Vitalik Buterin is urging Ethereum developers to set a hard ceiling on the proving cost associated with each unit of block gas, adding a resource-accounting challenge to the network’s effort to accelerate zero-knowledge proof generation.
The issue is not simply how quickly a prover handles an average block. It is whether the most demanding block that Ethereum’s rules allow can still be proved within a predictable hardware and time budget. That distinction could be important for independent operators deciding how much computing capacity they need to keep up with the network.
According to CryptoSlate, Buterin made the proposal in an October 15 post while praising the Airbender team at ZKsync. He identified three priorities for formal verification: ensuring that the mathematical representation of execution matches Ethereum’s specification, limiting proving costs relative to block gas, and protecting the proof system itself.
Faster execution does not automatically mean cheaper proofs
Ethereum’s gas system is designed to measure and constrain computational work in a block. A proving-cost limit would extend that accounting principle to the process of producing a cryptographic proof that the block was executed correctly.
The challenge is that one unit of execution gas does not necessarily represent the same amount of proving work for every operation. Some workloads may be inexpensive for ordinary CPU execution but relatively costly for a zero-knowledge prover. A meaningful ceiling would therefore need to account for deliberately difficult workloads, not only typical blocks or average benchmark results.
EIP-7904 illustrates the difference in scope. Its current analysis recommends no changes to compute gas prices and targets 100 million gas per second on optimized native execution clients. The proposal attributes the improved performance outlook partly to Block-Level Access Lists, which enable more parallel processing of operations that had previously created bottlenecks.
As of October 18, EIP-7904 was an Informational Ethereum Improvement Proposal in Last Call, with a November 1 deadline. Its analysis concerns execution on a bare CPU and expressly does not cover ZK-SNARK proof generation. Faster CPU execution therefore does not by itself establish that proof production will remain within a fixed operating budget.
ZKsync benchmark shows progress, but not a universal guarantee
ZKsync’s Airbender results provide evidence that proving performance is improving. In an October 13 announcement, the company said a single NVIDIA RTX PRO 6000 Blackwell processed 99% of a sample of 7,200 consecutive Ethereum mainnet blocks in less than 10 seconds each, including witness preparation. Every block in the sample was completed in under 12 seconds.
The announcement also referred to live ETHproofs configurations using two and four GPUs. Those results demonstrate performance on the selected blocks, but they do not establish a maximum proving cost for every workload permitted by Ethereum’s rules.
That distinction matters operationally. A normal block may fit comfortably within an operator’s equipment, while a block intentionally assembled to be expensive to prove could determine the reserve capacity required to produce proofs on time. A hard guarantee would need to specify the workloads, software implementations, security settings, hardware and deadline included in the calculation.
Research by Nethermind’s Lin Oshitani, based on January 27 measurements, found that cryptographic precompiles and several modular-arithmetic and division operations could be comparatively expensive to prove per unit of gas. The measured cost varied by prover, and ZK-cycle counts did not translate into identical elapsed times across operations.
The tests used SP1 v5.2.3 with sp1-cluster and RISC0 v3.0.4, alongside reth v1.9.3 and four NVIDIA GeForce RTX 4090 GPUs. They used representative or best-effort worst-case inputs, but did not include block-header validation and accidentally omitted BLOCKHASH. A complete assessment of costs dependent on argument values remained future work.
Possible ways to control the proving burden
The Ethereum Foundation’s January worst-case benchmarking framework links required proving throughput to a selected block gas limit and proving deadline. Under that approach, even a difficult block must fit the intended hardware and time envelope.
An end-to-end test is needed because proving performance is affected by more than the cost of individual operations. Fixed overhead, potentially nonlinear aggregation, hardware configuration and security parameters can all change the result. The expected mix of execution clients and proof systems also influences the calculation.
One possible approach described in the Nethermind study is weighted ZK-gas metering, combined with a per-block ceiling. Proof-intensive operations would consume a separate allowance and could be restricted before a block used all of its ordinary execution gas. Its effectiveness would depend on whether the assigned weights accurately captured difficult inputs.
Other choices would have different effects. Improving an expensive circuit or execution path could reduce the resources required to prove a transaction without changing its execution-gas charge. Opcode repricing, by contrast, would use the existing gas limit: more expensive operations would occupy more gas, leaving room for fewer of them in a block. At an unchanged positive effective gas price under EIP-1559, a higher gas charge for the same successful transaction would also mean a higher payment.
A resource limit and a fee mechanism are separate decisions. A cap can restrict the amount of proving work admitted into a block without determining what users pay or how provers are compensated. EIP-7999, a Draft proposal, takes a multidimensional fee-market approach and separates calldata into its own resource. EIP-8025, also Draft, proposes execution proofs but currently provides no prover incentives and retains re-execution even when proofs are available.
Moneyiar market picture
Moneyiar market data at 22:34 on October 19, 2026, shows Ethereum at $2,405.4494, up 0.97% over 24 hours but down 3.18% over seven days. Shiba Inu was $0.000005014, up 1.87% in 24 hours and down 4.03% over seven days. Uniswap stood at $7.2947, up 2.50% daily and 1.07% weekly, while Aster was $0.70897, up 2.20% daily and 2.08% over seven days.
The next engineering test for Ethereum’s proving strategy is therefore not only to make proofs faster. It is to establish a dependable upper bound for the hardest permitted workloads. EIP-7904’s November 1 review deadline addresses its CPU-performance analysis; the broader proving question requires a separate framework tied to explicit assumptions that operators can plan around.
