Base

The main Ethereum upgrade since the Merge is on its way: Dencun hard fork bringing EIP-4844 (Proto-Danksharding) will be there on March 13th aiming at making L2 transactions cheaper.

What impact will it have? Which L2 will win the fee war?

2/16

Methodology: 

1) Consists of evaluating transaction costs across various L2s (Starknet, ZkSync, Optimism, Scroll, and Arbitrum) comparing average DA to execution costs in %
2) Then project those costs into 3 scenarios of new DA cost post 4844 (pessimistic, moderate & DA=0)

3/16

- @Starknet

The attached graph reveals that DA costs currently represent about 90% of total costs, with execution making up the remaining 10%.

The green and red part shows the average DA cost on L1 (pre EIP-4844) and the purple strip represents the execution cost in %.

5/16

- @ZkSync

The attached graph indicates average DA costs at around 65%, with execution costs at 35%. -despite some meaningful peaks-.

6/16

- @Scroll

Scroll exp (scrollscan.com) suggests DA costs average 62-63% for swap/multicall & 90% for transfers. We approximate an average DA cost of 70%.
Without weighting the weight of each type of tx, we'll consider an average DA cost of 70% (vs 30% for exec)

7/16

- @Optimism (OP Stack)

According to their March 3rd article, OP claims a x70 decrease in tx fees post 4844, based on their "Superchain Savings Estimator" (SSE).

However, the exact method used justifying it isn't detailed.
We gotta rely on this data for now & wait for 4844

8/16

- @arbitrum

Unfortunately, couldn't get data to compare the weight of the DA to the execution cost so we won't be able to include them in this analysis.

Offchain Labs published a great article about it https://medium.com/offchainlabs/eip-4844-what-does-it-mean-for-l2-users-5e86ebc4c028

9/16

Ok now that we have the average cost of DA vs the execution cost for the L2s let's consider three scenarios post EIP-4844 to measure the impact on the price for each L2s

1) Pessimistic scenario
2) Moderate scenario
3) Extreme scenario: DA = 0

10/16

1) If the DA cost drops to 50%, the overall tx fee would decrease by:

- Starknet: x1.8 (with overall cost of 55% of current levels)
- Scroll: x1.5 (with overall cost of 65% of current levels)
- Zksync: x1.5 (with overall cost of 67.5% of current levels)
- Arbitrum: N/A

11/16

2) If the DA cost drops to 90%, the overall tx fee would decrease by:

- Starknet: x5 (with overall cost of 19% of current levels)
- Scroll: x2.7 (with overall cost of 37% of current levels)
- Zksync: x2.4 (with overall cost of 41.5% of current levels)
- Arbitrum: N/A

12/16

3) If the DA cost drops to 0, the overall tx fee would decrease by:

- Starknet: x10 (with overall cost of 10% of current levels)
- Scroll: x3.3 (with overall cost of 30% of current levels)
- Zksync: x2.9 (with overall cost of 35% of current levels)
- Arbitrum: N/A

13/16

Conclusion

🏆 If Optimism's projections hold, they'd win the 4844 fee war

BUT details are scarce, we must rely on them

They claim a x70 fee cut on Base, Zora, Mode
Meaning the DA cost of the OP Stack would be 98.5% of a tx fee in the scenario DA drops to 0

Let's see!

14/16

Conclusion (2):

The rest of the ranking would be as follows:
- Starknet 🥇
- Scroll 🥈
- Zksync 🥉
- Arbitrum: N/A

Now let's wait until March 13th when EIP-4844 is out ✨🫡

Feel free to share your thoughts in comments ⬇️

15/16

Sources/references:

- Nethermind great paper https://medium.com/nethermind-eth/starknet-and-zksync-a-comparative-analysis-d4648786256b#19b8
- OP 03/03/24 article: https://oplabs.notion.site/Superchain-Savings-Estimator-100e62675cd849488b6c275a2e09ca0c

16/16

Sources/references (2):

- https://github.com/ethereum/EIPs/blob/master/EIPS/eip-4844.md
- Scroll tx examples swap: https://scrollscan.com/tx/0x75e335b6c08f0a4a8dce0a8c6d288cf9296a580e4f18b4dd57fd823d7fa1d635

16bis/16

That means Base reduction cost post #EIP4844 would be x70 the current price if Optimism projections's hold

This analysis includes transactions of all types (transfer, swap, multicall...)

(Collecting the exact same & consistent source of data across each L2 was very challenging. Thus we will consider the data we had access to for each L2s.)

Now, let's get started 🏁

4/16