Two cameras, one station, and measurements that refuse to agree: that is the classic IEEE 1588 misunderstanding. PTP puts every GigE Vision camera in the cell on one clock, so the timestamps written into the images match across cameras to sub-microsecond level. It does not make the cameras expose together. If you enable PTP, see matching timestamps on two free-running cameras and treat the frames as simultaneous, a two-view measurement can be off by an entire frame period. This page separates the two jobs: what PTP aligns, what still needs a trigger or a scheduled action command, the arithmetic that shows how far apart free-running frames really are, and a worked example from a moving line.
What IEEE 1588 PTP does to a camera clock
PTP is the Precision Time Protocol of the IEEE 1588 standard, and it ships as part of the GigE Vision 2.0 feature set. Cameras exchange timing messages over the same Ethernet cable that carries the images, measure the delay to every device, and each camera except the elected master steers its clock onto the master's timebase. The protocol targets nanosecond-class clock agreement; on a small camera network, designing around one microsecond of timestamp agreement is realistic.
The camera presents PTP as a transport-layer feature: you enable IEEE 1588, then read the IEEE 1588 status feature until it reports a synchronized state. Locking takes several seconds because each camera trims its clock in steps. From then on, the device timestamp attached to every frame — the same value chunk data can carry inside each image — lives on the shared timebase. That is the entire job: one clock for timestamps and for scheduled events.
The step-by-step set-up, the master and slave roles, and the post-lock timestamp check are covered in IEEE 1588 PTP camera synchronization for GigE cameras and PTP GigE camera synchronization: master, slave and timestamp check. On the plant network side, NTP and PTP solve different grades of the problem — milliseconds for logs, sub-microsecond for events — and NTP versus PTP for industrial time synchronization compares both methods.
What PTP does not do: it does not frame-sync
Cameras are time-synced, not frame-synced. A free-running camera starts each frame on its own schedule, and PTP hands every camera the same ruler, not the same schedule. After lock, two cameras can stamp their frames with comparable times while the frame starts sit wherever power-up left them. Cameras at the same frame rate keep whatever phase offset they powered up with; cameras at different frame rates sweep through every offset, worst case one full period.
At 60 fps the period is 1/60 s ≈ 0.0167 s = 16.7 ms, so the worst-case gap between two exposures is 16.7 ms no matter what the timestamps claim. That is the trap: PTP removes the clock error and leaves the schedule error alone, so a timestamp delta check passes while the images sit milliseconds apart.
Some camera lines add a feature that uses the shared PTP time to pull free-running frame starts slowly into phase. Treat it as a soft servo: it converges over seconds, a dropped frame can disturb it, and it is not an event you can schedule against.
Worked example: two 60 fps cameras on a 100 mm/s line
The inputs: two identical global-shutter GigE cameras free-running at 60 fps (a common inspection rate); conveyor speed 100 mm/s (light assembly); the inspection needs the two views to agree within 0.1 mm (the example's tolerance, substitute your own); PTP agreement designed at 1 µs; event precision 10 µs, because GigE Vision action commands are specified in the microsecond class. The protocol figures come from the GigE Vision material; the frame rate, speed and tolerance are the example's assumptions.
The rule: with no shared event, the worst-case offset between the two exposures is one frame period, and the shift you see between views is part speed × capture offset.
The arithmetic:
- Frame period: T = 1/60 s ≈ 0.0167 s = 16.7 ms.
- Free-running, worst case: shift = 100 mm/s × 16.7 ms = 100 mm/s × 0.0167 s = 1.67 mm, about 1.7 mm. That is 16.7 times the 0.1 mm budget, so the two-view measurement fails on its own.
- PTP enabled, still no event: the clock term drops to 1 µs, but the schedule term stays at up to 16.7 ms, so the worst case remains about 1.7 mm. PTP relabels the clock; it does not close the gap.
- Action command or wired trigger at 10 µs event precision: shift = 100 mm/s × 10 µs = 100 mm/s × 0.00001 s = 0.001 mm = 1 µm. Inside the 0.1 mm budget with two orders of magnitude to spare.
- Exposure check: a 100 µs exposure adds 100 mm/s × 0.0001 s = 0.01 mm = 10 µm of motion blur, small against the 0.1 mm budget.
Result: free-running fails by a factor of about 17; a scheduled or wired event passes with 100× margin.
The decision: put the tolerance against part speed × one frame period. If the tolerance exceeds 1.7 mm here (16.7 ms × 100 mm/s), PTP timestamps alone are enough — pair frames by time and skip the wiring. If the tolerance is tighter, the exposures need a shared event: action commands scheduled on PTP time, or a hardware trigger for the tightest skew. On this line the crossover sits between 1.7 mm and 1 µm; the two approaches differ by three orders of magnitude, which is why they are not interchangeable.
A station like this also needs two clean network ports. The MV-GE2002P dual-port PoE NIC covers it from one slot, one cable per camera.
- Interface
- GigE
- Ports
- 2 ports
Three ways to make the exposures share an instant
1. A shared hardware trigger. One pulse source wired to every camera's trigger input starts all exposures within microsecond-class skew. It is the most deterministic option and the only common event in a mixed cell that also holds USB3 or Camera Link cameras. The frame-rate ceiling on a trigger is exposure plus readout time — maximum frame rate on a hardware trigger: exposure plus readout time works through the sum. Two related failure modes have their own pages: several GigE cameras firing at once can saturate the link (multiple GigE cameras, same trigger: why frames arrive incomplete), and a shared strobe is best fired by one camera (multiple cameras share one strobe: let one camera fire it).
2. Action commands scheduled on PTP time. A GigE Vision action command is a broadcast event on the camera network: arm every camera with the same action settings, schedule one instant, and each camera fires on the shared PTP clock with no trigger wire at all. GigE Vision puts action-command precision in the microsecond class while the PTP clock itself aligns to nanosecond-class, so the event mechanism, not the clock, sets the capture skew. This is the standard answer for an all-GigE cell where pulling trigger cable to every camera is impractical.
3. PTP-based frame-rate alignment, where the camera offers it. The shared clock can also drive a slow servo that nudges free-running frame starts into phase. It is soft sync: seconds to converge, disturbed by dropped frames, suited to loosely coupled pairs rather than a fast moving line.
| Method | What it aligns | Typical agreement | Extra wiring | Choose it when |
|---|---|---|---|---|
| PTP timestamps only | device clocks | about 1 µs (protocol targets nanoseconds) | none | frames are paired or ordered, exposures need not coincide |
| Action commands on PTP time | exposure starts | microsecond-class | none | all-GigE cell, simultaneous capture, no trigger cabling |
| Shared hardware trigger | exposure starts | microsecond-class or better | one line per camera | tightest skew, mixed interfaces, deterministic cells |
| PTP frame-rate alignment | frame-start phase | soft lock, converges over seconds | none | loosely coupled camera pairs |
Where PTP timestamps earn their keep
Once every camera stamps frames from one clock, the timestamp becomes usable engineering data:
- Pairing. Pick, per part, the frames that belong together across cameras by timestamp instead of arrival order; pair frames from two synced cameras by timestamp shows the logic.
- Per-image provenance. Chunk data carries the frame timestamp and the used exposure inside each image, so downstream code sees capture time without a side log: chunk data on camera frames: carry timestamp and exposure per image.
- Clock bookkeeping. Camera time and PC receive time differ by buffering delay; knowing which one you hold matters when you correlate with logs — camera frame timestamp vs system time: why clocks differ.
- Motion fusion. When the robot or motion controller also disciplines its clock by PTP, camera timestamps and axis positions share one time axis; IEEE 1588 PTP for robot synchronization and motion control covers that side of the cell.
- Beyond cameras. The same timebase extends to lidar and IMU data: PTP timestamp sync for GigE cameras, lidar and IMU.
- Plant data. Camera events that land in a historian align by shared timestamp for the same reason process tag history does; aligning tag history to shared timestamps shows that principle on the process side.
Setting up the network the timestamps run on
Keep the camera network dedicated. PTP shares the image cable, so office traffic and a busy uplink add queueing jitter to the very messages that set the clock. Give the cell its own network card instead of sharing the IT switch, and size the ports to the station:
- Two cameras: the dual-port PoE NIC featured above, one port per camera, power over the same cable.
- Three or four GigE cameras: the MV-GE1104 four-port GigE frame grabber.
- Four cameras at 10GigE, or an aggregate well past one GigE line rate: the MV-GT2104 quad-port copper 10GbE NIC.
| Specification | MV-GE1104 | MV-GT2104 |
|---|---|---|
| Interconnect | PCI-E gen2 ×4 | PCI-E gen3 x 8 |
| Delivery bandwidth | 860 MB/s (peak bandwidth), 800 MB/s (sustained bandwidth) | 7800 MB/s (max. transmission bandwidth); 6700 MB/s (sustained transmission bandwidth) |
| Standard | GigE Vision | IEEE802.3 |
| Interface | 4 × RJ45 | RJ45 × 4 |
| Camera type supported | GigE industrial camera | 10 GigE industrial camera |
| Camera connection speed | 1 Gbps | 1000/2500/5000/10000 Mbps |
| Jumbo frame | 2/4/8 KB | 9 KB |
| Power consumption | 10 W | 12.5 W (excluding PoE) |
| Dimension | 181 mm × 126.3 mm × 21.6 mm | 181 mm × 126.3 mm × 22.4 mm |
| Temperature | Working temperature: 0 °C to 60 °C (32 °F to 140 °F); Storage temperature: –20 °C to 70 °C (–4 °F to 158 °F) | Working temperature: 0 °C to 55 °C (32 °F to 131 °F); Storage temperature: –30 °C to 80 °C (–22 °F to 176 °F) |
| Operating system | Windows 7/10/11, Ubuntu 18.04 to 24.04 on 64-bit Linux | Windows 10, Windows 11, Linux |
| Certification | WEEE, RoHS, REACH | RoHS, CE, KC |
| Price | $699.00 | $1,429.00 |
All 2 share: Camera quantity supported 4 at most; Stream 4; Humidity 10% RH to 95% RH (no condensation). Values as printed in each product's data sheet; open a part for its drawing and full specifications.
Switches: ordinary switching holds microsecond-class agreement on a small, lightly loaded camera network, and every store-and-forward hop adds some queueing variation; nanosecond-class agreement across several hops is a job for PTP-aware switching. Keep all cameras in one PTP domain, let them elect a master, and force the role only where the cell layout is fixed.
Bring-up is short: switch the IEEE 1588 feature on in the camera's transport-layer controls, watch the IEEE 1588 status feature until it reports synchronized, and give it a few seconds.
Checks before you order
- PTP support is per camera model. IEEE 1588 shows up in a model's feature list, not in every model by default: filter the GigE area scan cameras and read the feature table before you standardize, or ask us when shortlisting.
- Decide the event question first. Simultaneous exposure means action commands or a wired trigger in the panel plan; timestamps only means PTP alone and no trigger wiring.
- Ports and bandwidth. One port per camera in tight cells, and budget the aggregate rate: a GigE port carries on the order of 115 MB/s of image data, so a group pushing more than that steps up to 10GigE.
- Cabling. PoE halves the cable count and wired events add one line per camera — both are panel decisions, not retrofit decisions.
- Where to pick the parts. The frame grabbers category sorts network cards by ports and interface, and the camera finder and accessories finder filter the same choices from the other direction.
When the usual answer is wrong
- PTP is on, so we are synced. The worked example's failure: about 1.7 mm of hidden offset against a 0.1 mm budget, with timestamp checks that pass.
- Two cameras, so PTP. If one source triggers both cameras and you never compare timestamps across devices, PTP buys nothing.
- Nanosecond synchronization. The clock reaches nanosecond-class agreement; the capture skew is set by the event mechanism at microsecond-class. Quote the second number to whoever signs off the tolerance.
- PTP for the whole mixed cell. USB3 and Camera Link cameras cannot join the PTP domain; the common event there is a wire.
- Frame-rate alignment will hold it. On a fast line, a soft servo that takes seconds to converge and can drop lock is not a capture plan.
The rule to take away: PTP for comparing, events for exposing, wire when the network event is not tight enough. More timing and synchronization material sits under the systems topic. If you are standardizing a multi-camera cell, start from the camera configurator or bring the frame rates and tolerances to the contact page and we will help you size the ports, the NIC and the event plan.
Questions
Does IEEE 1588 PTP make two cameras capture at the same instant?
No. PTP aligns the clocks, not the exposure schedule. Run the check before you trust it: the shift you must tolerate is part speed × frame period — 100 mm/s × 16.7 ms ≈ 1.7 mm at 60 fps. If your tolerance is smaller than that, add a shared hardware trigger or schedule action commands on the PTP time; if it is larger, timestamps alone are fine.
How closely do camera clocks agree with PTP?
Designing around one microsecond of timestamp agreement on a small, lightly loaded camera network is realistic. Capture skew is a separate number: action commands and hardware triggers are microsecond-class. Verify by reading both cameras' frame timestamps for the same scheduled event and confirming the delta stays near one microsecond.
Why do the timestamps match but the measurements still disagree?
Because clocks and exposure schedules are different things. Check three things in order: whether the cameras free-run (frame starts drift apart by up to one frame period), which clock the analysis reads (camera timestamp or PC arrival time), and whether frames were paired by timestamp or by arrival order. The first needs a trigger or scheduled action command; the other two are software fixes.
Can PTP synchronize a USB3 or Camera Link camera with a GigE camera?
No. IEEE 1588 rides the network, so only GigE Vision devices take part. In a mixed cell the common event is a wired hardware trigger: one pulse source fires every camera regardless of interface, within microsecond-class skew. The GigE cameras can still run PTP so their timestamps share a timebase with each other and with the plant network.
What is a GigE Vision action command?
A broadcast event on the camera network. Every camera armed with the same action settings reacts to the same scheduled instant on the shared PTP clock and starts its exposure without a trigger wire. Unlike a software command sent camera by camera, one broadcast reaches all of them with no per-camera send delay. GigE Vision specifies microsecond-class precision for this mechanism.
How many cameras can one PTP domain hold?
The protocol scales well past cell sizes; the practical bound is traffic and the precision you need. A dedicated multi-port NIC or one lightly loaded switch comfortably serves the two to four cameras of a typical cell at microsecond-class agreement, and around a dozen cameras stay workable while the links stay unloaded. Past that, or for nanosecond-class skew, move to PTP-aware network infrastructure.
