Glamsterdam Devnet-7 Tests a 300M Gas Limit

Ethereum developers launched Devnet-7 on July 14, the last dedicated test network for the Glamsterdam upgrade, and followed it with a mild freeze of the specification. The network is testing a block gas limit of 300 million, against a mainnet range currently near 60 million and well above the 200 million that had been quoted as the Glamsterdam target through June. Devnet-7 also carries EIP-8037, a repricing of state access that reached final draft and makes deploying a contract roughly ten times more expensive and creating a new account about 8.5 times pricier. It replaces Devnet-6, which ran at only about 80 percent participation while developers chased a storage-read disagreement in the Nethermind client. None of this touches the live chain.

A disposable network built to break

A devnet is a throwaway blockchain that core developers spin up among themselves to run an upgrade's code before anyone depends on it. It carries no real value and can be abandoned and restarted when something breaks, which is the point: its failures are cheap. Devnet-6 showed the pattern. Participation sat around 80 percent, meaning a meaningful slice of that network's validators were not attesting normally, and the cause traced to Nethermind, one of several independent programs that implement Ethereum's rules, disagreeing with the others about a storage read. Two clients reading the same state and returning different answers is precisely the class of bug a devnet exists to surface. The mild spec freeze that came with Devnet-7 means the changes are meant to stop moving while they are tested.

Five times mainnet, and past the stated target

Gas is Ethereum's unit for work: every operation a transaction performs costs a set amount, and the block gas limit caps how much total work one block may contain. Raising it lets more transactions fit per block, the direct lever on throughput, at the cost of every node executing, propagating and storing more. The 300 million under test is roughly five times the current mainnet range near 60 million, and it sits well clear of the 200 million quoted through June, so the devnet is deliberately being pushed past the number the upgrade has been discussed around, not to it. The overshoot is the test. A parameter of 300 million on a devnet is a setting, not a mainnet commitment, and as of mid-July the live chain still runs near 60 million.

EIP-8037: charging for state by the byte

State is the data every node must keep to validate the chain: account balances, contract code, contract storage. Unlike transaction history, it cannot be pruned away without cost. EIP-8037 reprices state access on a cost_per_state_byte model, parameterised in bal-devnet-6 to target around 60 GiB of annual state growth at a 300 million gas limit. The arithmetic behind the roughly tenfold rise in deployment cost is mechanical. Quintupling block capacity at unchanged prices would let state grow several times faster, so the price of the operations that add bytes has to rise to hold growth near the target. Deploying a contract writes its whole bytecode into state permanently, which makes it the operation repriced hardest; creating a new account, a smaller write, gets about 8.5 times pricier. Charging for permanent storage rather than one-off execution also points the fee schedule in the same direction as the long-term push to shrink the chain toward near-zero state.

Between a devnet figure and a shipped one

Devnet-7 also fixes timings for ePBS, the change that writes the split between block proposers and block builders into the protocol itself: a 3-second attestation deadline, 6 seconds for the payload and 9 seconds for the payload-timeliness committee. EIP-7688 and EIP-8282, the builder deposit and exit contracts, were late additions to the set. Whether 300 million survives testing is a separate question. The repricing is still a draft and 300 million a devnet figure; the higher deployment costs bite only if both carry through. Gaps between announced plans and delivered code are familiar across the stack. Base's native token standard arrived two weeks after the hardfork meant to carry it, and Glamsterdam's remaining testing still has to clear before a date is worth quoting.