⚡
Hydra Head — Live Demo
A real Hydra Head between two nodes (Alice & Bob) is open on a local Cardano devnet. Pre-built transactions are fired all at once — TPS measures pure head throughput, not CLI overhead.
Head Status
...
Total L2 Txs
0
Alice
0 ADA
Bob
0 ADA
UTxOs
0
Best TPS
---
Fanout → K single-output txs, batched in snapshots. Peak ~180 TPS at 20–30 txs.
30
Realtime throughput (2-s windows)
Send a burst to see live throughput
Comparison
Cardano L1 (typical)~6 TPS
actual mainnet average
This Hydra Head---
single head, 2 nodes, this server
Live TxValid stream0 recent
How this works
1Pre-build (offline)
Parallel mode: fanout sender's UTxO into K chunks, then build K independent transfers. Chain mode: consolidate + build N chained txs. cardano-cli overhead — NOT counted in TPS.
2Fire (measured)
All pre-built transactions are submitted at once via WebSocket NewTx commands. The clock starts here.
3Validate
The head multi-signs each tx off-chain. Parallel mode fits them into one snapshot → much higher TPS. Clock stops when all are confirmed.
Why this is fair: TPS measures only the Hydra head's actual transaction processing speed. On L1, transactions wait 20s for a block. In Hydra, they confirm in milliseconds with zero fees. This 2-node head on a shared VPS reliably hits 180–200 TPS (small bursts of 20–30 txs) — that's 30–33× Cardano L1. The limit here is hydra-node's single-threaded validation pipeline, not CPU or bandwidth. Scaling further requires running multiple independent heads in parallel — exactly how Hydra is designed to reach the often-cited 1 M TPS aggregate.
GitHub | hydra.family | hydra-node 1.3.0, Conway era, magic 42