jonas089 / sha256-zk-benchmark

Risc0, Risc0-Groth16, SP1 bechmarks for SHA256.

Sha256 benchmarks in SP1, Risc0 and Risc0 groth 16

This repository includes different crates for computing sha256 hashs in Risc0 and SP1 execution environments.

The focus of the benchmarks is primarily on proof size.

SP1 circuit

    let mut output_hashes: Vec<Vec<u8>> = Vec::new();
    let rounds = sp1_zkvm::io::read::<u32>();
    for i in 0..rounds {
        let sha_input = [1u8; 32];
        let mut hasher = Sha256::new();
        hasher.update(&sha_input);
        output_hashes.push(hasher.finalize().to_vec());
    };
    let output: Output = Output {
        sha256_hashes: output_hashes,
    };
    sp1_zkvm::io::commit(&output);

sha256 library used: sha2-v0-10-8 = { git = "https://github.com/sp1-patches/RustCrypto-hashes", package = "sha2", branch = "patch-v0.10.8" }

See here

Risc0 circuit

    // read the input
    let rounds: u32 = env::read();
    let mut output_hashes: Vec<Digest> = Vec::new();
    // test sha256
    for i in 0..rounds {
        let sha_input = [1u8; 32];
        let sha_hash: Digest = *Impl::hash_bytes(&sha_input);
        output_hashes.push(sha_hash);
    }
    let output: Output = Output {
        sha256_hashes: output_hashes,
    };
    env::commit(&output);

sha256 library used: use risc0_zkvm::sha::{Digest, Impl, Sha256};

See here

SP1 proof size benchmark

1 hash	10 hashs	100 hashs	1000 hashs
1.44 MB	1.52 MB	2.81 MB	x

SP1 proof speed benchmark

1 hash	10 hashs	100 hashs	1000 hashs
1s	4s	27s	x

Risc0 proof size benchmark

1 hash	10 hashs	100 hashs	1000 hashs
215.6 KB	216.2 KB	256.7 KB	x

Risc0 proof speed benchmark

1 hash	10 hashs	100 hashs	1000 hashs
19s	19s	78s	x

Risc0-Groth16 proof size benchmark

1 hash	10 hashs	100 hashs	1000 hashs
485 B	773 B	3653 B	x

Risc0-Groth16 proof speed benchmark

1 hash	10 hashs	100 hashs	1000 hashs
223s	223s	286s	x

Disclaimer: I had to run the SP1 benchmarks on my Macbook, on Ubuntu I encountered this error:

Machine used for SP1: M2 Macbook Air, 8GB Ram.

The reason why a different machine was used for the SP1 benchmarks is this issue related to curve25519-dalek v4.1.2.

The machine used for Risc0 and Risc0-groth16 was a 32GB Ram, 8-core x86-64 Ubuntu server.