Native AMD AMF encoder - test build
Pre-release
Pre-release
⚠️ Unofficial test build — native AMD AMF encoder
This is a test build, not an official Vibepollo release. It exists so people with AMD cards can help validate the native AMD AMF encoder (PR #295, tracking issue #172) before it's merged.
What's new
A native AMD AMF encoder that talks to the AMF SDK directly instead of going through FFmpeg's *_amf wrappers:
- ~44–46% lower per-frame encode latency measured on a 7900 XT (1.7 ms vs 3.0 ms at 1080p60; 2.65 ms vs 4.9 ms at 1440p60)
- Reference-frame invalidation — recovers from packet loss without a full keyframe
- HDR + D3D11 zero-copy input
How to test
- Back up your existing Vibepollo/Sunshine config first if you have one.
- Install/run this build. It's wired native-first: the native encoder (
amdvce) is tried automatically and falls back to the FFmpeg path (amdvce_legacy) if it can't initialise — so it should just work. - To force a path:
encoder = amdvce(native) orencoder = amdvce_legacy(old FFmpeg path, for comparison). - New tuning knobs in the AMD encoder tab:
amd_ltr_frames(RFI depth) andamd_input_queue_size.
Please report
- Your GPU + driver version
- Whether the native path was selected — check the log for
Found H.264/HEVC/AV1 encoder: *_amf [amdvce] - Anything that breaks or regresses
…on PR #295 or issue #172. Especially keen on RDNA2 / older cards and APUs, which I couldn't test myself.
Encoder work is ported/adapted from foundation-sunshine.