Transactions Per Second (TPS)

Transactions per second (TPS) is a measure of throughput: the number of transactions a network processes in one second. It is used for card networks, bank payment systems and, most often, blockchains, where it has become a common way to compare one network with another and with established payment systems.

TPS is one of several performance measures. Others are latency (how long one payment takes to be included) and finality (how long until it cannot be reversed). A network can have high TPS but slow finality, or the reverse. TPS figures are also among the most frequently exaggerated numbers in crypto marketing, so it matters whether a figure is a measured average, a peak, a test result or a theoretical maximum.

Think of a motorway toll plaza. TPS is how many cars pass the barriers each second. It depends on how many booths are open and how quickly each one works. It says nothing about how long the drive took or whether the toll was actually paid.

A sign saying "this plaza can handle 500 cars a second" is also different from the number of cars that actually passed through yesterday. Some operators quote the best number they could reach in a test with empty roads and the simplest cars. When comparing networks, it pays to ask which of those numbers is being shown.

For a blockchain that produces blocks at regular intervals, average throughput is:

• $N$ is the number of transactions included in a block (or the average across many blocks).
• $t$ is the average time between blocks, in seconds.

Example: the Bitcoin blockchain targets one block about every 10 minutes (600 seconds). If a block carries 3,000 transactions, throughput is $3{,}000 \div 600 = 5$ transactions per second. Bitcoin's throughput is limited by its block size rules and by how large each transaction is, which is why estimates of its maximum vary.

On networks that meter computation, capacity depends on the block's gas budget. The theoretical maximum for the simplest transfers is:

• $G_{\text{block}}$ is the gas limit per block.
• $g_{\text{tx}}$ is the gas used by one transaction (21,000 for a plain ether transfer).
• $t$ is the block time in seconds (12 on Ethereum since the Merge in September 2022).

Example: with a 30 million gas limit, the level Ethereum used for several years after 2021, $30{,}000{,}000 \div (21{,}000 \times 12) \approx 119$ TPS. That is an upper bound for the simplest possible transactions; token swaps and other smart contract calls use far more gas, so real throughput on the Ethereum blockchain is lower.

The most quoted benchmark is Visa. Visa's fact sheets have described VisaNet as capable of handling more than 65,000 transaction messages per second, a capacity figure Visa dated to August 2017. That is a stated capacity, not the average load, which is much lower. It is also a count of messages through a network that authorises payments and settles later through banks, so it is not directly comparable with a blockchain that settles each transaction on its own ledger. Wholesale systems such as real-time gross settlement (RTGS) systems process far fewer, much larger payments, so TPS says little about their importance.

• Theoretical versus observed. Many projects publish lab results or maximums. Ask for the measured average over a stated period.
• What counts as a transaction. On some networks, consensus messages are counted as transactions. On Solana (SOL), for example, validator vote transactions make up a large share of the total, so analysts often quote "non-vote" TPS separately.
• Transaction size. A plain transfer is much smaller than a multi-step DeFi trade, so a TPS figure based on the simplest transfers overstates capacity for real use.
• Decentralisation trade-offs. Higher throughput often requires stronger hardware or fewer validators. A chain run by a handful of data-centre machines can post high numbers more easily than one with many home-run full nodes.
• Finality. Fast inclusion does not mean final settlement. On chains with probabilistic finality, users may wait for several confirmations.

Approaches to scaling blockchains fall into a few groups. Base layers can raise block limits or shorten block times, at a cost in node requirements. They can split work into shards. Or activity can move off the base layer to Layer 2 rollups and channel networks such as the Lightning Network, where many payments happen off-chain and only the results are settled. When demand exceeds capacity, transactions queue in the mempool and users bid up transaction fees, so fees are often a better sign of congestion than TPS alone.

• ethereum.org: Scaling
• Visa fact sheet (July 2019)