EVM Compute Repricing

This dashboard studies the EVM's compute operations — arithmetic, bitwise, comparison, stack, control-flow, hashing, memory, context, and every precompile — from empirical client benchmark runtimes, around a 100 Mgas/s throughput anchor (the rate a plain ETH transfer sustains, which the chain can't exceed). It asks two paired questions:

  1. Bottlenecks — which operations are already too slow to keep up, i.e. run below 100 Mgas/s in the worst case? Those cannot be repriced cheaper.
  2. Reprice — of the operations that do keep up, which are overpriced for their measured runtime and should become cheaper, and by how much?

0 of 74 compute operations fall below the 100 Mgas/s ceiling (the Bottleneck page). Of the rest, at this anchor 67 of 79 fitted parameters could be reduced — open Repricing and drag the anchor to see the costs update live.

Bottlenecks

Bottleneck table →

Per-operation worst-case throughput (Mgas/s) across clients, flagging the ops that run below the 100 Mgas/s ceiling.

Bottlenecks Methodology →

How the worst-case throughput is measured from the benchmarks and compared against the 100 Mgas/s ceiling.

Repricing

Repricing table →

Interactive: set the anchor, see current vs. proposed gas, the diff, and which ops get cheaper — recomputed live in the browser.

Throughput loss →

How much throughput each repricing choice wastes — no reprice, reprice + round, reprice + fractional — per op and over real mainnet traffic.

Runtime model →

Per-opcode NNLS fits of runtime against opcode count, filterable by opcode and client.

Glue runtime →

Per-client fits for the priced glue opcodes that the target coefficients are netted against.

Test reference →

The opcode ↔ benchmark mapping: which test and fixture parameters price each operation, and the gas parameters it produces.

Repricing Methodology →

Benchmark suites, regression model, anchor, worst-case selection, and the throughput-loss method.