GigE Vision Multi-Camera Bandwidth: Sizing the Uplink Before the Cable
Quick answer / TL;DR: Short answer: add up the actual transmit rate of every camera, not the camera's maximum. Size the uplink above that total with headroom, use a 10G uplink for four or more gigabit cameras, and keep every copper run within 100 metres. Bandwidth exhaustion shows up as dropped frames and resends; a cable fault shows up as errors and a link that will not train.
Headroom is built: separator, 23 AWG copper, foil and drain.One GigE Vision camera on one cable is a solved problem. Four cameras on one switch is an arithmetic problem that most installations solve by accident and then diagnose by replacing cable.
Each camera can transmit close to 1 Gbps continuously. A switch with four camera ports and a single 1 Gbps uplink is oversubscribed four to one, and the buffer that hides this at low frame rates disappears at production rate.
This article gives the sizing arithmetic, the cable implications of a 10G uplink, and the symptoms that distinguish a bandwidth problem from a cable problem.
GigE Vision is maintained by the Automated Imaging Association and runs on IEEE 802.3 Ethernet. The camera streams frames at whatever rate the application requests, up to the link rate, and the protocol adds a packet resend mechanism for lost packets.
Resend is the mechanism that hides the problem. A camera whose packets are dropped by an oversubscribed uplink will resend, and the resends consume even more of the same bandwidth.
The Sizing Arithmetic
| Cameras | Aggregate at 1 Gbps each | 1 Gbps uplink | 10 Gbps uplink |
|---|
| 2 | 2.0 Gbps | 2:1 oversubscribed | Fine |
| 4 | 4.0 Gbps | 4:1 oversubscribed | Fine |
| 8 | 8.0 Gbps | 8:1 oversubscribed | Fine |
| 12 | 12.0 Gbps | Not viable | 1.2:1, tight |
The arithmetic uses actual transmit rate, which is usually below the link rate because the camera is configured to a frame rate and a region of interest smaller than the full sensor. A camera at 300 Mbps is common in practice, which changes the conclusion entirely.
The mistake is to size from the port speed because it is the number printed on the datasheet, and to ignore the configured rate that the application actually produces.
Where the Cable Comes In
Once the uplink goes to 10G, the cable becomes Cat6A at minimum, and the run length becomes a real constraint. Cat6 at 10G is limited to 55 metres in the worst case, which is often shorter than the cable route in a plant.
The camera-side cable stays at 1 Gbps and Cat6, because the camera port is gigabit. Only the uplink and the switch-to-switch runs need the higher category.
- Size the uplink from configured camera rates, not from port speeds
- Use a 10G uplink for four or more cameras at high frame rates
- Keep the camera-side runs Cat6 shielded, and upgrade only the uplink to Cat6A
- Enable jumbo frames end to end, because they reduce per-packet overhead at the same rate
- Monitor resend counters, which rise before frames are visibly dropped
Real-World Applications
A bottling inspection station runs four cameras at 400 Mbps each. The aggregate is 1.6 Gbps, so the switch has a 10G uplink and the camera ports stay gigabit. The initial design used a 1 Gbps uplink and dropped frames whenever the line ran above 60 percent speed.
A print inspection system runs eight line-scan cameras synchronised to the web. Each camera is configured to a narrow region of interest, so the aggregate is 5 Gbps and a single 10G uplink carries it with headroom.
A robot cell runs two cameras on a switch with a 1 Gbps uplink to the cell controller. Two cameras at 500 Mbps each fit, and the design deliberately caps the frame rate rather than upgrading the uplink, because the application does not need more.
A logistics hub runs 20 cameras across three switches with 10G uplinks to a fibre backbone. The copper never exceeds 80 metres, and the backbone carries the aggregate rather than any single switch uplink.
Standards and Verification
GigE Vision is specified by the Automated Imaging Association, with the transport on IEEE 802.3 and device control through GenICam. 10GBASE-T is specified in IEEE 802.3an, and copper channels follow ISO/IEC 11801 and TIA-568.
Verification should include a channel certification for each run, a negotiated-rate check at every port, and a sustained frame-rate test at production speed with the resend counters monitored. A test at idle proves nothing about bandwidth.
Where frames are being dropped, check the switch buffer and uplink utilisation before replacing cable. The cable fault signature is errors and a link that will not train; the bandwidth signature is resends with a clean link.
Related reading: GigE Vision cable — blogs/news/gige-vision-cable
Related reading: Cat6 vs Cat6A for 2.5G — blogs/news/cat6-vs-cat6a-2-5gbase-t
Related reading: Cat6A ethernet cable guide — blogs/news/cat6a-ethernet-cable-guide
A switch with a 1 Gbps uplink and four gigabit camera ports is not a gigabit switch in any useful sense. The uplink is the network, and everything else is a local port.
Frequently Asked Questions
How many GigE Vision cameras can share a 1 Gbps uplink?
It depends on the configured transmit rate. At 300 to 400 Mbps per camera, two fit comfortably and three is marginal. At full rate, one camera saturates the uplink.
Do I need 10G cable for the cameras?
No. The camera ports are gigabit, so Cat6 shielded is sufficient. Only the uplink and switch-to-switch runs need Cat6A for 10G.
Why do frames drop only at high line speed?
Because the aggregate camera rate crosses the uplink capacity. At low speed the switch buffer absorbs the burst; at production speed it does not.
How do I tell a bandwidth problem from a cable problem?
Bandwidth problems show rising resend counters with a clean link and no errors. Cable problems show CRC errors and a link that trains at a lower rate or not at all.
Sourcing cables for a project, a store, or a private-label line? Contact our B2B sales team for specifications, certification evidence, and volume terms – FOB and CIF options available. Request a quote or sample.
