The new Ethereum execution client processes 1.7 billion gas units per second on a disk footprint of just 240 gigabytes, delivering a 20-fold improvement in block persistence and removing the execution layer as a bottleneck for the network's scaling roadmap.
Paradigm released Reth 2.0 on Tuesday, an Ethereum execution client that processes 1.7 billion gas units per second — 70 per cent above the one-gigagas-per-second target the firm set two years ago and roughly 20 times faster than the previous version at persisting large blocks to disk.
The numbers are dry, but their implications are not. Ethereum's current mainnet processes around 15 million gas per block every 12 seconds, which works out to approximately 1.25 megagas per second. Reth 2.0 operates at more than 1,300 times that throughput. The gap between what the client can handle and what the network currently demands is so large that it effectively removes the execution layer as a bottleneck for any foreseeable scaling roadmap — including the high-throughput rollups and appchains that are supposed to absorb Ethereum's transaction growth over the next several years.
The performance gains come from three architectural changes. A sparse trie cache keeps an in-memory representation of the state trie alive across blocks, reducing state root calculation to between one and two milliseconds per mainnet block. A partial proofs system eliminates redundant disk lookups by fetching only unique trie portions, cutting proof requests per block by roughly half. And a tiered storage model splits hot and cold data between MDBX and a specialised append-only datastore backed by RocksDB indices, dropping the disk footprint to around 240 gigabytes for a full mainnet sync — or 170 gigabytes via snapshot download.
That storage number deserves attention. Running an Ethereum full node has historically required upwards of a terabyte of SSD space, a cost that discourages hobbyist operators and contributes to network centralisation. Reth 2.0's minimal mode fits on hardware that costs a fraction of what archival nodes demand. Paradigm benchmarked the client on an AMD EPYC server with 128 gigabytes of RAM and four NVMe SSDs — not consumer-grade equipment — but the disk savings alone make Reth competitive on more modest machines.
Georgios Konstantopoulos, Paradigm's CTO and general partner, led the development effort. The firm has been running Reth 2.0 in production on both Ethereum mainnet and Tempo, Paradigm's own high-performance chain designed to push the physical limits of network throughput. Tempo serves as a live stress test; if the client can handle gigagas-sized blocks there without choking, it can handle anything the Ethereum roadmap is likely to throw at it for years.
The release arrives at a moment when Ethereum's infrastructure layer is under more scrutiny than usual. The network's recent upgrades — from the Shapella withdrawal unlock to the Pectra changes rolling out this year — have focused primarily on the consensus and validator layers. Execution performance has received less attention, partly because existing clients like Geth and Nethermind have been adequate for current traffic levels. But "adequate" is a low bar when the stated ambition is to support millions of transactions per second across a sprawling ecosystem of rollups, each of which settles back to the base layer.
Reth's architecture is also designed for extensibility. The SDK model lets developers fork the client and build custom execution environments without maintaining a full codebase — an approach that has attracted teams building app-specific chains and high-frequency DeFi infrastructure. A 20-fold improvement in block persistence, from 8.4 seconds down to 400 milliseconds for gigagas-sized blocks, means those teams can design around tighter latency assumptions than were previously possible.
Whether Reth 2.0 shifts market share among Ethereum clients remains an open question. Geth still dominates with roughly 60 per cent of the network's execution layer nodes, and client diversity has been a persistent concern since the Merge. Reth's performance advantage is clear on paper, but migration costs and operational familiarity tend to keep node operators on their existing stack until something forces a change. The disk savings may prove a more compelling argument for switchers than raw throughput.
Paradigm framed the release as the beginning of its infrastructure ambitions, not the culmination. Saving a standard block now takes an average of 40 milliseconds; even a massive gigagas-sized block persists in just 400 milliseconds. A 1.7 gigagas execution client running on 240 gigabytes of disk — validated in production on two live networks — sets a new benchmark for what an Ethereum node can do.
