SDI vs NDI: Cabling, IP Networks & Conversion | Callaba
More live video workflow notes and product updates.
SDI carries a production signal over a dedicated serial interface, commonly on coaxial cable. NDI carries media across an IP network and adds discovery, naming and software-routable sources. The useful choice is rarely “old cable or new network.” Decide where you need deterministic point-to-point boundaries, where you need flexible routing, and who will operate each failure domain.
SDI and NDI move the routing problem to different places
Callaba operates on the NDI side of this architecture
Callaba lets you select NDI interfaces and Discovery Servers, synchronize discovered devices, and publish supported SRT, RTMP or video-call sources as named NDI outputs through adapters. That is a real NDI workflow. Callaba does not currently document native SDI capture cards, SDI router control or SDI hardware output as part of the product contract.
| Requirement | Callaba path | External boundary |
|---|---|---|
| Find NDI sources | Configure interfaces and discovery, then synchronize devices | The network must route the required traffic |
| Publish a managed source as NDI | Create and start a named NDI adapter | The receiver must see and decode that source |
| Bring SDI into the workflow | Feed Callaba from a supported converted source | Use dedicated SDI/NDI gateway hardware or software |
Put the gateway on the architecture diagram and monitor both sides. “SDI camera into NDI” sounds like one step, but it hides a cable, a receiver, format handling, an encoder or packetizer, a network interface and a discoverable NDI sender. Each is a separate checkpoint.
SDI gives each signal a visible physical path
SMPTE ST 292-1 specifies a 1.5 Gb/s bit-serial interface over coaxial cable for defined HD formats. SMPTE ST 424 covers a 3 Gb/s serial digital interface. These figures describe interface rates and format families, not an automatic promise about cable reach, connector quality or the behaviour of every converter.
SDI troubleshooting starts with a bounded chain: transmitter, cable, patch or router, receiver and format. A dedicated link does not compete with office backups or multicast, but flexibility usually requires more ports, patching and hardware routing. Redundancy also means deliberately duplicating paths rather than trusting one cable because it feels deterministic.
NDI turns the production fabric into a network service
The official NDI bandwidth guidance recommends sizing from measured average traffic and describes 1080p60 high-bandwidth video around the 130–150 Mb/s range. Treat those figures as planning references, not fixed reservations. Source type, format, implementation and simultaneous flows change actual utilization.
With NDI, multiple software receivers can select a source by name. That is useful for switching, graphics, monitoring and remote production, but the shared fabric still needs capacity for both normal and failure states. Check uplinks, oversubscription, switch buffers, multicast policy, VLANs, interface selection and receiver fan-out.
Discovery is part of availability
On one subnet, discovery can feel invisible. Across routed, cloud or segmented networks it becomes an explicit service. NDI's official Discovery Server documentation describes registration and lookup without relying on mDNS between every participant. Record the configured server addresses, reachable interfaces, sender name and receiver view.
A discovered source is not proof of flowing video, and active packets are not proof the expected name will appear on a receiver. Monitor discovery and media independently. Name collisions, stale registration and a receiver on the wrong interface can imitate a codec or bandwidth failure.
Latency is a chain, not a protocol label
Measure glass-to-glass latency through camera processing, SDI conversion, NDI encoding, switching, receiving, frame synchronization, rendering and displays. A protocol comparison made from marketing numbers omits the devices that dominate the result. Test the exact production modes, including interlace or progressive conversion and any frame-rate change.
Also measure stability: dropped frames, late frames, jitter, audio alignment and recovery after a link or switch interruption. The fastest average path may be the wrong choice if its tail behaviour creates visible cuts during peak traffic.
Choose conversion boundaries on purpose
Keep SDI where a short, dedicated hardware path is easy to understand and replace. Convert to NDI where one source must reach several software tools or travel through a managed IP fabric. Convert back only when a downstream device truly requires SDI. Repeated SDI-to-NDI-to-SDI stages add format negotiation, latency and extra failure points.
For each gateway, state accepted and emitted formats, audio mapping, reference or clock behaviour, name ownership, failover, monitoring and configuration recovery. Keep a known-good local monitor on each side so operators can determine whether a complaint begins before or after conversion.
Example: every NDI preview freezes at the same time
Several receivers freeze during an overnight backup, while the SDI confidence monitor remains clean. The team first replaces an SDI cable because that is the visible connection near the camera. Instead, compare switch-interface counters, uplink utilization, packet loss and NDI receiver timestamps across the incident. The simultaneous receiver failure and healthy pre-conversion monitor point to the shared IP path. Move backup traffic off the production window, correct the oversubscribed uplink or isolate the media fabric, then replay the peak load. Verify stable frames, audio continuity and enough remaining headroom with every intended receiver active.
Use a handoff test for every boundary
- Identify the signal. Record video format, frame rate, audio channels and source name.
- Prove the input. Monitor the signal immediately before conversion.
- Observe the gateway. Capture link, process and interface state.
- Prove the network. Measure actual flow, loss, capacity and discovery from the receiver side.
- Prove the output. Watch decoded frames and listen through the final production device.
SDI vs NDI FAQ
Is NDI simply SDI over Ethernet?
No. NDI adds an IP media implementation, discovery and network routing. Its operating and failure model is different from a dedicated SDI serial link.
Is SDI always lower latency?
Not by label alone. Measure the complete camera-to-display chain because conversion, synchronization, processing and display buffering can dominate either design.
How much network capacity does NDI need?
Measure the selected format and implementation, multiply by simultaneous paths and fan-out, then preserve headroom for bursts and failure routing. Do not size only from one nominal stream figure.
Can Callaba ingest an SDI cable directly?
Native SDI hardware ingest is not currently documented. Use a suitable gateway to present the signal through a supported Callaba input, then operate the NDI workflow from that boundary.