GPU encoding for low-latency live streaming

Stream Bridge works with an NDI, OMT, desktop or DeckLink/SDI source available to it - cameras, mixers, graphics, presentations, multiviews and programme outputs are typical picks.

The source needs to stay active and reachable by the paired device. Discovery can take around 30 seconds, and local-sources-only filtering helps simplify busy networks.

Programme and additional audio can be routed according to the chosen source and viewer configuration, and up to 16 channels are available where the source supports it.

Viewer links can include additional audio sources with independent level controls. Audio Return and Group Audio can be switched on separately when the remote person needs to speak back.

Encoding can run on NVIDIA NVENC or the CPU. NVENC generally gives the lowest latency and CPU use; CPU mode is useful on systems without compatible hardware or where GPU capacity is already committed.

Default limits are four simultaneous GPU encodes and two CPU encodes per device. Use the meters to raise those cautiously and spot overload before it becomes dropped frames.

The default limit is four simultaneous GPU encodes per device - only raise it after checking GPU load and real-world frame delivery. Every viewer is a separate connection, and your bitrate applies to each - two viewers at 5Mbps needs roughly 10Mbps upload.

Bitrate changes apply live. A practical starting point is 3-5Mbps for most 1080p30 feeds and 5-10Mbps for high-motion or 60fps content, depending on the weakest connection in the path.

No VPN or inbound port forwarding is required. Stream Bridge uses ICE to find a workable path between the sending device and each receiver, preferring a direct connection.

If strict NAT or firewall rules prevent that, the session uses our TURN network instead. It improves reachability, but the extra hop can add latency, so wired internet and sensible bitrate headroom still matter for production-critical feeds.

Viewer links can use readable custom URLs, optional passwords and player customisation, so a client, director, operator and internal team can each get different controls from the same stream.

Treat device connection keys and viewer links as credentials. Keep connection keys private, password sensitive feeds, and remove links or bridges once an audience no longer needs access.