Mining algorithms: SHA-256, Scrypt, RandomX — and why hardware differs
The algorithm a coin picks decides what machine you need: a laptop CPU, a rack of GPUs, or a purpose-built ASIC. How algorithms steer hardware, why some resist ASICs, and what that means for who gets to mine.
The algorithm is the puzzle
Every proof-of-work chain runs one specific hash function as its lottery — that choice is the mining algorithm. The algorithm doesn't change the game (guess a hash below the target, see the mining lesson) but it decides which hardware plays it well, because each function stresses different parts of a machine: raw compute, memory size, memory speed, or branchy program logic.
That makes the algorithm the single biggest fact about a coin for a miner: it sets your hardware bill, your electricity cost per coin, and who you're competing against.
The three hardware worlds
| Class | Design philosophy | Typical algorithms |
|---|---|---|
| CPU-friendly | memory-hard or program-heavy designs that exploit big caches and complex execution — the things CPUs have and GPUs don't | RandomX (Monero and followers) |
| GPU-friendly | parallel-friendly math with a large-but-shared memory footprint; GPUs win on price per stream processor | KAWPOW, Equihash, Groestl, Ethash-family |
| ASIC-dominated | simple, repetitive hashing that fixed-function chips do orders of magnitude more efficiently than any general hardware | SHA-256 (Bitcoin), Scrypt (Litecoin family) |
RandomX is the strongest CPU design in production: it generates and executes random programs in a virtual machine over a ~2 GB dataset sized to overwhelm CPU caches and punish small-memory hardware. Think of a test that requires reading random pages from a thick encyclopedia and following references between them — one person with the book on their desk beats a thousand people each holding a photocopied page. Since its 2019 deployment, no viable RandomX ASIC has shipped; to beat it you'd have to build… a general-purpose CPU.
At the other end, SHA-256 and Scrypt are simple enough that purpose-built chips (ASICs) hash them thousands of times more efficiently than GPUs. Once an ASIC exists for an algorithm, general hardware is priced out — GPUs don't stop working, they just stop being competitive.
ASIC resistance: three techniques
| Technique | How it blocks fixed-function chips | Seen in |
|---|---|---|
| Memory hardness | demand fast random access to gigabytes of RAM — expensive to attach to every hash core | RandomX (2 GB), Ethash-family (DAG) |
| Random programs | make the computation different every nonce, so there's nothing fixed to hardwire | RandomX, KAWPOW/ProgPow |
| Algorithm rotation | change the puzzle periodically to brick whatever ASICs were built | Monero's pre-RandomX era |
How to read an algorithm name on a coin page
- CPU-tagged algorithm → you can mine on machines you already own; think electricity cost, not hardware capex.
- GPU-tagged → hardware is the investment; compare coins per algorithm, since rigs switch freely inside one algorithm family.
- ASIC-tagged → entry means buying dedicated hardware that mines *only that algorithm* — capex risk concentrates on one coin family's economics.
- Merged-mining capable (usually Scrypt-family) → the same hash can pay a second coin; see the AuxPoW lesson.
See it live on pow.family
Project pages show each coin's algorithm and hardware class up front, and the Hardware section lists real benchmarked hashrates and watts per device — the numbers to plug into any rental or purchase decision.