hardware monopolies. Stratum V2 drew partly on the lessons of this episode, giving miners more autonomy over transaction selection as a structural answer to the centralization risk covert AsicBoost exposed. And BIP320 version-rolling became mundane baseline infrastructure — the technique that nearly split Bitcoin in 2017 is now something every pool and miner simply speaks. Bitcoin survived an insider-hardware advantage, and the response left the system more transparent and more resilient.

What this means for your miner today

If you own anything from an S19 onward, any Bitaxe, or any Whatsminer M30+: AsicBoost is enabled by default in firmware, and you just need a pool that supports version-rolling for the gain to be active — SoloFury does, on all five chains. You can confirm it’s working in your AxeOS dashboard (Bitaxe), the Bitmain status page (Antminer), or the pool stats, where an active session shows “version-rolling: yes” or similar. The practical warning: if you move from a version-rolling pool to one without it, your hashrate stays the same but your power draw rises by ~10-13% — same work, more energy. Always confirm your pool speaks version-rolling. On custom firmware (Braiins OS+, VNish, LuxOS) the setting is exposed per-rig, but the default is on, and most users should leave it there.

The bottom line

AsicBoost is one of those quiet optimizations that’s invisible from the outside — your miner runs, your hashrate is what it is — while underneath, every modern ASIC skips the unnecessary chunk and saves the unnecessary watts. The 13% doesn’t sound like much until you scale it: across a network running near 980 EH/s, AsicBoost accounts for well over 100 EH/s of effective hashrate that wouldn’t otherwise exist — more than the entire Bitcoin network had in 2018, freed by a SHA-256 quirk and a patent fight that ended up making the system stronger. Every block you find — every Bitaxe lottery win, every BCH solve, every XEC subsidy — is partly a product of it. The math was always there in SHA-256; the community had to fight over how to use it, and overt AsicBoost is what won.