The lazy way to sync a database is to copy the whole thing. It works — until it's 200 MB and the connection is a phone. The GF Cloud DB ships only what changed: 500–1,450× smaller than the database at real scale, measured.
An offline-first data layer has one job when the connection returns: catch up fast, over whatever link exists. That makes payload size the number that matters — not throughput on a datacenter network. A delta that fits in a few kilobytes syncs over hotel Wi-Fi, a field tablet, or a folder on a shared drive. A full copy doesn't.
The numbers below are the delta-efficiency category of the GF Cloud DB Benchmark, measured 2026-07-03.
The delta bar is drawn at minimum visible width — at true scale it would be invisible next to the file.
| Delta compute for 10K changed rows8 MB peak RAM — runs on anything | 0.07 s |
| Full sync of 10K rows, end to end57,610 rows/s through the whole pipeline | 0.17 s |
| Payload at 100K-row scale53× smaller than the database file | 126 KB gz |
| Payload at 1M-row scale531× smaller than the 68 MB file | 128 KB gz |
| Payload at 3M-row scale1,448× smaller than the 207 MB file | 143 KB gz |
The full database file moves exactly once — when a brand-new node is seeded. After that, only deltas ever cross the wire, over HTTP or even a plain folder on disk.
See how the benchmark proves no change is ever lost — under conflicts, kills, and 1,500-op fuzz runs.
