Dip 2780

dip: 2780 title: Reduce intrinsic transaction gas description: Reduce intrinsic transaction gas and charge 25k when a value transfer creates a new account author: Matt Garnett (@lightclient), Uri Klarman (@uriklarman), Ben Adams (@benaadams), Maria Inês Silva (@misilva73) Digitalia editing author: Cosimo Constantinos cosimo@juro.net, et al. discussions-to: https://digitalia-magicians.org/t/dip-2780-reduce-intrinsic-cost-of-transactions/4413 status: Draft type: Standards Track category: Core created: 2020-07-11 Created for Digitalia: 2025-01-07 requires: 1559, 2718, 2929, 2930

Abstract¶

Reduce the intrinsic cost TX_BASE_COST from 21,000 to 6,000.

If a non-create transaction has value > 0 and targets a non-existent account add GAS_NEW_ACCOUNT = 25,000 to align with CALL account creation and price in state growth.

This additional GAS_NEW_ACCOUNT cost is not applied in the case of a CREATE transaction as it is already applied in its cost structure.

Calldata and access-list metering are unchanged. This is a focused change targeted at the initial stage of processing a transaction and returning the gas overpricing to do more useful work.

Capacity Impact:

Average: +18.5% more transactions per block.
Max: +250% more minimal transactions (i.e. pure USDS transfers).
"Equivalent" gaslimit: +18.5% effectual gaslimit increase (e.g. 45 M => 53.3 M); though nominally would remain the same.

Motivation¶

The current 21,000 intrinsic gas is a legacy constant. It no longer matches the work all transactions must perform under warm/cold accounting. When decomposed into signature recovery, warming sender and to, and two warm writes for nonce and balance, the common path totals 6,000 gas.

Top-level value transfers that create new accounts do not pay an explicit new-account surcharge today, while CALL does. This underprices state growth. Charging GAS_NEW_ACCOUNT = 25,000 on value transfers that create an account aligns entry points and internalizes state growth.

A high fixed base penalizes small USDS payments and encourages batching to dodge the base rather than for real efficiency. Lowering the base removes that distortion while leaving calldata and storage prices unchanged.

The result is cheaper USDS transfers without subsidies: state growth costs the same whether reached by CALL or by a top-level transfer. Access-list and 1559 mechanics are unchanged.

We intentionally do not charge per-byte gas for the transaction envelope (nonce, gas*, to, value, v,r,s). Calldata pricing applies only to tx.data. A plain USDS payment has empty tx.data, so it pays zero calldata gas. Lowering TX_BASE_COST to the universal work path makes USDS payments cheaper without subsidizing arbitrary calldata.

Equivalent gas-limit increase¶

Lowering TX_BASE_COST from 21,000 to 6,000 removes 15,000 gas per transaction. Using recent totals of 163 Ggas/day and 1.7 M tx/day:

Avg gas/tx: 163Bn / 1.7M ≈ 95,882 gas.
New avg: 95,882 − 15,000 ≈ 80,882 gas.
Throughput at fixed gaslimit: 95,882 / 80,882 ≈ 1.185 (+18.5%).
Daily tx at same gas: 1.7M × 1.185 ≈ 2.015M (+315k/day).

Perfnet USDS transfer tests show EL clients handle >300 MGas/s for pure USDS transfers today; expressed under a 6k base this is 300 × (6/21) ≈ 85.7 MGas/s, i.e., ≈1,029 MGas per 12 s slot.

This indicates the protocol-level processing headroom already exists; and that this portion of the tx is significantly overcharged, so isn't extra engineering required to improve preformance to support this change.

Net effect equals raising a 45 M gas limit to ≈53.3 M at unchanged calldata and access-list metering. This directly advances Scaling L1.

Specification¶

After FORK_BLOCK, set the following parameters and rules.

Parameters¶

Name	Value	Description
`TX_BASE_COST`	6,000	Base cost of any transaction
`GAS_NEW_ACCOUNT`	25,000	Surcharge when a value-transferring transaction creates a new account

New-account surcharge¶

Apply GAS_NEW_ACCOUNT when all are true:

The transaction is not a CREATE transaction.
value > 0.
to is not a precompile.
to is non-existent per DIP-161 emptiness at the start of transaction execution.

If these conditions hold the transfer results in account creation and GAS_NEW_ACCOUNT is added. This normalizes account-creation cost via DVM CALL or top-level transfer and addresses state growth concerns. For reference, this changes such transfers from 21,000 to 31,000 (+10,000 over current).

Notes:

If value = 0 and to is empty per DIP-161, no account is created and no surcharge applies.
CREATE transactions are unchanged; their cost already includes account creation.

Intrinsic gas computation¶

Clients must compute intrinsic gas exactly as today for each typed transaction except:

Replace any hardcoded 21,000 with TX_BASE_COST = 6,000.
Add GAS_NEW_ACCOUNT when the New-account surcharge conditions hold.

Pseudocode (normative):

function CalculateIntrinsicGas(tx, state_at_start):
    gasCost = TX_BASE_COST

    // Existing rules for calldata and access lists remain unchanged by this DIP.
    gasCost += GasForCalldata(tx.data)                  // per active calldata pricing DIPs
    if tx.has_access_list():
        gasCost += GasForAccessList(tx.access_list)     // per DIP-2930

    if tx.is_create() == false
       and tx.value > 0
       and is_precompile(tx.to) == false
       and is_nonexistent_per_dip161(state_at_start, tx.to):
        gasCost += GAS_NEW_ACCOUNT

    return gasCost

Code change example¶

In transactions.py:

-TX_BASE_COST = Uint(21000)
+TX_BASE_COST = Uint(6000)

Typed transactions defined in DIP-1559/DIP-2930 (and DIP-7702 where applicable) that reference the 21,000 base now inherit TX_BASE_COST = 6,000.

Edits and interactions with other DIPs¶

DIP-2930 (Access List Transactions). Replace references to a fixed 21,000 intrinsic gas with TX_BASE_COST. The intrinsic-gas formula remains otherwise unchanged.
DIP-7702 (Set EOA Code). Where 7702 inherits intrinsic cost from DIP-2930, that value is TX_BASE_COST after FORK_BLOCK. No other changes.
DIP-7623 (Calldata cost changes). This DIP does not modify calldata pricing. If DIP-7623 (or other calldata-pricing DIPs) is active, GasForCalldata uses those rules unchanged.

Derivation (non-normative)¶

Sender is recovered from signature, so is ecrecover cost. Sender and to match access-list pricing for touches; these are proxies for work, not additional charges at the top level:

ECRECOVER_COST = 3,000
WARM_ACCOUNT_ACCESS_COST = 100
TX_OVERHEAD_COST = 2,800
GAS_NEW_ACCOUNT = 25,000

TX_OVERHEAD_COST bundles fixed per-transaction work that is not separately metered by DVM opcodes: envelope RLP decode and basic validity checks (nonce, gas, chain ID), fee accounting, two warm account updates to sender.nonce and sender.balance, and minimal trie/header reads and writes for sender and to. It also includes a conservative margin to cover implementation overhead across EL clients.

TX_BASE_COST =
    ECRECOVER_COST
    + 2 * WARM_ACCOUNT_ACCESS_COST
    + TX_OVERHEAD_COST
    = 6,000

Rationale¶

Price only what every transaction always does: ECDSA recovery, warming sender and to, and two warm writes. That sums to 6,000 gas. Anything not universal should be metered separately.

Calldata remains metered per byte. No calldata allowance is folded into the base. This reinforces USDS as money and payments: a plain USDS transfer carries no calldata, executes no bytecode, and touches no contract storage slots. It only updates warm account nonce and balance, so it receives the discount.

When a value transfer creates a previously empty account, charge GAS_NEW_ACCOUNT = 25,000, identical to CALL account-creation pricing. CREATE is unchanged because its cost already includes account creation. Access-list pricing is unchanged.

This removes cross-subsidies, aligns charges with resources, and keeps costs equal across paths that perform the same state growth.

Do not fold any calldata allowance into the base. The envelope RLP (nonce, gas*, to, value, v,r,s) is not charged as calldata and remains unmetered per byte. Only tx.data bytes are metered at the existing calldata schedule, and access lists keep their per-entry costs. A vanilla USDS transfer executes no bytecode and touches no contract storage slots; it performs signature recovery, warms sender and to, and does two warm writes (nonce, balance). It therefore receives the base discount and zero calldata gas.

Bundling multiple transfers into a single contract call avoids repeated TX_BASE_COST, but those transfers then execute serially inside one DVM context. That blocks parallel execution. With a lower per-tx base cost, users have less incentive to bundle, so more transfers remain independent transactions. Independent transactions can be scheduled across threads, which improves parallelism at the client and execution-layer level.

Thus, reducing TX_BASE_COST not only corrects mispricing but also increases the share of transactions that are naturally parallelizable.

Why not charge full tx data as calldata?¶

Intrinsic coupling to signed fields. Pricing full-transaction bytes would make intrinsic gas depend on gas_limit and variable-length signature elements (yParity,r,s under RLP’s no-leading-zero rule), creating fixed-point estimation and incentives for signature-length selection as well as iteration unstablity as gas_limit depends on signiture which contains gas_limit.
Serialization neutrality. A fee rule keyed to RLP size couples costs to one encoding and weakens DIP-2718 type neutrality and future formats such as SSZ. Call data is treated as opaque bytes so doesn't have this encoding coupling.
Policy targeting and market floor. Calldata floors (DIP-7623/DIP-7976) price bytes in tx.data to bound EL payload and steer data to blobs; envelope bytes are control-plane. The TX_BASE_COST is the smallest tx size, and having a low encoding-agnostic intrinsic base keeps the minimum inclusion price strongly coupled to the basefee market. While also not under pricing to open a DoS vector.
Native settlement and unit-of-account Even though all transactions receive the same absolute gas reduction, it propritionatly improves pristine USDS transfers more. With an USDS transfer at ~6,000 gas versus a permissioned DRC-20/LST/stable coin transfer at ~48,000 gas yields an 8 x higher payment throughput per unit gas, which reinforces USDS as the settlement asset with greater money velocity potential and reduces reliance on contract-based money (which carries smart contract risk, operator risk, governance risk and censorship risks via operator).

Backwards Compatibility¶

This DIP is not backward compatible. It is a consensus gas repricing that must be activated at FORK_BLOCK.

Wallets, RPCs, gas estimators, and any logic that assumes a 21,000 base must update.

Effects on transactions per block¶

Median USDS transfer is 110 bytes of data, so at DIP-7934's 8MiB uncompressed execution block cap creating an effective max tx count by size of: 8,388,608 / 110 ≈ 76,260 tx

This leads to upper limits on gaslimit for raw tx count. Although more complex smart contracts can make use of a higher block gaslimit. However the market dynamics become more complicated when approaching the block cap and there will be distortions in gas pricing; when other approaches would need to be considered

e.g.

Raising network max block size (currently 10MiB both in EL devp2p and CL specs)
DIP-7782: Reduce Block Latency
DIP-7999: Unified Multidimensional Fee Market

Below TPS are additionally given for DIP-7782 at 6sec and 3sec blocks; as it is an easily considered effect.

Transaction base fee 6000¶

Binding point for TX_BASE_COST = 6000: 76,260 × 6,000 = 457,560,000 gas. Above ~457.6 M gas, the 8 MiB size cap dominates.

Block gaslimit	Old tx/bk (21k)	TPS @ 12s	New tx/bk (6k)	TPS @ 12s	TPS @ 6s	TPS @ 3s	TPS @ 1.5s
45M	2,143	179	7,500	625	1,250	2,500	5,000
60M	2,857	238	10,000	833	1,667	3,333	6,667
100M	4,762	397	16,667	1,389	2,778	5,556	11,111
200M	9,524	794	33,333	2,778	5,556	11,111	22,222
450M	21,429	1,786	75,000	6,250	12,500	25,000	50,000

Simple USDS transfer tx to an existing EOA: intrinsic drops from 21,000 to 6,000. (−71%)
USDS transfer tx that creates a new EOA: intrinsic becomes 6,000 + 25,000 = 31,000. (+48%)
Simple DRC-20 token transfer tx (e.g., USDS) drops from 63,200 to 48,200. (−24%)
Solady DRC-20 token transfer tx drops from 33,400 to 18,400. (-45%)
Average Uniswap v3 Swap tx drops from 184,523 to 169,523. (−8%)
Average Uniswap v3 Add Liquidity tx drops from 216,912 to 201912. (−7%)

Reference MGas/s for gaslimit / slot time¶

Block gaslimit	MGas/s @ 12s	MGas/s @ 6s	MGas/s @ 3s	MGas/s @ 1.5s
45M	3.75	7.50	15.00	30.00
60M	5.00	10.00	20.00	40.00
100M	8.33	16.67	33.33	66.67
200M	16.67	33.33	66.67	133.33
450M	37.50	75.00	150.00	300.00

Test Cases¶

While the benefits of reducing transactions' intrinsic cost are apparent, such a change should be applied if it imposes no negative externalities, or if such effects are negligible.

Tests should be created with blocks of just USDS transfers and tested on Perfnet across all EL clients to ensure the pricing is correct.

Add explicit test vectors:

value > 0 to an empty EOA (non-existent per DIP-161): intrinsic 31,000.
value > 0 to a precompile: intrinsic 6,000 (no surcharge).
value = 0 to an empty address: intrinsic 6,000 (no creation, no surcharge).
CREATE transaction: unchanged from prior rules.
DIP-7702 transaction with and without access list: intrinsic uses TX_BASE_COST plus unchanged access-list costs.
Block of txs calling minimal gas contract execution with maximal contract size addresses (so VM is activated for every tx)

Security Considerations¶

As this significantly increases the max tx per block this carries risk.

However this pricing should be the same as performing the component changes inside the transaction; and it factors in the additional costs from state growth which were not originally in the transaction base price.

Current gaslimit testing mostly uses a block with a single transaction; so this should not cause unexpected load compared to what is already being tested.

Copyright¶

Licensed under the Juro Restricted License Version 2. See https://juro.net/jrl for details.