How many hours of video fit a given Wh budget, by hardware & codec at a target VMAF.
ⓘ About this test (click to expand)
Transcode a source video and measure the server's wall-power draw during the encode.
Accepted: MP4, MOV, MKV, AVI, WebM, TS · Max 1GB.
Baseline measured 10s before each run · P110 + thermals at 1s intervals.
All GPU presets use the full VAAPI pipeline (hardware decode + encode + scale) — representative of live encoding workflows.
Rate control is ABR (constant bitrate target) across all 6 presets so CPU and GPU receive identical tasks.
Scope: device layer only — network, CDN, and client devices (CPE) excluded.
H.264
H.264 CPU
libx264 · ABR · 1080p
detailsSoftware encode across all 24 cores.
H.264 GPU
h264_nvenc · ABR · 1080p · full pipeline
detailsHardware decode + encode. Full GPU pipeline — representative of live encoding.
H.264 Both
CPU then GPU · same file
detailsSide-by-side energy + thermal report with analysis.
H.265
H.265 CPU
libx265 · ABR · 1080p
detailsSoftware HEVC encode.
H.265 GPU
hevc_nvenc · ABR · 1080p · full pipeline
detailsHardware decode + encode. Full GPU pipeline.
H.265 Both
CPU then GPU · same file
detailsSide-by-side H.265 CPU vs GPU comparison.
AV1
AV1 CPU
libsvtav1 · ABR · 1080p
detailsSVT-AV1 software encode.
AV1 GPU
av1_nvenc · ABR · 1080p · full pipeline
detailsHardware decode + AV1 encode on the GPU's dedicated AV1 engine.
Scored after the measurement window closes — never part of the energy figure, but it adds
minutes of wall time. Uncheck for a faster energy-only run. Single-preset runs never score.
Confidence flag
🟢 Repeatable≥95% confident the task draws above idle, with enough task polls to confirm. Reliable enough to cite.
🟡 Early insight≥80% confident above idle, or too few polls for green. Directional, needs a longer run.
🔴 Need more dataNot yet distinguishable from idle. Don't cite yet.
confidence = Φ(ΔW / SE), SE from this run's noise + the calibrated idle floor · methodology