Simulations put one-dimensional Ethereum fees against EIP-7999

Victoria Shinder23 September 2026, 08:212 min read

An Ethereum Research post published this week by Fei Wu, with work partly conducted during an internship at the Ethereum Foundation, compares a post-Glamsterdam one-dimensional fee market against selected EIP-7999 configurations under identical simulated demand. The results are reproducible from a linked repository.

The question is whether moving from a single base fee shared by all resources to separate prices per resource actually buys anything. The post tests three one-dimensional benchmarks. Here "one-dimensional" means a shared fee with two-dimensional metering: the shared base fee responds to the larger of regular gas and state gas, whereas EIP-7999 maintains separate execution, data and state base fees with a fake-exponential update rule.

The reported numbers:

• Baseline (64 gas per counted byte floor from EIP-8131/8279, CPSB = 1530): delivers 82.0M-92.6M execution gas per block but permits ~288-403 GiB/year of state growth.
• EIP-8368-style CPSB recalibration: state growth falls to ~121-122 GiB/year, but the higher shared base fee cuts delivered execution to 67.5M-70.2M.
• EIP-8372-style calibration: recovers much of that, delivering 151.9M-177.9M, though both branches leave regular and normalized-state utilization below target, producing ~86-88 GiB/year.
• EIP-7999: maximum-throughput configurations deliver 252.9M-272.6M execution gas; historically anchored ones 173.6M-223.0M. Both hold state growth at 120 GiB/year.

What it means

The comparison isolates the thing a shared base fee cannot do. One price has to clear two markets, so it can be calibrated to hold state growth down or to let execution through, and the benchmarks show it trading one against the other: the configuration that fixes state growth loses roughly a fifth of execution throughput relative to baseline.

State growth is the constraint with the longer half-life. Execution gas is a throughput number that affects fees today; GiB per year accumulates into what every future node must store and sync, and a mechanism that permits 400 GiB/year is making a commitment on behalf of operators who have no vote in it.

⚠️ Read the status carefully. This is research under constructed demand conditions, not a measurement of mainnet and not a decision about any EIP. The demand model, the metering multipliers and the February-May 2026 activity anchor are all assumptions, and the post publishes them so they can be argued with. Nothing here is investment information.