A camera that streams at its full rate in the desktop viewer, then falls to roughly half after a driver node touches it, is not wearing out. The node wrote an AcquisitionFrameRate value the current pixel format cannot deliver, the camera clamped the register to the bandwidth maximum, and that value stays in the camera for as long as it is powered. This guide is for anyone driving a GigE area scan camera from a robotics or acquisition node. You will read the log line that names the true maximum, restore the full rate in one step, and configure the node so the clamp cannot come back.
The log line that names the real maximum
The sequence repeats exactly. A GigE camera streams at 54 FPS in the desktop viewer. A driver node starts, requests 40 FPS, and logs a warning that reads in part: 'Value 40.000000 must be smaller than or equal 25.873647', thrown while calling AcquisitionFrameRate.SetValue(). The node then streams about 26.67 FPS. Close the node and reopen the viewer: about 26 FPS. Power-cycle the camera: 54 FPS again.
Every part of that warning comes from the camera, not from the host. The camera computed 25.873647 FPS as the maximum AcquisitionFrameRate for the settings written so far and refused the 40. The message shape — a feature name, a SetValue call, an out-of-range throw — is standard GenICam reporting, so any compliant client produces it the same way.
The decision that follows: treat the thrown maximum as the camera's bandwidth ceiling for the configuration written so far, and treat the low rate that persists afterwards as stored register state, not as a hardware fault.
The arithmetic behind a cap like 25.87 FPS
Two numbers set the ceiling: the bytes each frame needs and the bytes per second the link may carry. A GigE link tops out at 1 Gbit/s, which is 125,000,000 bytes per second. The camera's DeviceLinkThroughputLimit feature can cap that further when it was lowered for link sharing.
Worked example, with illustrative inputs of a 1624 × 992 ROI on a camera running at the full GigE rate:
- BGR8: 1624 × 992 × 3 B = 4,833,024 B per frame, so 125,000,000 / 4,833,024 ≈ 25.9 FPS — within a tenth of a percent of the 25.873647 in the log.
- Mono8: 1624 × 992 × 1 B = 1,611,008 B, so the link would allow 125,000,000 / 1,611,008 ≈ 77.6 FPS; the sensor's own limit (54 FPS in this scenario) governs instead.
- At the requested 40 FPS, BGR8 needs 4,833,024 B × 40 = 193,320,960 B/s, about 193 MB/s — 1.5 times the entire link. The same 40 FPS in Mono8 needs 64.4 MB/s, comfortably inside it.
That is why one frame-rate value can be valid for one camera and impossible for the same camera in another pixel format. Halve DeviceLinkThroughputLimit and the ceiling halves too: 62,500,000 / 4,833,024 ≈ 12.9 FPS. A 2 MP-class colour stream on GigE therefore tops out in the mid-20s FPS; the BGR versus Bayer pixel format walkthrough works through the same arithmetic for the formats GigE cameras actually ship.
Why the cap survives closing the node
GenICam features live in volatile registers: they hold their values while the camera is powered, and closing the client does not reset them. Nothing reaches non-volatile memory unless you execute a user-set save, and most cameras boot their Default user set — which is why the rate returns after a power cycle.
Persistence is the diagnostic fork. Bandwidth and exposure limits re-evaluate every frame: shorten the exposure or relight the scene and the rate moves within a second. A stored register does not move until something writes it or the power cycles. Rate unchanged by light, unchanged by exposure, healed by power cycle: register state.
One wrinkle makes the fault deterministic instead of random. Driver nodes often load a startup user set, and a set saved while the camera was clamped re-installs the low rate at every launch. The user set guide covers saving sets safely, and the leftover-settings checklist makes the same point from the test bench: start every session from a known state.
Confirm it in two minutes
Leave the camera powered, close every client, and work down this list; each step ends in a yes or a no.
- Read the range. In one GenICam viewer, read the AcquisitionFrameRate value and its allowed range. An upper bound near 26 where you expect 54 means the register holds the clamp.
- Read ExposureTime. The stuck rate of 26 FPS has a 1 / 26 ≈ 38.5 ms frame period. The session's 14990 µs (15.0 ms) exposure supports rates up to 1 / 0.01499 s ≈ 67 FPS, so exposure cannot be the limiter. If ExposureTime instead reads close to 1 / observed rate, you have the exposure case, not this one.
- Check the budget. Frame bytes times the requested rate: 4,833,024 B × 40 = 193,320,960 B/s, well past the 125,000,000 B/s link. The request was impossible, not lost; retries cannot fix it.
- Power cycle. Full rate returns: confirmed. The state lived in the registers, not in the sensor, the cable or the host.
Restore full frame rate
Fastest path without downtime: load a known-good user set. Select UserSetSelector = UserSet1 and execute UserSetLoad. The frame-rate register returns to the stored value and the stream comes back immediately.
If no known-good set exists yet, power-cycle: most cameras boot their Default user set, which restores factory values — the 54-class rate in this scenario.
Then make the fix permanent. Configure the camera once for production: pixel format, ROI, exposure, and a frame rate inside the reported maximum. Select UserSet1 and execute UserSetSave. From then on, the node's first action after connecting is UserSetLoad — one write replaces a dozen per-feature writes and cannot inherit a clamp left behind by an earlier run.
Configure the node so it cannot recur
The clamp happens when a node writes features in an order that makes the frame-rate request stale, then ignores the camera's refusal. Write features in this order and clamp against the camera's own number:
- Width, Height, OffsetX, OffsetY, binning: the ROI first, because it sizes every frame.
- PixelFormat: Mono8, YUV422 or BGR8, which multiplies the frame size by 1, 2 or 3.
- DeviceLinkThroughputLimit: only when you share a link between cameras; otherwise leave the default.
- Read the AcquisitionFrameRate range again. Its upper bound has moved after steps 1-3, and that bound is the only trustworthy ceiling.
- Set AcquisitionFrameRateEnable to true and AcquisitionFrameRate to 80-90 % of that bound, clamped in code rather than trusted from the configuration file.
Run exposure and gain fixed for a deterministic rate. Auto exposure stretches ExposureTime toward the frame period whenever the light drops, and the rate follows it down — a different mechanism with the same symptom, covered in why GigE frame rate drops at longer exposure times. Writing the features from a script follows the identical pattern; the GenICam exposure example shows the read-back-before-set habit, and it transfers directly to AcquisitionFrameRate.
Do not copy frame_rate values between configurations. The 40 above was valid for a Mono8 stream and impossible for BGR8 on the same camera, from the same arithmetic.
Same symptom, different causes
A collapsed rate has four usual origins, ordered by how often each explains a post-node collapse. The table separates them by what responds to what.
| Pattern | Signature | Quick test | Fix |
|---|---|---|---|
| Stored register (the fault here) | Cap survives closing every app; power cycle restores | Read the AcquisitionFrameRate range in a fresh viewer | UserSetLoad or power cycle, then fix the write order |
| Exposure cap | Rate tracks scene brightness; ExposureTime ≈ 1 / rate | Set ExposureTime to 1 ms; the rate jumps at once | Fixed exposure, or more light |
| Link bandwidth | Cap only with heavy pixel formats; Mono8 reaches the sensor limit | Switch to Mono8 and compare | Lighter format, smaller ROI, or a faster camera class |
| Waiting on triggers | Delivered rate equals trigger rate; AcquisitionFrameRate is irrelevant | Count trigger pulses against delivered frames | TriggerMode Off for free-run, or repair the trigger source |
On a USB3 camera the bandwidth row comes from hub budget rather than a throughput register; the USB hub bandwidth walkthrough covers that math. On GigE, remember the link itself can be the bottleneck: when the link drops below gigabit, the ceiling falls with the negotiated speed. For the trigger row, this triggered frame rate troubleshooting walkthrough works a delivered-rate-versus-commanded-rate diagnosis on an area scan camera. The rest of the frame-rate fixes sit on the systems topic page.
Hardware that holds the rate, and the decision
When the application genuinely needs 40 FPS or more in colour, the fix is link budget, not configuration. In the area scan cameras range, filter by interface: USB3 Vision carries roughly three times GigE's throughput and 10GigE carries ten times it, provided the host NIC negotiates the faster link — the 10GigE throughput fix shows how a misconfigured port silently falls back to 1 Gbit/s.
Two C-mount GigE cameras in the resolution classes the arithmetic above assumes, 2.5 MP and 6 MP:
- Sensor format
- 1/1.8"
- Lens mount
- C-mount
- Sensor format
- 1/1.7"
- Lens mount
- C-mount
The MV-CU020-90GC is a 2.5 MP, global-shutter, colour model — the class of camera behind this scenario. The 6 MP model carries 6 MP at the same GigE interface, where the byte budget tightens further and a lighter pixel format or a smaller ROI matters even more.
On the host side, give the camera a port of its own instead of sharing the workstation NIC with other traffic. The vision controllers line is built for it:
- LAN ports
- 3-4 LAN ports
- USB 3 ports
- 2 USB 3 ports
- Light outputs
- 4 built-in channels
- Memory
- 8 GB
The MV-VC2140-128G40-1T carries 3-4 LAN ports, 2 USB 3 ports and 4 built-in light channels, so the camera link stays dedicated while the node also drives lighting.
Decision, restated: if the collapse follows a node run and a power cycle heals it, the fix is the write order above plus a known-good user set — no new hardware required. If the rate you need exceeds the link budget in your pixel format, pick the camera class and host that carry it; the parts above are the two starting points. To size the whole chain, open the camera configurator, or write to us through contact with your required rate, pixel format and resolution, and we will check the budget before you order.
Questions
Why does a power cycle bring the frame rate back?
Camera registers hold their values while the camera is powered, and boot loads the Default user set. The clamped frame-rate write disappears from the registers, the factory value returns, and the stream goes from about 26 FPS back to 54. Nothing was repaired, because nothing was broken: the low rate was stored state, not damage.
Does the out-of-range write damage the camera?
No. The out-of-range exception is the camera refusing a value its current configuration cannot deliver; nothing is written to non-volatile memory and no wear occurs. The only cost is throughput until the register is restored. If the camera reached its full rate before the node ran, the hardware is fine, and a firmware update is unrelated to this fault.
What frame rate should I put in the driver configuration?
None as a fixed number. Write pixel format and ROI first, read the AcquisitionFrameRate range, and clamp the requested value in code to 80-90 % of the reported maximum. For the scenario above that is about 21-23 FPS in BGR8 on GigE, or the full sensor rate in Mono8. A hard-coded 40 is what produced the clamp.
How do I store a known-good configuration in the camera?
Configure the camera for production, write UserSetSelector = UserSet1, and execute UserSetSave. Verify by power cycling, loading UserSet1 with UserSetLoad, and confirming the expected frame rate and pixel format. Only then point the node's startup at that set. Save from a proven-good state: a set saved while the camera was clamped stores the fault permanently.
Our camera idles near 5 FPS in the viewer even with defaults. Same fault?
Not necessarily. Check ExposureTime first: a 15 ms exposure supports roughly 67 FPS, so 5 FPS with a short exposure is not an exposure cap. Then check TriggerMode, then DeviceLinkThroughputLimit and the negotiated link speed. If shortening the exposure does not move the rate but a power cycle does, the stored-register fault described here is the likely cause.


