Performance Characteristics

Native Hash Rate

Metric Hemera Plonky3 Goldilocks t=12 Ratio
State width 16 elements 12 elements 1.33x
Total rounds 72 30 2.40x
Permutation field muls ~3,648 ~2,050 1.78x
Input bytes per permutation 56 56 1.00x
Estimated hash rate ~53 MB/s ~86 MB/s 0.62x
Perms for 1 KB 19 19 1.00x

38% native hash rate reduction comes from the wider permutation and additional partial rounds. Throughput per permutation is identical between the two designs. Partial rounds are lightweight (~19 field multiplications each vs ~304 for full rounds), so the round count increase (72 vs 30) overstates the actual computational cost difference.

Proving Cost

STARK trace dimensions change from Plonky3 Goldilocks to Hemera:

  • Trace width: 12 -> 16 columns (~1.33x)
  • Trace length: 30 -> 72 rows (~2.40x)
  • Combined: ~3.2x proving cost per hash

System-level impact depends on what fraction of total proving time is spent on hashing. If hashing is 20% of total proving time, the system-level overhead is ~0.44x (20% x 3.2 + 80% x 1.0 = 1.44x). If hashing is 40% of total proving time, overhead rises to ~0.88x (40% x 3.2 + 60% x 1.0 = 1.88x). Wider state provides security margin that justifies this cost at both operating points.

Steady-State Adequacy

At scale: 10²⁴ cyberlinks with 1% annual update rate.

  • 10^24 x 0.01 / (365.25 x 86,400) = ~317B cyberlinks/sec required
  • Each particle = 64 bytes = ~1 permutation
  • Single core at ~53 MB/s = ~946,000 permutations/sec
  • Single core handles steady-state with ~3x headroom

Burst scenarios (bulk import, migration, recovery) benefit from parallelism. The permutation is independently computable per chunk, scaling linearly with core count. A 64-core machine sustains ~60 million permutations/sec, sufficient for bulk rehash of large datasets.

Dimensions

performance
zheng/docs/explanation/performance
performance characteristics zheng produces proofs that are larger than pairing-based schemes and smaller than FRI-based STARKs, with verification speed that matches or beats both. the concrete numbers matter for system design, so this page presents them without hedging. proof sizes proof size…

Local Graph