SHA-256 vs. Scrypt: Why ASICs Won Both Mining Races.

Bitcoin, Litecoin, and Dogecoin operate on the same basic principle: computers keep searching for a valid number combination until a block meets the network's requirements. Nevertheless, a Bitcoin miner cannot be pointed at Litecoin or Dogecoin and vice versa. The reason lies in a single design decision from 2011: Litecoin replaced SHA-256 with Scrypt, and Dogecoin adopted this algorithm two years later at its launch. This article explains what distinguishes the two algorithms, how it came about historically, and what practically follows from it for mining operators.

Key Takeaways

• SHA-256 is a pure computing function. It requires practically no memory and can therefore be packed extremely densely into silicon. Bitcoin, Bitcoin Cash, and Bitcoin SV use it.

• Scrypt requires a scratchpad memory in addition to computation, around 128 kilobytes per hash operation for Litecoin. Litecoin, Dogecoin, and a number of smaller chains use it.

• Scrypt was chosen to slow down ASICs. That worked for about two and a half years: the first Scrypt ASICs hit the market in 2014, and since 2015 GPU mining on Scrypt has no longer been economical.

• The hashrates of the two worlds are not directly comparable. Bitcoin sits at around 900 EH/s, Litecoin at around 3 PH/s, a factor of roughly 300,000. Measured in energy, the difference is only around a factor of 20 to 30, because a Scrypt hash costs roughly 20,000 times more electricity than a SHA-256 hash.

• Merged Mining is standard practice on the Scrypt side: a machine does not only mine Litecoin but delivers the same computational work simultaneously to DOGE and several smaller coins. Dogecoin alone contributes around 35 to 45 percent of revenue depending on price levels.

• For operators, the same calculation decides in both worlds: efficiency times electricity price versus revenue per hashrate. The infrastructure such as racks, power, cooling, and noise protection is practically identical, the machines are not.

Two Algorithms, One Principle.

Proof of Work functions identically for all coins discussed here. A miner assembles a block header, modifies a counter variable within it, and runs the whole thing through a hash function. The result is a seemingly random number. If it falls below a target value set by the network, the block is valid and the miner receives the reward. If it falls above, the counter variable is incremented and the process starts again. Billions of times per second.

The network adjusts how low this target value sits via the difficulty, so that the block time remains constant on average. When computing power is added, the difficulty rises; when computing power disappears, it falls.

The only relevant difference between Bitcoin and Litecoin at this level is the question of which function performs the dice-rolling work and which resource it consumes in the process. SHA-256 consumes computing cycles. Scrypt consumes computing cycles and memory accesses. Everything else follows from this one difference: different chips, different manufacturers, different metrics, and two completely separate hardware markets.

The History: How Two Algorithms Became Competitors.

The split was not a technical necessity but a reaction. Bitcoin's hardware arms race displaced all hobby miners within a few years. Scrypt was the attempt to prevent this development. That the attempt failed explains why both worlds are dominated by industrial ASICs today and why Dogecoin would hardly have survived without Litecoin.

SHA-256 and the Launch of Bitcoin.

SHA-256 belongs to the SHA-2 family, standardized by the US agency NIST in 2001. It was never designed for mining, but as a fast, collision-resistant checksum for signatures and integrity checks.

When the Bitcoin network generated its first block on January 3, 2009, it used SHA-256 in a double application: the block header is hashed twice in succession. The target block time was ten minutes, the reward 50 BTC. Mining was done on ordinary processors, and that was explicitly the intention. The vision of a network in which every computer carries roughly equal voting weight was part of the original design.

The SHA-256 Hardware Race: From CPUs to the First ASICs.

This vision lasted about 18 months. SHA-256 consists of a fixed sequence of simple integer operations with no memory requirement, and precisely that makes it ideal for specialized hardware.

In 2010, it became clear that graphics cards handle the task far better than processors because they run thousands of identical computing units in parallel. FPGAs followed in 2011, freely programmable logic chips with better consumption per hash. In early 2013, the first ASICs reached miners: chips developed for a single task, computing Bitcoin hashes.

The efficiency leap was brutal. The first ASIC generation of 2013 sat at around 1,200 joules per terahash. Current devices are below 10 J/TH, an improvement of more than a factor of 100. Within a few months, mining on a home PC was no longer merely unprofitable but hopeless. Anyone who had not ordered ASICs was out of the game.

Scrypt and the Promise of ASIC Resistance.

Scrypt does not come from the cryptocurrency world. Colin Percival developed the function in 2009 for his backup service Tarsnap as a key derivation function for passwords. Its purpose was to make brute-force attacks with specialized hardware expensive. The idea: if every guess occupies not only computing power but also memory, an attacker with a thousand parallel computing units loses their advantage, because they have to pay for memory a thousand times over.

This property is called memory hardness, and it became highly interesting in the Bitcoin environment from 2011 onward. In September 2011, Tenebrix appeared as the first cryptocurrency with Scrypt as its Proof-of-Work function. On October 7, 2011, Litecoin followed, developed by then-Google engineer Charlie Lee: Scrypt instead of SHA-256, 2.5-minute block time instead of ten, 84 million coins instead of 21 million.

The argument was explicit: with Scrypt, graphics cards and processors were supposed to remain competitive, keeping the network more broadly distributed. Notably, Lee himself qualified this position early on, stating in 2014 that ASIC resistance was not a goal of Litecoin. The practical effect in the early phase was more modest than the promise anyway: graphics cards were only about ten times faster than processors on Scrypt, whereas on SHA-256 they were many times faster than that. The gap between hardware classes was smaller, but had not disappeared.

Dogecoin and the Scrypt Wave.

In 2013, Scrypt became the standard for new coins. The reason was economic: all the GPU miners who had lost Bitcoin to ASICs were sitting on functioning hardware and looking for a use. Scrypt coins offered one.

On December 6, 2013, Dogecoin launched, developed by Billy Markus and Jackson Palmer as a fork of Luckycoin, which itself descended from Litecoin. Block time: one minute. Reward: initially random, later a fixed schedule. Alongside Dogecoin, dozens of other Scrypt chains emerged, most of which disappeared within two years.

2014: The ASICs Reach Scrypt Too.

Memory hardness delayed specialized hardware, it did not prevent it. From early 2014, Gridseed shipped the first Scrypt ASIC, a device that could even compute Scrypt and SHA-256 simultaneously via a switch. Innosilicon followed with the A2 Terminator, the first Scrypt chip in 28-nanometer fabrication. KnCMiner also took pre-orders for the Titan, but repeatedly revised the promised performance upward during development and ultimately shipped late. Scrypt mining finally reached the mass market in 2017 with Bitmain's Antminer L3+, which displaced the pioneer devices and whose successors, the L7 and L9, shape the market to this day.

For graphics cards, the matter was settled within a year. But the new ASICs also shifted the balance between the Scrypt chains themselves. It became particularly critical for Dogecoin: its own hashrate declined, and miners constantly switched between Litecoin and Dogecoin depending on which chain was currently yielding more. This pendulum movement destabilized both networks.

The answer was Merged Mining. On August 4, 2014, the Dogecoin team announced Auxiliary Proof of Work (AuxPoW), which was activated at block 371,337 around September 11, 2014. From that moment, Litecoin miners could submit the same computational work for both chains. Within one month, Dogecoin's hashrate and difficulty rose by around 1,500 percent. Since then, both networks have run practically in lockstep; as early as 2019, the vast majority of Dogecoin's hashrate came from Litecoin pools.

The balance after 15 years: Scrypt delayed the arrival of ASICs by about two and a half years and in doing so created a lasting second hardware economy.

How the Two Algorithms Work Technically.

The following sections describe what actually happens in a hash operation, without formulas, but precisely enough to understand the consequences for chips and power consumption. Important upfront: Scrypt does not replace SHA-256 but contains it. Internally, Scrypt uses SHA-256 as a building block and adds a memory phase.

SHA-256 Explained

SHA-256 takes data of any length and generates a 256-bit fingerprint from it. The function is deterministic, meaning the same input always produces the same result, and it is practically irreversible: the input cannot be calculated back from the fingerprint. Change a single bit of the input, and the result looks completely different.

Internally, SHA-256 splits the data into blocks and pushes them through 64 identically structured rounds. Each round consists of additions, bit shifts, and logical operations on eight intermediate values. There are no branches, no jumps, and no memory access. In mining, the 80-byte block header is run through this procedure twice and the result is compared with the target value.

For chip design, this is an ideal case. A SHA-256 computing unit is so small that a modern chip can run hundreds of thousands of them in parallel. Scaling here simply means: more copies of the same circuit on more area.

Scrypt Explained

Scrypt runs in three phases.

First, the function generates a starting value from the input; this is where SHA-256 comes in. Then follows the decisive phase: from this starting value, a large table of pseudorandom data is built and stored in memory. This table is then traversed again, with the order of accesses depending on the content of the table itself. With each access, the data is mixed and computation continues. Finally, the result is condensed into a hash.

The trick lies in the second and third phases. Because the access order depends on the intermediate result, it cannot be predicted and the table cannot be shrunk. Anyone wanting to save memory has to recompute the values on every access, and that costs more than it saves.

How large this table is, is defined by a parameter. Litecoin uses a setting that results in about 128 kilobytes per hash operation. Dogecoin and the other Scrypt chains use the same values, which is why the same hardware works everywhere.

Why the Memory Requirement Makes the Difference.

On a chip, space is the scarce resource, and memory needs a lot of it. While computing units can be shrunk further with every new chip generation, this applies to memory cells only to a limited extent. Each Scrypt computing unit needs its own working buffer, which occupies the largest share of the area. Instead of hundreds of thousands of computing units as with SHA-256, significantly fewer therefore fit on a chip.

This is the reason for all the differences in metrics: fewer hashes per second per device, significantly higher energy expenditure per hash, and a smaller gap between general-purpose hardware and specialized hardware. Precisely this is what the hope for ASIC resistance was based on.

Why it still did not work out has a structural reason. 128 kilobytes is little for a modern chip; today, thousands of such buffers can be placed on a single chip. A significantly larger parameter was never an option, though: every full node must verify every block header itself. If a verification required hundreds of megabytes of memory, running a node on ordinary hardware would be impossible, jeopardizing exactly the decentralization that was meant to be protected. Memory hardness in the block header is therefore capped, and within this window, ASICs win.

Scrypt and SHA-256 in Direct Comparison.

Why Hashrates Are Not Directly Comparable.

Bitcoin sits at around 900 exahashes per second, Litecoin at around 3 petahashes per second. That is a factor of about 300,000, and this number says very little.

The reason: a Scrypt hash is a completely different piece of work than a SHA-256 hash. Converting the current top values, a Scrypt hash on today's hardware costs about 20,000 times more energy than a SHA-256 hash (0.21 J/MH versus 9.5 J/TH).

A more meaningful comparison is via energy input. Extrapolating the network hashrate with the efficiency of current devices, the Bitcoin network consumes about 15 to 20 gigawatts, the Scrypt networks about 0.6 to 0.8 gigawatts. That is a factor of about 20 to 30, which is substantial, but far removed from 300,000. Litecoin's security budget is therefore not microscopically small, it is a good order of magnitude smaller.

A second point concerns the numbers themselves. Hashrate cannot be measured, only estimated from difficulty and observed block times. Different observation windows produce different values. For Dogecoin, figures between 1.7 and 4.2 PH/s were found in 2026; for Bitcoin, depending on the provider, values between about 900 EH/s and over 1 ZH/s. Anyone using such numbers in calculations should note the source and averaging period. Difficulty values of different chains are likewise not comparable with each other, because each chain uses its own reference basis.

Block Times, Rewards, and Emission.

Bitcoin targets ten minutes per block and currently pays 3.125 BTC. The reward halves every 210,000 blocks; the next event sits at block height 1,050,000 and is expected for April 2028, when it drops to 1.5625 BTC. Currently, about 450 new BTC are created per day.

Litecoin targets 2.5 minutes and pays 6.25 LTC. Halving occurs every 840,000 blocks; the next event is expected around the end of July 2027. The last halving took place in August 2023, and the network hashrate has since risen from under 0.8 PH/s to around 3 PH/s, more than tripling despite the halved reward.

Dogecoin is the special case. Since block 600,000, the reward has been fixed at 10,000 DOGE per block with a one-minute block time. There are no further halvings and no supply cap. Around 14.4 million DOGE are created daily, about 5.3 billion annually. Because the amount remains absolutely constant, the relative inflation rate decreases year by year.

Which Coins Run on Which Algorithm.

An ASIC can only compute the algorithm it was built for. The algorithm therefore also determines which other coins a device can switch to. The following sections show which chains come into question in each case and how realistic a switch is.

SHA-256 Coins

Besides Bitcoin, Bitcoin Cash and Bitcoin SV are the most relevant, both split off from Bitcoin in 2017 and 2018 respectively and with an identical reward schedule: both likewise pay 3.125 units per block at a ten-minute block time. Further candidates are eCash and Fractal Bitcoin, plus a few dozen older forks, most of which no longer play a role.

Since the same devices work on all SHA-256 chains, profitability settles at a similar level everywhere: if a smaller chain temporarily becomes more rewarding, miners switch there until the difficulty catches up and the advantage disappears. On smaller chains, an individual miner finds a block themselves more often, but is paid in coins that are harder to sell. A special case is DigiByte: the chain uses several algorithms in parallel, including both SHA-256 and Scrypt.

Scrypt Coins

The load-bearing chains are Litecoin and Dogecoin. Added to these is a group of smaller, mostly Dogecoin-adjacent projects that hang off Litecoin via Merged Mining: Bellscoin, Pepecoin, Luckycoin, Junkcoin, and Dingocoin. Older Scrypt coins such as Verge, Auroracoin, or GoldCoin still exist but no longer play an economic role.

The structure is notable: apart from Litecoin and Dogecoin, practically none of these chains can be mined independently. They live on being secured as a byproduct of Litecoin's work.

Merged Mining: One Machine, Two Chains.

Merged Mining exploits the fact that a Proof of Work only proves that work was done, not for whom. A smaller chain can specify that it accepts the work of a larger chain as valid proof, provided this work contains a reference to its own block. The miner computes once and submits the same result to both networks, without additional computing or electricity expenditure. The ledgers, rules, and rewards remain completely separate; only the computational work is shared.

The technique dates back to Bitcoin's early days: first described as a concept in 2010, Namecoin has been mined together with Bitcoin since 2011. Dogecoin adopted the mechanism three years later.

Today, the distribution looks like this. On the SHA-256 side, Namecoin, Elastos, Hathor, Syscoin, Rootstock, and Fractal Bitcoin, among others, hang off Bitcoin. On the Scrypt side, Litecoin is the parent chain for Dogecoin, Bellscoin, Pepecoin, Luckycoin, Junkcoin, and Dingocoin; large pools such as F2Pool, ViaBTC, and AntPool activate these chains automatically.

The economic difference between the two worlds is substantial. For Bitcoin, merged mining revenue is a footnote compared to the block reward. For Scrypt, the side chains contribute a significant share of income; Dogecoin alone accounts for about 35 to 45 percent depending on the price ratio. Running a Scrypt device without Merged Mining therefore does not give up a few percent, but roughly a third of revenue.

Three caveats apply. First, the pool decides which chains run alongside, not the operator. Second, revenue accrues in partly very illiquid coins, for which minimum payout thresholds apply and whose sale may involve discounts. Third, the lineup is not stable: AntPool discontinued support for several smaller chains as of July 10, 2026. Calculations built on the entire coin basket should therefore rest on the core, LTC and DOGE, and treat the rest as a bonus.

The Infrastructure Is Algorithm-Neutral.

Anyone designing a site for ASIC mining does not design it for one algorithm. Connected load, circuit protection, airflow, filters, noise protection, networking, and monitoring are practically the same for a 3.5-kilowatt Scrypt device and a 3.5-kilowatt SHA-256 device. A site that can run SHA-256 hardware can also run Scrypt hardware.

Two differences are still relevant for planning. First, cooling technology: on the SHA-256 side, water and immersion cooling are established, with corresponding requirements for cooling circuits, distribution, and maintenance concepts; on the Scrypt side, air remains the rule. Second, power density: a single 3U rack unit with 1.16 PH/s notionally replaces several air-cooled devices, but at around 11 kW also draws correspondingly more power in the same space. Power supply and circuit protection in the rack must be dimensioned for this. Anyone wanting to run both plans for the more demanding case.

The machines themselves, however, are not transferable. A SHA-256 ASIC cannot compute Scrypt and vice versa. That is the actual lock-in, and it sits in the device, not in the building.

Diversification Across Algorithms.

What splitting across both algorithms actually delivers: different revenue drivers and different calendar dates. The Scrypt side depends on LTC and DOGE plus side chains, the SHA-256 side on BTC. The halvings are staggered, Litecoin expected in July 2027, Bitcoin expected in April 2028, so both revenue streams do not collapse at the same time. Dogecoin has no cliff at all anymore.

What it does not deliver: independence from the crypto market. Both revenue sides correlate strongly with the overall market environment: in a broad downturn they fall together, in an upturn they rise together.

What it costs: duplicate manufacturer relationships, duplicate pool and wallet setup, duplicate spare parts inventory. The Scrypt secondary market is considerably thinner, which makes exiting a fleet more expensive. And because the hardware cannot be repurposed, every purchase decision is a bet on one algorithm over the entire service life.

Both Algorithms from a Single Provider: Mining with Bitkern.

Anyone who does not want to choose between the two worlds after reading this does not have to. Bitkern offers hardware and hosting for both algorithms: on the SHA-256 side, current Bitcoin miners such as the Antminer S21 XP; on the Scrypt side, the Antminer L9 for Litecoin and Dogecoin including Merged Mining. Through Bitkern PRO, more than 20 hosting locations worldwide are available for this, with electricity prices from $0.049 per kWh and 24/7 support.

DISCOVER PRODUCTS

Anyone unsure which algorithm or which setup fits their own strategy is best served by booking an appointment with a Bitkern expert.

BOOK A CONSULTATION

Sources
https://hashrateindex.com/blog/hashrate-index-roundup-august-17-2026-2/
https://cryptorank.io/news/feed/ba5a0-bitcoin-miner-exit-hash-rate-slump
https://www.asicminervalue.com/coins/ltc-litecoin
https://www.kucoin.com/blog/dogecoin-mining-profitability-2026-guide
https://www.coindesk.com/markets/2014/05/14/kncminer-reveals-additional-titan-scrypt-asic-specs
https://litecoin.watch/articles/understanding-the-scrypt-mining-algorithm
https://www.coindesk.com/markets/2014/08/04/dogecoin-to-allow-litecoin-merge-mining-in-network-security-bid
https://www.binance.com/en/research/analysis/merged-mining
https://support.viabtc.com/hc/en-us/articles/11477430615439-LTC-Merged-Mining-Coins-Mining-Tutorial
https://blocksbridge.substack.com/p/hidden-message-bitmain-antminer-s23
https://xonmining.com/product/bitmain-antminer-s23-xp-hyd/
https://asicminerinsider.com/blog/antminer-l9-17-gh-s-review-is-it-still-profitable-in-2026
https://bt-miners.com/products/elphapex-dg2-litecoin-dogecoin-miner-20-5gh-s-bt-miners/
https://www.coingecko.com/en/coins/bitcoin/bitcoin-halving
https://en.wikipedia.org/wiki/Bitcoin