What problem does it solve?
Measuring real video codec performance on NVIDIA Jetson devices requires authenticated official samples, correct environment setup, and disciplined statistics; ad-hoc benchmarking produces unreliable or misleading FPS numbers. This Skill runs controlled encode/decode throughput, preset comparison, and codec-worker capacity benchmarks, or produces clearly labeled documentation-based planning estimates when media is unavailable.
Core Features & Use Cases
- Authenticated throughput benchmarks: Runs dry-run then execute passes against official AppEncPerf/AppDecPerf or PyNvVideoCodec performance samples, reporting per-repetition FPS and MP/s with mean/min/max statistics.
- Preset and surface comparisons: Compares P4/P5 presets or native versus Python surfaces while holding every other control constant, without inferring quality differences.
- Capacity and planning estimates: Measures strictly increasing codec-worker concurrency, or produces clock-scaled and resolution-scaled SDK 13.0 documentation estimates labeled as theoretical bounds.
- Use Case: Ask how many 1080p H.264 camera streams a Jetson can encode; the Skill either runs a measured worker-capacity sweep on your media or returns a clearly labeled theoretical codec-stream bound from documented SDK tables.
Quick Start
Ask your agent to measure the H.264 encode throughput in FPS of a specific video file on this Jetson using the jetson-video-benchmark skill.
