loading…
—
runs
—
machines
—
last updated
Across machines
Proof sizes
Root proof size in KiB per aggregation workload — what gets gossiped to validators each slot. Deterministic per topology (same value across all machines for a given combo); read from any one run that recorded it.
Scaling on c4 (timing vs cores)
Granite Rapids only — same per-core arch across the four c4-standard-* sizes, so the only thing changing is core count. X axis is logical cores (vCPU) since that's what GCP charges per and what Rayon's threadpool sizes itself to; per-workload mean (best run) on the y axis.