09 Open questions and benchmarks¶
Status: Open. These need practical tests; contributions with real numbers are welcome.
Questions¶
- What is the lowest achievable end-to-end latency with cloud GPU rendering?
- What is the best way to send live camera video into cloud-hosted Unreal Engine?
- Can green screen compositing be done entirely inside Unreal Engine for a live booth, without OBS?
- How reliably can Aximmetry work with a remote rendering pipeline?
- What GPU and network bandwidth does each architecture need?
- Is a local GPU workstation cheaper than cloud rendering over a multi-day event?
- Can a portable booth stay reliable on event Wi-Fi or a dedicated 5G link?
- Which AI matting model gives the best hair edges on wigs and lace at 1080p, and at what speed?
- How many frames per second does COLMAP need for a reliable solve of a 5-second handheld phone shot?
How to report a benchmark¶
Open an issue or pull request adding a row to benchmarks/results.csv (create it if missing) with:
| Field | Example |
|---|---|
| date | 2026-11-02 |
| question | 1 |
| setup | Booth v0, RTX 4070 laptop, 1080p50 capture card, OBS 31 |
| method | Stopwatch filmed through system, 20 samples |
| result | median 112 ms, p95 140 ms |
| notes | Window capture; NDI added 30 ms |
Give the method. A number without a method can't be compared.