A live benchmark experiment

Is your phone using a hash function from 2001?

SHA-256 is everywhere — passwords, certificates, blockchains, file verification. It was designed before the first iPhone existed. There's a faster, structurally stronger alternative. Let's prove it.

Scroll to learn more

Two algorithms. Two eras.

2001

SHA-256

Designed by the NSA. Published by NIST. Built for an era of single-core processors.

Sequential chain
Block 1
Block 2
Block 3
Block 4

Each block must wait for the previous one. One core, no matter how many you have.

  • Sequential — can't parallelize a single message
  • Relies on hardware acceleration to match modern alternatives
  • Vulnerable to length extension attacks
  • 25 years old, showing its age
2020

BLAKE3

Designed by cryptographers Jack O'Connor, Jean-Philippe Aumasson, Samuel Neves, and Zooko Wilcox-O'Hearn. Built for the modern era.

Parallel tree
Root
Node
Node
B1
B2
B3
B4

All blocks processed simultaneously. More cores = more speed. Linearly.

  • Parallel — scales with every core
  • Fast in pure software, no hardware needed
  • Immune to length extension attacks
  • Supports verified streaming

Your device

Detecting...

The benchmark

Four tests, four data sizes. BLAKE3 single-threaded vs. BLAKE3 multi-core (using all your CPU cores) vs. SHA-256 in WebAssembly (fair fight) vs. SHA-256 through your browser's hardware-accelerated Web Crypto API (home advantage).

The verdict

Run the benchmark to see your results.

Open questions

This benchmark is a living experiment. Here's what we're still investigating — and where your device's results might help.

Does multi-core BLAKE3 overtake hardware SHA-256 at 10 MB?

BLAKE3's tree hashing should dominate at large data sizes, but Web Worker overhead is real. We're looking for the crossover point where parallelism definitively wins.

How much would WASM SIMD accelerate BLAKE3?

BLAKE3 is designed to exploit SIMD instructions (NEON on ARM, SSE/AVX on x86). The current WASM implementation doesn't use WASM SIMD — enabling it could significantly widen the gap.

Why do BLAKE3 and SHA-256 WASM tie at 1 KB?

Both show identical throughput at very small sizes. Is this the browser timer resolution floor? Fixed per-call overhead? We're investigating whether this is a measurement artifact or a real performance ceiling.

How do older iPhones compare?

We've tested on recent devices, but iPhone SE, iPhone 11, and other older models have different chips and fewer cores. If you're on an older device, your results are especially valuable.

Should parallel BLAKE3 use full Merkle tree assembly?

Currently, multi-core BLAKE3 hashes independent chunks across workers. A full Merkle tree implementation would combine chunk hashes into a single root — matching the BLAKE3 spec more faithfully and potentially improving large-data throughput.

Curious about the methodology? Read our sources and methodology — or view the source code directly.

Future directions

We're exploring where to take this project next. These are early UI prototypes showing three possible directions — mock data, not yet wired up. Which one resonates?