Industrial Camera Packet Loss Issues (Network Card Settings Chapter, Camera to Network Card)

I. First, Clarify the Data Transmission Process from Industrial Camera to Network Card

Taking a common GigE industrial camera as an example, image data roughly passes through the following path:

This article uniformly adopts the structure where the industrial camera is directly connected to the computer network card:

This article only discusses the scenario where an industrial camera is directly connected to a computer network card via a single network cable, focusing on the single path from the camera sending data, through the network cable, into the computer network card, and into the network card receive buffer.

It should be noted that the industrial camera sends a continuous data stream. When the speed at which the camera sends data exceeds the speed at which the network card, driver, or operating system can process it at a given moment, reception congestion or packet loss may occur.

II. Why Do Network Card Settings Affect Industrial Camera Packet Loss?

Many users believe that as long as the computer network card supports 1 Gbps or 10 Gbps, it can stably receive industrial camera data. In reality, the actual receiving capability of a network card depends not only on its nominal bandwidth but also on the following factors:

• Actually negotiated link speed;
• Network card driver;
• Receive Buffers;
• Network card interrupt handling mechanism;
• RSS (Receive Side Scaling);
• Flow Control;
• Jumbo Frame;
• Network card power saving features;
• Network card hardware offload features;
• Computer CPU and PCIe bus resources.

Therefore, when troubleshooting industrial camera packet loss, one should not only check "whether the network card is gigabit", but also further inspect the actual operating status and advanced parameters of the network card.

III. First Check: Actual Connection Speed Between Camera and Network Card

First, confirm the actual link speed established between the industrial camera and the computer network card.

For example, if the industrial camera supports 1 Gbps, but the computer network card, due to the network cable, connector, switch port, or negotiation anomaly, actually only establishes a 100 Mbps link, then it is very easy to understand that data congestion will occur under high-speed continuous acquisition.

Common link speeds include:

• 100 Mbps;
• 1 Gbps;
• 2.5 Gbps;
• 5 Gbps;
• 10 Gbps.

In Windows, you can check the connection status of the corresponding network adapter through "Settings → Network & Internet → Advanced network settings", or through "Control Panel → Network and Internet → Network Connections".

IV. Second Check: Network Cable and Physical Link

Before modifying advanced network card parameters, it is recommended to first rule out physical link problems.

Industrial cameras require significantly higher stability from network links during long-term continuous transmission compared to ordinary office networks. It is recommended to check:

• Whether the network cable category meets the current link speed requirements;
• Whether the network cable has obvious bends, damage, or loose connectors;
• Whether the crystal connector or industrial connector has reliable contact;
• Whether the network cable length is within the range recommended by the device;
• Whether the connectors and interfaces at both ends of the network cable are normal;
• Whether the computer network card experiences frequent renegotiation.

If packet loss is significantly reduced after replacing with a reliable network cable, then priority should be given to solving the physical link problem, rather than continuing to modify network card parameters.

V. Third Check: Network Card Driver

The network card driver is responsible for connecting the network card hardware, the Windows network system, and upper-layer applications. The driver version and configuration method may directly affect the receiving performance of the network card.

In "Device Manager → Network Adapters", it is recommended to confirm:

• Specific model of the network card;
• Driver version;
• Driver date;
• Whether the network card is working properly;
• Whether there are device errors or abnormal identifiers.

If the system is currently running stably, it is not recommended to frequently change drivers for the sake of "optimization". A more reasonable approach is to record the current driver version and perform targeted testing based on the official driver provided by the network card manufacturer when problems arise.

VI. Key Check: Receive Buffers

For industrial camera packet loss issues, Receive Buffers is one of the key items to check in the advanced network card settings.

The industrial camera continuously sends data, and the data received by the network card is not all handed over to the application for processing at the same instant. The network card first needs to temporarily store the data using the receive buffer, and then continue processing through the driver and operating system.

If the camera data arrival speed is high and the receive buffer is set too low, buffer insufficiency may occur.

In different network card drivers, this parameter may be displayed as:

• Receive Buffers;
• Receive Buffer;
• RX Buffers;
• Receive Buffer (Chinese);
• Receive Descriptor.

If the driver allows adjustment, you can appropriately increase the receive buffer according to the supported range of the network card and conduct actual acquisition tests.

VII. Do Transmit Buffers Need Adjustment?

Some network card drivers also provide the Transmit Buffers parameter.

Industrial cameras mainly send image data to the computer, so the receive direction is usually the focus of troubleshooting. However, the communication between the camera and the computer is not completely one-way; control commands, status information, and GigE Vision related control communications still need to be transmitted over the network.

Therefore, the transmit buffer can also be considered as a troubleshooting item, but generally it should not be treated as the first priority for industrial camera image packet loss issues.

VIII. Check Flow Control

Flow Control settings can be seen in the advanced properties of many network cards.

Common options include:

• Disabled;
• Rx & Tx Enabled;
• Rx Enabled;
• Tx Enabled.

Flow control can help alleviate pressure caused by data bursts in specific network environments. However, the specific settings in an industrial camera system need to be tested in combination with the camera, network card, switch, and the entire network topology.

If there is only one camera in the system and the camera is directly connected to the computer network card, the network structure is relatively simple; this article does not discuss network topologies such as switches and multi-camera aggregation, and only provides explanations for the scenario of a single industrial camera directly connected to a single computer network card.

IX. Check Energy Efficient Ethernet and Other Power Saving Features

Modern network cards usually include power-saving related features, such as:

• Energy Efficient Ethernet (EEE);
• Green Ethernet;
• Power Saving;
• Energy-saving Ethernet.

These features are mainly aimed at reducing power consumption. On dedicated acquisition network cards for industrial cameras, if the system experiences occasional communication anomalies, abnormalities when restarting acquisition, or problems after long-term operation, these power-saving features can be considered as troubleshooting items.

If testing requires disabling a certain power-saving feature, only one major parameter should be modified at a time, followed by re-running continuous acquisition tests.

X. Jumbo Frame and Industrial Camera Packet Loss

Some industrial cameras and network cards support Jumbo Frame. Larger network frames can reduce the number of data packets corresponding to a unit of data volume, which helps reduce some network processing overhead in suitable network environments.

However, Jumbo Frame is not simply a matter of setting the network card to 9000 to solve packet loss problems.

If there are devices in the network link that do not support the corresponding MTU, communication anomalies may occur. Therefore, before using Jumbo Frame, confirm:

• Whether the industrial camera supports it;
• Whether the network card supports it;
• Whether the network card driver supports it;
• Whether the computer network card supports it;
• Whether a consistent MTU configuration can be used between the industrial camera and the computer network card.

If the system has not been fully verified, it is recommended to first use standard configuration to establish a stable baseline, and then conduct Jumbo Frame tests.

XI. Check Interrupt Moderation

After the network card receives network data, it needs to notify the CPU for processing. Modern network cards usually improve processing efficiency through interrupt mechanisms.

Interrupt Moderation can reduce the number of interrupts received by the CPU, reducing system overhead through a certain degree of data coalescing.

For high-throughput industrial camera systems, if packet loss, latency, or high CPU network processing pressure occurs, this parameter can be included in the testing scope.

It is important to emphasize that one should not simply assume that "disabling interrupt moderation is always best". The final configuration should be determined through actual acquisition tests.

XII. Check RSS (Receive Side Scaling)

If the industrial computer uses a multi-core CPU, the network card may support RSS (Receive Side Scaling), which distributes network receive processing across multiple CPU cores.

For high-bandwidth industrial camera systems, RSS can help improve the CPU distribution of network data processing.

If it is found that one CPU core has significantly high load while other cores have low utilization during industrial camera operation, further check RSS, network card interrupt distribution, and related driver settings.

XIII. Check Network Card Hardware Offload Features

Common network card hardware offload features include:

• IPv4 Checksum Offload;
• TCP Checksum Offload;
• UDP Checksum Offload;
• Large Send Offload;
• Receive Segment Coalescing.

These features are mainly used to reduce the CPU's network processing burden.

Due to differences in combinations of different industrial cameras, acquisition SDKs, Windows versions, and network card drivers, it is not recommended to disable all offload features without testing basis.

If troubleshooting compatibility issues, a method of modifying items one by one and verifying them one by one should be adopted, and each test result should be recorded.

XIV. Relationship Between Industrial Camera Packet Size and Network Card Settings

In GigE Vision industrial cameras, an important parameter commonly seen is Packet Size.

It determines the size of data packets used when the camera sends image data. The data sent by the camera ultimately needs to be received by the network card, so Packet Size has a direct relationship with the network card, switch, and the entire network path.

If the Packet Size set by the camera exceeds the supported range of a device in the network link, communication anomalies may occur.

Therefore, when adjusting Packet Size, the capabilities of the network card and switch should be confirmed simultaneously, rather than only modifying the camera parameters.

XV. Packet Delay and Network Card Receive Pressure

When a high-speed industrial camera sends image data, data packets will continuously arrive at the computer network card through the network cable in a very short time.

This article does not discuss the situation where multiple industrial cameras send data simultaneously, but only analyzes the data transmission from a single industrial camera to a single computer network card.

Some industrial cameras support parameters such as Packet Delay, which can reduce instantaneous network traffic pressure by appropriately increasing the sending interval between data packets.

Therefore, even with a single camera directly connected to the network card, the sending rhythm on the camera side and the receiving capability of the network card should be checked simultaneously.

XVI. This Article Uniformly Adopts the "Industrial Camera → Network Cable → Computer Network Card" Direct Connection Mode

To make problem localization more clear, this article does not introduce network devices such as switches, routers, or multi-camera aggregation, and only discusses the situation where one industrial camera is directly connected to a computer network card via one network cable.

Under this structure, the troubleshooting path can be clearly divided into:

1. Whether the industrial camera network interface is sending data normally;
2. Whether the network cable can stably transmit data;
3. Whether the computer network card establishes a link at the correct speed;
4. Whether the network card driver is working properly;
5. Whether the network card Receive Buffers are sufficient;
6. Whether advanced network card features cause reception anomalies;
7. Whether the system can process data in time after the network card receives it.

The biggest advantage of this single-path structure is that the fault range is relatively clear. When packet loss occurs, you can check from the camera end all the way to the network card end without simultaneously analyzing factors such as switch ports, switch caches, and network aggregation.

XVII. Single Industrial Camera Also Needs Bandwidth Check

Even with only one industrial camera, bandwidth issues cannot be ignored.

For example, when a high-resolution, high-frame-rate camera continuously outputs a large amount of image data, if the actual link is only 100 Mbps and the camera data traffic already approaches or exceeds this level, obvious data transmission pressure may be generated.

Therefore, under single-camera direct connection mode, focus on confirming:

• Maximum network rate of the camera interface;
• Maximum network rate of the computer network card;
• Current actual negotiated link speed;
• Actual image data traffic of the camera;
• Whether the network card receive buffer can withstand data bursts.

If the actual link speed is lower than the data traffic generated by the camera, simply modifying advanced network card parameters cannot fundamentally solve the bandwidth insufficiency problem.

XVIII. Recommended Troubleshooting Sequence for "Camera to Network Card"

1. Confirm camera working mode: Confirm whether the camera is in a normal continuous acquisition state.
2. Confirm actual link speed: Check whether the expected network speed is established between the camera and the computer.
3. Replace with a reliable network cable: First rule out the most basic physical link problem.
4. Confirm network card driver: Record the network card model and driver version.
5. Check Receive Buffers: Confirm that the receive buffer is not set too low.
6. Check Flow Control: Test according to the actual network topology.
7. Check power saving features: Troubleshoot EEE, Green Ethernet, and other features.
8. Check Jumbo Frame: Confirm whether the entire link supports a consistent MTU.
9. Check RSS and Interrupt Moderation: Further test for high-load and multi-core CPU environments.
10. Check camera Packet Size: Confirm that the camera data packet size matches the network environment.
11. Check Packet Delay: Focus on instantaneous traffic in multi-camera environments.
12. Conduct long-term testing: Do not test for only a few seconds; continuous acquisition verification should be performed.

XIX. Why Re-test After Modifying One Network Card Parameter?

Industrial camera packet loss problems often involve multiple parameters. If Receive Buffers, Flow Control, Jumbo Frame, RSS, and Interrupt Moderation are modified simultaneously, even if the problem is improved, it is impossible to determine which parameter actually worked.

A more reasonable approach is to establish a test baseline:

Record original settings → Modify one parameter → Continuous acquisition → Record results → Compare with baseline.

If conditions permit, you can also record:

• Continuous acquisition time;
• Number of acquired frames;
• Number of lost frames;
• Number of camera alarms;
• Network card statistics;
• CPU usage;
• Memory usage.

This can turn "feeling more stable" into comparable data.

XX. Industrial Camera Packet Loss Cannot Be Judged Only by the Network Card

The focus of this article is on the section from "camera to network card", but packet loss problems in actual engineering may come from multiple links.

XXI. Basic Configuration Ideas for Dedicated Industrial Camera Network Cards

If the computer uses an independent network card specifically connected to an industrial camera, you can establish a basic configuration according to the following ideas:

The core principle of this process is: first ensure link stability, then adjust the network card; first establish a baseline, then optimize item by item.

XXII. Common Misconceptions

Misconception 1: The network card is gigabit, so there will definitely be no packet loss

Actual link speed, network card cache, driver, CPU, switch, and camera sending parameters all affect the result.

Misconception 2: Setting Receive Buffers to maximum is definitely the best

A larger receive buffer can increase cache space, but it does not mean that all systems can achieve better overall performance as a result. It should be determined based on actual tests.

Misconception 3: Jumbo Frame will definitely solve packet loss

Jumbo Frame requires support from the entire communication link. If there are devices in the link that do not support the corresponding MTU, new communication problems may arise instead.

Misconception 4: Disabling all advanced network card parameters

This method may change the overall network processing method of the system and does not equal optimization. The correct approach is to verify item by item based on the actual problem.

Misconception 5: Testing for only a few seconds and considering the problem solved

Industrial cameras usually need to run continuously for long periods. Occasional packet loss can only be more easily detected after longer testing.

XXIII. Summary

Industrial camera packet loss is a typical system-level network problem.

From the perspective of "camera to network card", one should first ensure physical link stability, then confirm actual link speed and network card driver status, and further check parameters such as Receive Buffers, Flow Control, power saving features, Jumbo Frame, RSS, and Interrupt Moderation.

At the same time, it is also necessary to comprehensively analyze the Packet Size, Packet Delay, frame rate, and multi-camera bandwidth on the industrial camera side.

The truly effective troubleshooting method is not to modify a large number of parameters at once, but to gradually verify in the order of:

Physical Link → Actual Speed → Network Card Driver → Receive Buffer → Advanced Network Card Parameters → Camera Sending Parameters → Long-term Acquisition Test

For industrial automation and machine vision systems, only by analyzing "camera, network cable, network card, driver, and acquisition software" as a complete data transmission link can the real cause of industrial camera packet loss be more accurately located.