FFmpeg-based RTMP streaming container with GPU acceleration support for the MarchProxy Unified NLB Architecture.

• RTMP Protocol Support: Full RTMP server implementation
• FFmpeg Transcoding: Industry-standard video transcoding
• CPU Encoding: x264 (H.264) and x265 (H.265) software encoders
• GPU Acceleration:
  • NVIDIA NVENC (H.264/H.265)
  • AMD AMF (H.264/H.265)
• Adaptive Streaming:
  • HLS (HTTP Live Streaming)
  • DASH (Dynamic Adaptive Streaming over HTTP)
• Multi-Bitrate: Automatic adaptive bitrate ladder (1080p, 720p, 480p, 360p)
• gRPC Integration: ModuleService implementation for NLB communication

RTMP Input → RTMP Server → FFmpeg Transcoder → HLS/DASH Output
                ↓                  ↓
           Session Mgmt      GPU Detection
                ↓                  ↓
           gRPC Server     Encoder Selection

Three variants available:

1. CPU: Software encoding (x264/x265)

   docker build -f Dockerfile -t marchproxy-rtmp:cpu .

2. NVIDIA: Hardware encoding (NVENC)

   docker build -f Dockerfile.nvidia -t marchproxy-rtmp:nvidia .

3. AMD: Hardware encoding (AMF)

   docker build -f Dockerfile.amd -t marchproxy-rtmp:amd .

docker run -d \
  -p 1935:1935 \
  -p 50053:50053 \
  -v /path/to/streams:/var/lib/marchproxy/streams \
  -e RTMP_ENCODER=x264 \
  marchproxy-rtmp:cpu

docker run -d \
  --runtime=nvidia \
  --gpus all \
  -p 1935:1935 \
  -p 50053:50053 \
  -v /path/to/streams:/var/lib/marchproxy/streams \
  -e RTMP_ENCODER=nvenc_h264 \
  -e NVIDIA_VISIBLE_DEVICES=all \
  -e NVIDIA_DRIVER_CAPABILITIES=compute,video,utility \
  marchproxy-rtmp:nvidia

docker run -d \
  --device=/dev/kfd \
  --device=/dev/dri \
  --group-add video \
  --group-add render \
  -p 1935:1935 \
  -p 50053:50053 \
  -v /path/to/streams:/var/lib/marchproxy/streams \
  -e RTMP_ENCODER=amf_h264 \
  marchproxy-rtmp:amd

Create /etc/marchproxy/rtmp.yaml:

host: 0.0.0.0
port: 1935
grpc-port: 50053
log-level: info

encoder: auto
preset: medium

output-dir: /var/lib/marchproxy/streams
enable-hls: true
enable-dash: true
segment-duration: 6

max-bitrate: 10
max-streams: 100
max-resolution: 1080

encoder-params:
  tune: zerolatency
  profile: high

The container automatically detects available GPUs and selects the best encoder:

1. NVIDIA GPU detected → nvenc_h264
2. AMD GPU detected → amf_h264
3. No GPU → x264 (CPU)

Override with RTMP_ENCODER environment variable.

1. Settings → Stream

   • Service: Custom
   • Server: rtmp://your-server:1935/live
   • Stream Key: your_stream_key

2. Settings → Output

   • Output Mode: Advanced
   • Encoder: x264 (will be transcoded by container)
   • Bitrate: 6000 Kbps

ffmpeg -re -i input.mp4 \
  -c:v libx264 -preset veryfast -b:v 6000k \
  -c:a aac -b:a 192k \
  -f flv rtmp://your-server:1935/live/your_stream_key

http://your-server/streams/your_stream_key/master.m3u8

http://your-server/streams/your_stream_key/dash/manifest.mpd

• Single stream: ~2-3 CPU cores
• 1080p30: ~50-70% CPU on modern CPU
• Latency: 6-12 seconds

• Single stream: ~5-10% GPU, <5% CPU
• Can handle 10+ concurrent 1080p streams on RTX 3060
• Latency: 6-12 seconds

• Single stream: ~5-10% GPU, <5% CPU
• Can handle 8+ concurrent 1080p streams on RX 6600
• Latency: 6-12 seconds

Default ladder (configurable):

Implements ModuleService for NLB integration:

• GetStatus() - Module status and stats
• GetRoutes() - Active streams
• GetMetrics() - Transcoding metrics
• HealthCheck() - Health status

# CPU version
go build -o rtmp-proxy ./cmd/rtmp

# Run
./rtmp-proxy \
  --host 0.0.0.0 \
  --port 1935 \
  --grpc-port 50053 \
  --encoder auto \
  --preset medium

Limited AGPL3 - See LICENSE file

• Documentation: https://docs.marchproxy.com
• Issues: https://github.com/penguintech/marchproxy/issues
• Website: https://www.penguintech.io