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.