zkEVM2026-07-08_006

L1-zkEVM Breakout #006

2026-07-08 3 decisions 462 transcript lines

Transcript

At a glance

ELF loading standard adopts execute-and-read semantics for executable segments; execute-only interpretation is discouraged
Direction adopted: Ethereum guest programs must be built with unmodified stock toolchains to limit trusted computing base
Bump allocator not recommended for production guest programs due to OOM liveness risk at higher gas limits; sophisticated allocators preferred
Recursive proof chain-prefix PoC spec and implementation — Francesco, coming days
Next zkEVM release to include both test fixtures and Geth-filled benchmark fixtures

Detailed notes

Allocator Safety

  • •Bump allocator (BAM) is fastest but causes OOM at current ~0.5 GB limit for high-gas blocks; increasing to 1.5 GB bypasses failures but leaves little headroom — sophisticated allocators (e.g., DLmalloc) recommended for safety00:53:07
  • •OpenVM 2.1 will support 4 GB of guest memory (chat: Yi Sun); most other zkVMs may not exceed 1 GB (Marcin)00:57:10

Spec And Test Updates

  • •zkEVM spec v0.50 released: rebased on Glamsterdam and devnet 6; supports external transition tools (Geth) for benchmark fixture filling00:07:21
  • •SSZ max-length refinements in guest input will change hash tree root calculations — all implementers must review and adjust00:08:43
  • •Hive dashboard now tracks ethrex and Besu execution-witness compliance plus first guest program test runs in a single view00:13:11
  • •Bug fixed in Geth execution-witness generation (merged); benchmark fixtures refilled and looking correct00:14:04

Evm Formal Specification

  • •EVM SAIL spec (evm-sail) achieves 100% EAST test compliance; enables extraction to Lean, Isleris/Iris for binary compliance proofs and RISC-V guest program verification00:41:44
  • •Goal: SAIL spec reaches feature parity with Python execution specs, allowing core devs to contribute directly to a single formal specification00:43:01

Guest Api Standardization

  • •8 guest API standards accepted (ISA I/O, crypto accelerators, termination, memory, static library, linker script); ELF loading and accelerated memory operations open for review00:19:13
  • •ELF loading standard: recommendation is execute-and-read semantics for executable segments; execute-only (Zisk current) is discouraged as no toolchain guarantees instruction-only text sections00:20:13
  • •Precompiles in static libraries must not allocate on the guest heap; a separate isolated scratch region is required — other vendors should audit for this collision00:24:14
  • •Direction flagged: Ethereum guest programs should be built with stock unmodified toolchains to minimize trusted computing base00:24:59

Sync Protocol And Recursive Proofs

  • •Weak-subjectivity sync can require up to 113,000 slots (~26.9 GB of proof content); recursive execution proofs (SSZ Merkle tree or MMR) proposed to enable gossip-based chain-prefix authentication00:26:56
  • •Recursive proof approach would simplify consensus clients by eliminating complex request-response sync; proof-of-concept spec and implementation planned imminently00:31:44

Prover Infrastructure And Benchmarking

  • •EF cluster scalability: GPU proving scales near-linearly from 1 to full machine on large blocks; CPU-bound bottleneck observed between 2 and 4 GPUs00:34:16
  • •OOM kills still observed on mainnet EST 6M-gas blocks; root cause under investigation00:35:04
  • •New ZK-Boost release adapts to current compatible input format; proof-request and verification APIs updated to require chain config as public input00:35:58

Action items

  • •Review zkEVM spec v0.50 SSZ max-length changes and update hash tree root calculations accordinglyAll guest program implementers
  • •Add Nethermind (and other clients) to execution-witness Hive dashboard by adding a branch entry to the dashboard repo JSONNethermind and other EL teams
  • •Audit precompile static libraries for guest-heap allocation collisions; add isolated scratch region if affectedzkVM vendors

Agenda

From the call issue #2148Show

Key decisions

  • ELF loading standard adopts execute-and-read semantics for executable segments; execute-only interpretation is discouraged

    To limit trusted computing base, Ethereum guest programs must be built with unmodified stock toolchains.
  • Direction adopted: Ethereum guest programs must be built with unmodified stock toolchains to limit trusted computing base

    To limit trusted computing base, Ethereum guest programs must be built with unmodified stock toolchains.
  • Bump allocator not recommended for production guest programs due to OOM liveness risk at higher gas limits; sophisticated allocators preferred

AI Disclaimer: Some content or metadata on EIPsInsight may be AI-inferred or automatically compiled. If you find any discrepancy, please contact us at dev@avarch.org.