Files
SnapOtter/apps/docs/guide
SnapOtterandGitHub 8574615ef5 docs(deployment): refresh hardware requirements from the 2.0 benchmark (#436)
Update the Hardware Requirements section with fresh six-machine benchmark data:
- Tiers corrected to the resource-sweep floor (2c/2G minimum; 512MB cannot start,
  1GB is single-file-only, batches need 2GB).
- Add the 64-bit-only architecture requirement (Pi 4/5 yes; 32-bit ARM and
  512MB boards no).
- Correct the AI-on-CPU viability (colorize/face-enhance are ~10s and usable, not
  "marginal to no") and add the AI RAM lever (~360MB idle without bundles vs
  ~2.6GB with all seven installed).
- Fix bogus GPU speedups (noise-removal/blur-faces are CPU-bound, ~1x, not
  13,400x/100x); real wins are upscale ~47x, face-enhance ~12x, transcribe ~4.5x,
  remove-bg ~4x; photo-restoration is CPU-bound even on a GPU.
- Call out video transcode as the one CPU-heavy tool; refresh concurrency numbers.

Claude-Session: https://claude.ai/code/session_01UvVCMNUBrgpghk8gye5gav
2026-07-05 20:19:24 +08:00
..