Splicer

Got questions or doubts? Ping me on Discord

Splicer · speed

Measured, not promised.

Every release is timed against the one before it on the same machine, and a step back of more than 5% holds it until Gamer Dad signs off. Release builds, only the timed command running under the GPU queue's bench class, two or three runs each with the median kept, on his own 2 h 07 min OBS recording (H.264 1920x804 with PCM sound in MKV, and its two-track copy). The charts are drawn by pito-compare, PITO MD's own chart tool.

  • 72 ms from opening a two-hour recording to its first frame on the desk
  • 24.6 s to hear all two hours and seven minutes on your own graphics card
  • 0.61 s to the first clips of a 122 GB recording, in 806 MB of memory
  • 4.1 ms a desk frame while an export runs beside it
Bar chart, opening a 2-hour recording with Splicer 1.0.6, lower is better: the desk's launch to its first 4K frame 233 ms, the recording's first frame on the desk 72 ms, and on the command line 59 ms.
Open a two-hour recording and its first frame is on the desk in 72 ms.
Bar chart: a recording of 2 hours 7 minutes against 24.6 seconds from opening it to its last clip heard, Splicer 1.0.6.
Two hours and seven minutes, heard on your own graphics card in 24.6 seconds.
Paired bars, lower is better: a 16:9 export with 300 cuts took 104.7 to 104.9 s in 1.0.4 against 126 s in 1.0.0-rc.1; a 1000-cut smart cut took 197 to 224 s in 1.0.1 against 319 s before it.
The same 300 cuts went from 126 s to under 105 s.
Paired bars, lower is better: the desk's frame p50 while a card export runs, 4.1 to 4.3 ms in 1.0.6 against 11.7 to 12.0 ms in 1.0.5; the worst frame while the window is resized, 5.8 ms against 386 ms.
The desk keeps drawing while an export runs and while you resize the window.

Release by release

What Measured Release
Open the two-hour recording: the first clips ready to keep or drop, on the command line 1.02 s 1.0.0-rc.1
Open the two-hour recording: its first frame, on the command line 59 ms 1.0.6
Open the two-hour recording: every clip heard, on the command line 24.6 s 1.0.6, against 24.8 s in 1.0.4
The desk while the two-hour recording opens: its own work per frame, p99 2.6 to 5.3 ms 1.2.0, four runs
A screenful of 27 rail pictures: a 10-minute MP4, then the two-hour MKV 24.4 ms, 60.8 ms 1.0.7
Open a 122 GB recording: first clips, memory peak 0.61 s, 806 MB 1.0.0-rc.1
An AV1 video, 720p, two minutes: first frame 49 to 52 ms 1.0.5, its own codecs, against 87 to 719 ms in 1.0.4
A Matroska file on a NAS over SMB: open to first frame 27 to 79 ms 1.0.1, against no picture for 30.6 to 34 s in 1.0.0-rc.15
Silences removed, the first 20 minutes of the two-hour file: smart cut 24.5 s 1.0.4, against 32.0 to 33.8 s in 1.0.2
1000 cuts of the two-hour, two-track file: smart cut 197 to 224 s 1.0.1, against 319 s before it
300 cuts of the same file, 16:9 export 104.7 to 104.9 s 1.0.4, against 126 s in 1.0.0-rc.1
A desk frame while a card export runs beside it, p50 4.1 to 4.3 ms 1.0.6, against 11.7 to 12.0 ms in 1.0.5

Hearing runs on your own graphics card: the two-hour recording is heard on an AMD Radeon RX 9060 XT. Every number comes from Splicer's speed baseline, which each release candidate is measured against.