EIP-4844

EIP-4844 is the Ethereum protocol upgrade (Dencun, March 2024) that handed rollups a discount lane for publishing data: up to six fixed-size blobs per block, priced by their own fee market, pruned after roughly 18 days. It was the single biggest cost cut in L2 history, and it is why the modular-Ethereum rollup model stopped needing trusted data committees to stay cheap. How it works Before Dencun, a rollup posted its compressed transaction batches to Ethereum as calldata — permanent data charged at ordinary execution-gas prices. After Dencun, the same batches ride inside blobs: 128 KB segments (capped on-chain at ~0.375 MiB per blob) attached to the block but executed by no one. No validator runs blob code; they only check that the blob exists and is referenced, which keeps blob data cheap. The economic machinery is a separate, self-clearing fee market: The blob gas target is 3 blobs per block, with a hard maximum of 6 — the buffer that absorbs short spikes. A base-fee multiplier (like EIP-1559 gas) raises blob fees when blocks exceed target and cuts them when they run below. Blobs carry a commitment-to-data pairing: the block header commits (KZG) to blob contents so light clients can verify that posted data matches what the rollup claims. Nodes keep blobs for ~18 days — long enough for optimistic-window fraud proofs or ZK verification — then prune them, so blob storage never inflates into permanent datacenter-grade history. A rollup never touches normal gas for its batches anymore. It stakes a small blob-fee deposit, posts the blob, and pays in the blob fee market — an order of magnitude cheaper per byte than the calldata lane. Why it matters for decentralization Data availability is the load-bearing wall of the modular stack. If posting batches to Ethereum cost 10x what it does, every rollup would face pressure to offload that data to a centralized committee — "just trust our sequencer and our DAC" — and the Ethereum-security inheritance that defines the L2 model quietly disappears. Blobs keep that wall standing cheaply. They also prove the road to danksharding: the base layer can scale its data capacity without raising the cost of running a validator into datacenter territory, which is what would actually kill home-node decentralization. Read the method behind our pillar scoring on the W3D methodology page, or compare how the four pillars play out on the Ethereum audit. Example: what Dencun did to L2 fees Look at any OP Stack or rollup chain after 13 March 2024. Sending an L2 transaction used to pay for L1 data posting at calldata prices; the week after Dencun, data-posting costs on most rollups fell 85–95%, and retail L2 transaction fees dropped to a few cents or less even during busy periods. The blob space that made Base's explosive growth possible — sub-cent transfers during peak memecoin mania — only exists because of this upgrade. DimensionBefore EIP-4844 (calldata)After EIP-4844 (blobs) Data permanencePermanent on-chain~18 days, then pruned (archives retain) PricingExecution-gas base feeSeparate blob base fee, up to ~10x cheaper per byte CapacityCompetes with normal tx space3 blobs target / 6 max per block Rollup fee impactFees track L1 congestionData-carrying fees down 85–95% on most L2s Node requirementsHistory retained foreverPruned — validator hardware stays modest You can price the difference yourself with the L2 gas estimator in the W3D Terminal, which models the blob fee market rollups actually pay into. Risks & limitations Finite blob space: at 3 blobs target, sustained demand peaks still price small rollups out during manias; capacity only grows via more blobs per block, bigger blobs, or data-availability sampling. Pruned history: after ~18 days the canonical chain no longer stores blob contents, so permanent history depends on third-party archivers and indexers — a data-availability window, not a permanence guarantee. Cheaper ≠ free: blob fees reintroduced a floor even during calm periods; small chains occasionally post more blobs than they need, wasting the allowance. Concentration tail risk: cheap data re-centralizes activity onto a handful of dominant sequencers — data cost is solved, sequencer centralization is not. Spec deep dive: the blob transaction and its prices The upgrade shipped as part of the Dencun hard fork on Ethereum mainnet on 13 March 2024, and its economics are precise enough to replicate: A type-3 (blob) transaction references blobs it brings in its body; the header exposes blob_gas_used, and execution clients burn exactly what the blob fee market charges. Target vs cap: blocks target 3 blobs but may include up to 6. Sustained above-target usage accumulates excess that drives the current blob base fee up by the EIP-1559-style multiplier; below-target usage decays it back toward the 1 wei minimum. KZG polynomial commitment: each blob is a polynomial evaluated by the prover; the header keeps a single versioned commitment to all blobs so anyone can later verify a blob was exactly what was committed — without downloading and re-hashing gigabytes of history. Point-evaluation precompile (0x0a): an EVM precompile that takes a commitment + value + proof and verifies the evaluation on-chain. Rollups call it inside their L1 transaction to prove "the blob I posted matches this state batch" entirely on-chain. Sidecar gossip: full blobs travel as sidecars attached to block gossip, so validators store them only during the ~18-day window and can stay light on disk forever after. For protocol researchers these are the four rails that let you rebuild the whole design from the spec: the transaction type, the fee market, the commitment, and the pruning window. Each one exists to keep one specific cost off the everyday execution market. Frequently asked questions Did EIP-4844 lower my wallet fees? It lowered L2 fees dramatically — an Arbitrum or Base swap went from dollars to cents — because rollup batches ride blobs. Your L1 wallet transaction is unchanged; Ethereum execution still costs execution gas. Are blobs permanent? No. Blob data is pruned after about 18 days by design — long enough for fraud or validity verification, not for archival. Rollups and explorers keep their own permanent copies; the protocol does not. What comes after EIP-4844? Full danksharding: many more blobs per block plus data-availability sampling, so light clients can verify blob availability without downloading everything. That is the upgrade path that keeps Ethereum's data lane both huge and home-node-friendly. Sources & methodology EIP-4844 spec (Ethereum foundation) — the blob transaction type, fee market math, and KZG commitments. L2Beat — independently verified data-availability and rollup risk profiles. W3D methodology — how the four pillars (infrastructure, capital, governance, software) are scored; dataset on GitHub. Related terms Blobs · Data availability sampling · Rollup · OP Stack · Shared sequencing Chain audits: Ethereum · Arbitrum · tool: L2 gas estimator