vpbgen
Prebuilt vpbgen binaries attached (linux-amd64, macos-arm64, macos-amd64).
The library itself installs from source: v install --git https://github.com/we-be/protobuf.v
protobuf.v benchmark report
- date: 2026-08-10 (UTC)
- cpu: AMD EPYC 7763 64-Core Processor
- v: V 0.5.2 , harness compiled with -prod
- go: go1.26.5, google.golang.org/protobuf v1.36.11
Wire compatibility
- V and Go encode 1000 identical messages to byte-identical output
- Go decode of V bytes: IDENTICAL 154752 bytes, 1000 people
- V decode of Go bytes: IDENTICAL 154752 bytes, 1000 people
Throughput
| payload | op | V ns/op | V MB/s | Go ns/op | Go MB/s | V/Go time |
|---|---|---|---|---|---|---|
| 86 B | encode | 366 | 235.3 | 641 | 134.3 | 0.57 |
| 86 B | decode | 889 | 96.8 | 794 | 108.3 | 1.12 |
| 15.5 KB | encode | 89150 | 173.7 | 155263 | 99.7 | 0.57 |
| 15.5 KB | decode | 154606 | 100.1 | 196799 | 78.7 | 0.79 |
| 1.6 MB | encode | 13592042 | 116.2 | 16818352 | 93.9 | 0.81 |
| 1.6 MB | decode | 17753860 | 89.0 | 23071845 | 68.5 | 0.77 |
V/Go time: 1.00 = parity, below 1 = V faster. Schema: bench.proto (people with
nested phone messages, packed varint tags). Both harnesses build identical data
(splitmix64, seed 42) and time identically (iterations scaled to ~1s per cell).
Benchmarks ran on a shared CI runner — good for V-vs-Go ratios, noisy for absolute numbers.