RTMP Target (Multi scaler)

Introduction

The RTMP Target (Multi scaler) streams the composed scene to up to four RTMP / RTMPS destinations simultaneously, each with its own resolution and bitrate, from a single component. One scene in, four independently-encoded videos out — perfect for multi-destination delivery, adaptive-bitrate (ABR) ladders, or program+backup redundancy without standing up four separate target instances and four separate encoder paths.

Compared to running multiple RTMP Target components in parallel — which works, but encodes the same source once per target — the Multi scaler is more efficient: a single CUDA multi-scaler kernel produces all four resolutions in one launch on the GPU, and a single NVENC encoder produces all four video tracks. The result is dramatically lower GPU and CPU load when you need a fan-out delivery from one Composer scene.

Encoding pipeline

The component uses a tightly integrated GPU pipeline:

CudaMultiScaler — one kernel invocation produces up to four scaled copies of the rendered frame at the configured destination resolutions. Cheaper than four separate scale passes.
NVENC — a single hardware encoder instance handles all video tracks in parallel; each destination can carry its own bitrate, profile, and codec parameters.
AAC audio — one shared audio program is mixed into every destination, so all four streams stay in lock-step audio-wise.
Per-destination FLV / RTMP mux writers — independent network connections, each with its own automatic reconnect. A failed destination retries on its own without disturbing the others.

Using the Multi scaler requires an NVENC-capable NVIDIA GPU. Most modern NVIDIA cards qualify, but the maximum number of concurrent NVENC sessions, supported codecs, and resolution limits vary by model — consult NVIDIA's compatibility matrix when planning a deployment.

Common use cases

Multistreaming to several platforms at once — push the same broadcast simultaneously to YouTube Live + Twitch + Facebook Live + a custom origin without provisioning four target components.
Adaptive-bitrate (ABR) ladder — deliver 1080p / 720p / 540p / 360p rungs into a single CDN ingest from one Composer scene, ready for HLS / DASH packaging downstream.
Program + backup redundancy — push the program feed to redundant origins on different network paths; the per-destination reconnect logic keeps each path live independently.
Contribution + distribution from the same scene — send a high-bitrate contribution feed to the master facility and a lower-bitrate distribution feed straight to viewers in one component.
Resource-constrained machines — when CPU / GPU headroom is tight, the shared scale + encode passes free up budget that running four separate RTMP Target components would consume.

URL macros

Each destination URL supports three substitution macros, evaluated at start time:

@@RandomId() — unique per destination, useful when each rung in an ABR ladder needs its own stream key.
@@SharedRandomId() — identical across all four destinations, useful when each destination is a different origin but the stream identifier should match.
@@HostName() — the machine's hostname.

Resilience

Per-destination automatic reconnect retries each network connection independently — a single dropped destination retries on its own without interrupting the other three. Combined bitrate, queue depth, and reconnect statistics are surfaced as readable properties for dashboards and the HTTP API.