A Forest of Levers: The Performance of Orchard
summary
Focuses on reducing verifier time and proof size for Orchard by using Halo 2 primitives, floor planning and gadget/chip abstractions. Circuit uses degree bound 9, 192 constraints and 10 advice columns; selector combining compresses ~40–50 selectors to ~15 and query-set pruning removes five query sets. API batches roughly ~100 inversions into a single inversion, and region-based floor planner packs merkle hashing across chips. Result: two-action Orchard proves in ~1.5s on an M1 with a base proof of ~2.7KB + ~2.2KB per action, improving prover speed and verifier cost. Halo 2, regions and chips
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:34What does "performance" mean? Provertime
- 3:36In one slide: Halo 2
- 5:47PLONK-style R1CS
- 6:34PLONK-style circuits - Querying cells
- 10:24How circuits become proofs in Halo 2
- 12:29What affects performance?
- 14:57Circuit structure - Rows vs Columns
- 18:04Circuit structure - Query optimisations
- 21:56Circuit structure - Degree bound
- 24:50API design - Regions
- 27:59API design - Gadgets and chips
- 35:01API design - Other affordances
- 38:36Feedback mechanisms
- 41:12Floor planning - Region layout packing
- 42:38Selector-combining optimisation
- 44:55Prover pipelining
- 47:50Benchmarks!
related talks
45mDesign a Proof Of Solvency protocol with Halo2 Workshop
43mPushing the Limits of Zero Knowledge Applications
02h 50mFunctional Array Funhouse Intensive - Part 3 of 5
22mPowered by Less: Low Power Circuit Synthesis
24mZK: Accelerating DA Interop
59mPerformance matters - new tricks for old dogs
44mFunctional Array Funhouse Intensive - Part 5 of 5
16mWriting Optimized Solana Programs