Transactions Per Second (TPS)

Transactions per second (TPS) is a key metric of the speed and scalability of payment and distributed ledger (DLT) networks. High TPS enables networks to support demanding real-world transactional loads without congestion.

Overview

Transactions per second (TPS) refers to the number of transactions a network can process each second. It is the most common measurement used to benchmark the performance and scalability of payment and DLT networks.

However, TPS is not the only way to measure transaction throughput. Other metrics include:

• Transactions per minute (TPM) - Used when transactions are complex and take longer than a second to process.

• Peak throughput - The maximum TPS achieved under ideal conditions. Gives insight into potential capacity.

• Sustained throughput - The average TPS maintained over time under real-world conditions. Measures actual performance.

• Latency - The time for a transaction to be processed and confirmed. Faster latency allows higher TPS.

• Concurrency - The number of transactions processed in parallel. Higher concurrency can increase TPS.

• Scalability - How throughput changes with increased load. Linear scalability means doubling nodes doubles TPS.

TPS can be measured by benchmarks and tests that simulate transaction loads. Common benchmarks for blockchains involve transferring assets between accounts or executing smart contract functions. The transactions sizes, complexity, and submission rates are parameterized to evaluate different scenarios.

When comparing TPS across networks, it is important to understand the assumptions and conditions of each test. Results can vary significantly based on factors like network size, geography, hardware, latency, consensus mechanism, transaction types, and concurrency. As with any benchmark, real-world performance may differ.

TPS Limits of Early Blockchains

Bitcoin and Ethereum put every transaction into a single chain of blocks that every node validates, so throughput is capped by how much a block holds and how often one is produced.

Bitcoin produces a block about every ten minutes, and its protocol limit on block size holds sustained throughput to about 7 TPS, according to the Bitcoin Wiki. The same page puts Visa's average at about 2,000 TPS.

Ethereum produces blocks far more often, which raises throughput, but high demand still slows transactions and raises fees. Both designs favour decentralisation and security over throughput, the tradeoff the blockchain trilemma describes.

Radix TPS

Radix's throughput figures come from test networks, and each test ran different software. The highest came in August 2019 from Tempo, a consensus design Radix later replaced. RDX Works engineers replayed ten years of Bitcoin transaction history, with full signature validation, on servers spread across 17 countries, at about 700 US dollars of Google Cloud capacity per run. Their write-up states that the results do not necessarily apply to Cerberus, the design that succeeded Tempo. A first public run in June 2019 peaked at over 1m TPS.

The Hyperscale public test in January 2026 ran real cross-shard swaps across 128 shards on commodity hardware, and more than 590 nodes joined it. The report on the day gave peaks over 700k TPS; the closing report a month later gave peaks above 800k, and recorded that a run on 64 shards had sustained about half the throughput, the linear scaling the design aims for.

Radix mainnet runs neither design at those rates. It runs as a single shard group, and the braided cross-shard consensus in the Cerberus paper has not shipped in any implementation (Radix Research). hyperscale-rs, a community-built Rust implementation, is the leading candidate to bring sharded consensus to mainnet through the Xi'an upgrade.