Understand **NDI bandwidth requirements** for **Full NDI, HX3, and HX2** workflows, and how to calculate network capacity for multi-camera broadcast studios.
Essential for preventing network congestion in IP video production. To plan your specific infrastructure, use our NDI network planner. Explore more broadcast engineering tools for IP video and media networking.
NDI (Network Device Interface) is a royalty-free standard developed by NewTek (now Vizrt) for software-defined video over standard IP networks. It enables cameras, graphics systems, replay machines, and vision mixers to share video streams over standard Ethernet without dedicated video cabling — making it foundational to modern AV-over-IP workflows in broadcast studios, live events, and corporate video production.
Unlike SMPTE 2110, which transports uncompressed video at multi-gigabit data rates requiring dedicated network infrastructure, NDI uses compression to enable video transport over standard 1 GbE networks. The choice of which NDI variant to use determines the tradeoff between bandwidth consumption, latency, and visual quality.
| NDI Variant | Codec | 1080p60 Bandwidth | Latency | Best For |
|---|---|---|---|---|
| Full NDI (HB) | SpeedHQ (intra-frame) | 135–150 Mbps | ~1 frame (16ms @ 60fps) | Production, switching, monitoring |
| NDI HX3 | H.264 / HEVC (short GOP) | 30–60 Mbps | ~1 frame (16ms @ 60fps) | PTZ cameras, 1G constrained networks |
| NDI HX2 | HEVC (long GOP) | 2–16 Mbps | 100–300 ms | Monitoring, recording, campus distribution |
Full NDI uses the SpeedHQ codec developed by FFmpeg — an intra-frame compression format where each video frame is independently compressed, similar in concept to Motion JPEG. Because every frame is self-contained with no temporal prediction, there is no decoder delay caused by I-frame interval waiting — the codec can produce a new frame immediately without waiting for a reference frame from a preceding I-frame. This is why Full NDI achieves near-zero glass-to-glass latency.
The cost of this intra-frame approach is high bandwidth. At 1080p60, a Full NDI stream typically requires 135–150 Mbps. At 4K, this increases to roughly 250 Mbps. If you are mixing NDI with other streaming protocols, use our Video Bitrate Calculator to compare Mbps for 1080p and 4K workflows. Up to 5–6 Full NDI streams can share a single 1 GbE switch uplink at a safe 80% load threshold.
NDI HX3 is specifically engineered to provide the near-zero latency of Full NDI while using significantly less bandwidth. By using H.264 or HEVC with an extremely short GOP (Group of Pictures), HX3 maintains 1-frame latency (approx 16ms at 60fps). A 1080p60 HX3 stream typically requires 30–60 Mbps — between 2× and 5× less than Full NDI with no perceptible lag.
This makes HX3 the ideal choice for modern PTZ cameras and 4K production on standard 1GbE networks where bandwidth is at a premium but latency must remain broadcast-compliant. It is well within EBU R37 tolerance and suitable for high-speed tracking compensation in Unreal Engine's LiveLink workflows.
NDI HX2 uses long-GOP HEVC, allowing reference frames to span many frames ahead, yielding extremely low bitrates of 2–16 Mbps. This makes HX2 suitable for distributing video to many destinations over low-bandwidth campus networks, or for recording proxy streams running in parallel with high-quality masters. However, the encoding and decoding latency of long-GOP HEVC is 100–300 ms — making HX2 completely unsuitable for live production switching, IEM monitoring, or any application where real-time matching of audio and video is important.
The fundamental rule of network planning for NDI is: never exceed 80% of your switch uplink capacity. The remaining 20% headroom absorbs burst traffic from protocol overhead, TCP retransmission, and background network activity. At 1 GbE, your usable bandwidth budget is 800 Mbps.
For larger deployments, use the NDI Network Planner to model mixed-codec topologies and get an exact bandwidth budget per switch uplink.
By default, NDI operates in unicast mode. Every receiver that subscribes to a source gets its own dedicated stream from the sender. If four workstations all pull the same 1080p60 Full NDI source at 200 Mbps, the sender transmits 800 Mbps of identical data — saturating a 1 GbE link completely with just four receivers.
Multicast solves this scaling problem. Rather than point-to-point, the sender transmits a single multicast stream to the network. Any receiver that has joined the multicast group receives the same packets without the sender needing to transmit them multiple times. In a correctly configured multicast network, the bandwidth consumed is fixed regardless of the number of receivers — one 200 Mbps stream serves 4 or 40 destinations equally.
NDI uses Multicast DNS (mDNS) for automatic source discovery on the local network. mDNS is a zero-configuration protocol — NDI sources advertise themselves and receivers discover them without any server infrastructure. However, mDNS is limited to the local subnet and does not cross router boundaries.
For multi-subnet environments (multiple VLANs, or NDI across WAN links), use NDI Access Manager — a centralised service that maintains a registry of NDI sources and makes them discoverable across subnets. Access Manager can be installed on any Windows or macOS machine and configured with remote source addresses.
NDI and SMPTE 2110 are complementary, not competing, technologies. SMPTE 2110 transports uncompressed video at exact studio quality (10-bit 4:2:2 at ~1.5 Gbps for 1080p50) with frame-precise synchronisation via PTP. It requires dedicated 10 GbE or 25 GbE infrastructure. NDI offers compressed, flexible routing across any standard Ethernet network with lower infrastructure cost and simpler configuration.
The choice depends on the use case: SMPTE 2110 for master control, production fabric, and live switching where uncompressed quality and precise synchronisation are non-negotiable. NDI for monitoring, remote contribution, screen sharing, and workflows where the slightly lower quality of a compressed codec is acceptable in exchange for network simplicity and cost.