I-IAL9000/shrinkerr v0.5.9

on GitHub

Added

• NVENC bit-depth choice in Settings → Encoding → NVENC card. New dropdown with three options:
  • 10-bit (main10 / p010le) — Default. Preserves the pre-v0.5.9 hardcoded behaviour. Best quality (reduces banding artifacts), slightly larger files on some sources, requires Pascal-or-newer NVIDIA (GTX 10xx, Quadro P-series, RTX). Excludes Maxwell silicon (GTX 9xx, GTX 750 Ti, Quadro M-series).
  • 8-bit (main / nv12) — Maxwell-compatible. Smaller files on most material, faster encode, fully sufficient for 8-bit source content. The right choice if your NVENC card is older than GTX 10xx, or if you've A/B-tested and prefer the size win.
  • Match source — Probes the source pix_fmt at job time and picks 10-bit only when source is already 10-bit (yuv420p10le, yuv420p12le, p010 surfaces), otherwise 8-bit. Avoids the 8→10 bit upconvert overhead for 8-bit sources without giving up quality on 10-bit ones. Resolution decision is logged per-job ([CONVERT] NVENC bit-depth auto: source pix_fmt='yuv420p' → encoding 8bit).
• Probe now captures video_pix_fmt alongside video_codec from ffprobe — driver for the auto-mode bit-depth resolution. Exposed on scanner.probe_file() return dict.

Fixed

• -threads N now emitted both pre-input AND post-encoder. Per the GitHub feature requester's observation: "ffmpeg will only apply it for that processing portion." -threads is a per-codec-context option, not a global. The v0.5.6 pre-input placement caps decoder threads, but software encoders (libx265 specifically) need the cap on the encoder side too — without it, the Parallel Jobs > 1 + FFmpeg Threads Per Job = 2 setting only constrained decode, while libx265 encode kept grabbing every available core. Now emitted at both boundaries. NVENC/QSV/VAAPI ignore the flag (GPU does the heavy lifting) so the redundant post-encoder copy is harmless on hardware paths and a real cap on libx265.

Behaviour notes

• The 10-bit default means existing v0.5.7/v0.5.8 users see no behaviour change unless they actively switch the new dropdown.
• Maxwell-era NVENC users (anyone whose NVENC capability detection works but who's been hitting profile/format errors on encode) should pick 8-bit to make NVENC usable. The capability detection (encoder_caps.py) can't yet distinguish Maxwell from Pascal+, so the dropdown defaults to 10-bit and Maxwell users have to flip it manually. Surfaced in the help text.
• "Match source" is the recommended setting for users with mixed libraries — 8-bit Blu-ray rips encode as 8-bit (smaller, faster), 10-bit anime / 4K HDR remuxes encode as 10-bit (preserves source precision). The decision is per-job, not per-session.
• The NVENC+NVDEC scale_cuda=format= filter (added in v0.5.8 to bridge the pix_fmt mismatch) now picks p010le or nv12 based on the resolved bit depth — keeps the HW decode pipeline correct in both modes.

Docker images

Pinned to this release:

docker pull ghcr.io/i-ial9000/shrinkerr:0.5.9         # portable (amd64 + arm64, CPU)
docker pull ghcr.io/i-ial9000/shrinkerr:0.5.9-nvenc   # x86_64 + NVIDIA GPU (NVENC)
docker pull ghcr.io/i-ial9000/shrinkerr:0.5.9-edge    # as above but ffmpeg master
docker pull ghcr.io/i-ial9000/shrinkerr:0.5.9-edge-nvenc

Floating tags (:latest, :edge, :nvenc, :edge-nvenc) track the most recent release of each lineage.