Sui overcomes consensus bottlenecks with Mysticeti V2

Sui maintains a 0.24 second p50 finality time across the last 24 hours, which outperforms Solana’s 0.40 second p50 measurement. This speed follows the implementation of Mysticeti V2, which moved transaction validation into the consensus process and replaced the old Quorum Driver with a Transaction Driver. The new driver sends each transaction to one chosen validator instead of all of them. This change reduced latency by 35% in Asia and 25% in Europe. While Mysticeti allows for high speeds, the network typically processes between 5,000 and 20,000 transactions per second on mainnet under real conditions.

The Sui object-centric model differs from the account-based model used by Ethereum and Aptos. In Sui, every piece of data lives as an object with a unique ID, an owner, a version number, and a digest of the last transaction that touched it. Transactions involving only owned objects skip consensus through a fast path. If multiple transactions use different objects, the network processes them in parallel. In contrast, Aptos uses the Block-STM engine to execute transactions speculically and re-runs those that conflict. This optimistic concurrency control creates re-execution overhead that increases during heavy contention.

Comparison of parallel execution and storage

Sui Move removes global storage functions like move_to or borrow_global because contracts operate directly on passed-in objects. This architecture simplifies sharding because the system manages objects as independent units without requiring global state synchronization. Aptos Move remains closer to the original Diem design, using global storage indexed by account addresses. On Aptos, a Move module defines a resource type and each account holds an instance of that resource. This account-centric approach requires developers to use the acquires keyword, which Sui eliminates.

Sei V2 approaches parallelization through optimistic parallelization and the Twin-Turbo consensus mechanism. This upgrade allows Sei to process transactions in parallel by grouping similar transactions or processing non-conflicting transactions simultaneously. Sei also provides EVM compatibility, which allows developers to deploy Ethereum smart contracts without code changes. While Sui focuses on object-centricity, Sei V2 targets 20,000 TPS through its parallel order execution.

Throughput benchmarks and market reality

Sui reached a peak of 40,614,180 TPS during a Basecamp test using Sui tunnels, which are offchain payment and state channels that settle to the mainnet. CertiK analyzed this test, which beat the July 4 record of 6,086,766 TPS. These figures measure the capacity of the offchain layer rather than base-layer transactions landing in checkpoints.

Aptos maintains a higher market cap in real-world asset tokenization, with over $1.2 billion in RWA value on-chain. BlackRock deployed its BUIDL fund to Aptos in late 2024. Sui has higher developer counts and DEX volumes, processing $38.3 billion in DEX volume compared to $10.8 billion on Aptos. You should ignore the 40 million TPS headline when evaluating base-layer stability.

Sei V2 aims for 200,000 TPS through its Autobahn consensus layer. The upgrade enables the chain to support many users while offering cheaper per-transaction costs. Will the Autobahn implementation match the theoretical throughput targets during periods of extreme network congestion?