When you spec an industrial switch, the gigabit-vs-10-gigabit question gets dumbed down to "the ports are ten times faster." The real dividing line is not the 1G camera ports on the front, but the uplink that runs from the access switch up to the NVR or the core. A gigabit switch's downlink ports are usually fine; it is the uplink that gets saturated first by dozens of camera streams piling up. Using the bitrate formula and the uplink bottleneck, you can work out exactly which sites must go 10G and which are just burning budget.
The difference is the uplink, not the downlink: a 24-port gigabit switch feeds all camera traffic through one uplink port — that is the throat.
Oversubscription is the metric that matters: 24×1G downlink to 1G uplink is 24:1; to 10G uplink it drops to 2.4:1.
A 10G switch only pays off where aggregated uplink exceeds ~940 Mbps — large 4K camera farms, machine-vision lines, multi-switch core convergence, and storage/vMotion backbones.
Take a typical network switch 24 port gigabit model. Its backplane is usually ≥48 Gbps (24×1G×2, full duplex), and its forwarding capacity is more than enough. The problem is convergence: every downlink port's traffic has to leave through a single uplink port. If that uplink is 1G, the theoretical ceiling is 1000 Mbps, and after frame gaps and protocol overhead the usable figure is about 940–950 Mbps. The moment total downlink bitrate crosses that line, the uplink becomes the whole device's bottleneck.
A lot of buyers assume "all-gigabit" means "never congested." All-gigabit only guarantees each downlink port runs 1G to the switch chip — it does not guarantee all 24 ports can simultaneously push 1G out to the world. Oversubscription is the real number: 24×1G downlink against a 1G uplink is 24:1; swap the uplink to 10G and it falls to 2.4:1. On a gigabit industrial switch, the uplink port rate deserves more attention than the downlink ports — the downlinks each feed one camera, the uplink carries the sum of the entire box.
Note: "All-gigabit" is a downlink promise, not an uplink promise. Always read the uplink port rate before you trust the label.
Cameras do not eat bandwidth by "megapixels" — they eat it by bitrate. Real-world H.265 numbers: 1080P (2MP) about 2–4 Mbps, 4MP about 4–8 Mbps, 4K (8MP) about 8–16 Mbps. Switch the codec to H.264 and the bitrate roughly doubles. Do not design from the vendor's peak label; plan at a 70% water level for stability.
The formula is simple: total uplink bitrate = number of cameras × per-stream bitrate. Example: a workshop with 48 cameras at 4MP, H.265, 6 Mbps — that is 288 Mbps, only about 30% of a 1G uplink, comfortably within budget. Push it to 120 cameras at 4K, H.265, 12 Mbps — 1440 Mbps — and you have already blown past 1G, so this is where you need 10G.
| Scenario | Camera scale | Per-stream (H.265) | Total bitrate | 1G uplink (~940 Mbps usable) | Uplink advice |
|---|---|---|---|---|---|
| Small store | 16× 1080P | 3 Mbps | 48 Mbps | Plenty of headroom | Gigabit |
| Mid factory | 48× 4MP | 6 Mbps | 288 Mbps | ~30% | Gigabit |
| Multi-building campus | 100× 4MP | 6 Mbps | 600 Mbps | ~64%, critical | Gigabit |
| Scenario | Camera scale | Per-stream (H.265) | Total bitrate | 1G uplink (~940 Mbps usable) | Uplink advice |
|---|---|---|---|---|---|
| Plaza / station | 120× 4K | 12 Mbps | 1440 Mbps | Saturated | 10G |
| Machine-vision line | 20× industrial cameras | 200 Mbps | 4000 Mbps | Saturated | 10G (link bundle) |
Large HD camera aggregation. 100+ 4K or dense 4MP cameras push combined bitrate past 940 Mbps; a 1G uplink drops packets and stutters the recordings.
Machine vision / industrial cameras. An AOI inspection camera on the line can run 100–400 Mbps on its own; a dozen of them break a gigabit, and a normal gigabit uplink cannot hold — you need 10G to aggregate them.
Multiple access switches converging to the core. Each access switch carries its own 1G uplink; ten of them is 10 Gbps into the core. If the core's downlink ports are 1G, it jams anyway — the core side needs 10G or even 40G.
Storage and virtualization backbone. NVR clusters, dual-active iSCSI storage, and VM live migration (vMotion) can each fill 1G on a single flow. Unlike camera traffic, this is "few connections, huge bandwidth," and a 1G uplink is an obvious short board.
Tip: Camera traffic is many small flows; storage and vMotion traffic is few huge flows. Both can saturate a 1G uplink, but for opposite reasons — size the uplink for the worst-case aggregate either way.
A 10G switch does not mean every port is 10G. The common build is 24×1G + 4×10G SFP+. The 10G only lives on those few uplink ports; camera downlinks stay 1G, which is fine because the bottleneck was always the uplink. Count the 10G SFP port count and their location before you buy.
A 1G uplink with a 10G module gets you nothing. Module rate must match at both ends. Plug an SFP (1G) into an SFP+ (10G) port and it still negotiates 1G. For true 10G, the switch uplink port, the module, and the peer port must all be 10G.
Distance matters too: 10GBase-SR multimode typically reaches ≤300m; crossing buildings or campuses needs single-mode 10GBase-LR (10km and up). An industrial DIN-rail PoE switch also has to clear -40~75°C wide temperature, DIN-rail mounting, and ring protocols (ERPS/RSTP) — do not shop on port speed alone.
Warning: A 1G SFP in a 10G SFP+ cage quietly runs at 1G. If the link still shows 1G after you "upgraded" the module, the switch uplink port or the peer SFP port is the limit — check all three ends.
Flip it around — these cases are fully served by a gigabit uplink, and 10G is wasted money:
A single site ≤50 cameras, mostly 1080P/4MP, total bitrate under 400 Mbps year-round;
Cameras scattered so each switch carries only a few streams and the uplink sits idle;
A pure PLC, instrument, and access-control industrial ethernet network that generates almost no video traffic.
The budget you save is better spent on PoE power and wide-temperature grade — many projects get blocked by total PoE power before they ever hit an uplink bottleneck. When selecting, weigh PoE budget and wide temp on an industrial PoE switch first; agonizing over uplink speed is usually less practical.
Note: This is a different question from a network switch vs router. A switch moves frames inside one LAN; a router moves traffic between networks. Picking gigabit vs 10G is about LAN uplink capacity, not about routing.
Why plan a 1G uplink at 940 Mbps instead of the full 1000?
Ethernet carries frame gaps, preamble, and TCP/IP header overhead, so usable throughput is about 94–95% of line rate. Designing at full bandwidth causes frequent drops; engineers leave a 20–30% margin.
A camera is rated 4 Mbps — will it really exceed that?
Yes. Dynamic scenes, bitrate fluctuation, audio overlay, and ONVIF signaling all push it up; H.264 runs about double H.265. Reserve 1.3–1.5× the rated value.
Is a 24-port gigabit switch's uplink 1G or 10G?
It depends on the model. Many ship a 1G uplink; others add 1–4 SFP+ (10G) uplink ports. Read the uplink port line on the datasheet — that, not the "24×1G" front, decides whether you can grow to 10G later.
A gigabit and a 10-gigabit design share the same downlink job; the choice lives on the uplink. Size the uplink from the bitrate formula, watch oversubscription (24:1 vs 2.4:1), and only pay for 10G where aggregated streams clear ~940 Mbps — large 4K farms, machine vision, multi-switch cores, and storage/vMotion. For everything else, a gigabit uplink with strong PoE and wide temp beats an expensive 10G port you will never fill. The best ethernet switch for your site is the one whose uplink matches the load, not the one with the biggest number on the box.
Customer Manager at Shenzhen Rayin Technology Co., Ltd.
