Introduction Costs to prove Ethereum blocks are down only. Competing zkVMs have drastically improved their performance over the past few years, with many predicting that proving Ethereum blocks in real time (under 12 seconds) will become a practical reality by 2025. However, if Ethereum is to rely on provers in the not-so-distant future, rigorous research is needed to determine whether certain EVM instructions (i.e., opcodes and precompiles) are particularly slow or computationally intensive to prove. In this post, we use a data-driven approach to identify "prover killer" blocks with particularly long proving times, and then analyze the opcode-level gas consumption patterns to pinpoint the specific opcodes that contribute most to these performance bottlenecks. Methods Data Acquisition and Preprocessing Proofs and associated metadata—including clusters, AWS instance pricing, and zkVM teams—were extracted from Supabase ethproofs.org data. This data was then used to calculate key derived metrics such as proving time and proving costs for each block. Opcode execution traces were obtained via Ethereum's debug_traceTransaction RPC call using Xatu. These traces provided the sequence of opcodes executed along with their gas consumption, allowing us to compute the total real gas used and the execution frequency for each opcode per block.

DALL·E Aug 27 Illustration (1) by Thomas Thiery – Tuesday; August 27th, 2024. Thanks to Julian Ma, Barnabé Monnot, Anders Elowsson and Max Resnick for the very helpful comments and feedback. Introduction A lot of attention has recently been focused on block co-creation by multiple proposers on Ethereum. The multiple proposers idea is broad enough to create compelling narratives and emphasize the need to strengthen the protocol’s censorship resistance (CR) properties. However, the discussion often fails to capture the nuances between different proposals and the various ways to achieve that goal. In this note, I hope to disambiguate gadgets like FOCIL from new protocols like BRAID by highlighting their drastically different approaches to CR and MEV. Disclaimers: