Skip to main content

Benchmarks

Cross-repo source of truth​

Primary benchmark methodology: privacy-ethereum/csp-benchmarks. Per that repo’s README:

  • Results are published quarterly on ethproofs.org/csp-benchmarks as a comparison UI.
  • The harness compares systems along proving time, peak memory, proof size, and preprocessing size.
  • As of Q3 2025, hosted runs use an AWS mac2.metal host (Apple M1, 8 cores arm64, 16 GB RAM) as a stand-in for client-class hardware.
  • A spreadsheet describes planned benchmarking scope: linked from the csp-benchmarks README.

Local CLI numbers​

wallet-unit-poc/ecdsa-spartan2/README.md records recent MacBook Pro M5 timings for prepare / show across 1k–8k JWT sizes, including prove / reblind / verify steps.

Mobile tables​

wallet-unit-poc/README.md lists Prepare and Show times on iPhone 17 and Pixel 10 Pro (setup, prove, reblind, verify).

CI guardrail​

GitHub Actions workflow ecdsa-spartan2.yaml runs the 1k compile + end-to-end prepare/show flow on each PR to prevent regressions (not a full benchmark sweep).