Solana activated Alpenglow on its mainnet on Oct. 21, replacing the TowerBFT consensus system that had governed the network since its launch. The upgrade is the network’s first major consensus replacement and changes how validators vote and certify blocks without changing Solana’s transaction execution, programs, fees or underlying virtual machine.
The live phase is Votor, which moves validator votes off the blockchain and has validators exchange votes directly before combining them into certificates. A block can become final in one voting round when roughly 80% of stake supports it, or in two rounds when participation is lower. The system replaces TowerBFT’s process, in which validator votes were recorded as transactions and finality depended on votes accumulating across multiple slots.
Alpenglow was designed around a finality target of about 150 milliseconds, compared with the roughly 12. 8-second figure associated with the previous system. Independent mainnet measurements recorded a more cautious result after launch: a 404-millisecond median over 24 hours, with the 90th percentile at 579 milliseconds and the 99th percentile at 765 milliseconds across more than 357,000 slots. Those measurements indicate a roughly 17-fold improvement from a pre-upgrade median near seven seconds, while the 150-millisecond figure remains a protocol target rather than an independently observed mainnet median.
The change also alters how network activity is counted. Validator vote transactions, which had represented about 62% of processed transactions in the week before activation, are no longer recorded in the same way, so raw transaction-per-second comparisons across the switch are not directly equivalent. Validators now pay a 0. 8-SOL admission-ticket charge per epoch instead of vote fees, and operators were required to use compatible Agave-based software and register the required validator credentials. Application developers generally do not need to migrate, although infrastructure that reads block and finality data must account for the new consensus behavior.
Alpenglow is being introduced in two phases. Votor handles voting and finality and is live; Rotor, the planned second phase, is intended to replace Turbine’s block-propagation system with a relay layer but has not been assigned a launch date. Before mainnet, the protocol went through testnet, devnet and repeated migration rehearsals, including a failed test-network migration that was later completed successfully. The remaining gap between the 150-millisecond target and public-node measurements reflects both network observation time and the fact that external nodes receive and process blocks before reporting finality.