What a capture card does
A capture card is hardware that receives a live video signal and delivers it to a computer as digital frames in memory, ready for software to record, stream, display or analyse. Cameras, medical scopes, broadcast equipment and other computers send video as a continuous signal meant for a monitor, which a computer cannot read directly. The capture card sits between the two.
Inside, a receiver chip decodes the incoming signal, such as HDMI, SDI or analogue composite. A processor or FPGA then converts the colour format, scales the picture if required and writes each frame across the PCIe or USB bus into system memory. Some cards also compress the video with H.264 or H.265 on board, which helps when the computer has to record many channels at once. Software then sees the card as a standard video device, through V4L2 on Linux or DirectShow and Media Foundation on Windows.
Internal vs external capture cards
Capture devices come in four main formats. The choice depends on how many inputs you need, the space inside the computer and how low the latency has to be.
| Format | Connects through | Typical channels | Latency | Best for |
|---|---|---|---|---|
| PCIe card | PCIe slot, x1 to x8 | 1 to 16 or more | Lowest, direct to memory | Multi-camera recording, 4K and 12G-SDI, AI servers |
| M.2 module | M.2 slot, usually M key | 1 to 8 | Low | Compact and fanless computers, Jetson systems |
| Mini PCIe card | Mini PCIe slot | 1 to 4 | Low | Small embedded boards and vehicle computers |
| USB device | USB 3 port | Usually 1 | Higher, varies with host | Portable use, laptops, quick setups |
For anything that runs every day, an internal card is usually the better fit: it cannot be knocked out, it shares no bandwidth with other USB devices and multi-channel cards keep cabling tidy. USB capture suits laptops, temporary installs and computers with no free slot.
What to look for in a capture card
Five specifications decide whether a card fits the job.
-
//
Inputs. The video connector and signal: HDMI for cameras and computers, SDI over BNC for broadcast and long runs, DVI or DisplayPort for computer outputs, composite or HD-TVI for analogue CCTV.
-
//
Resolution and frame rate. 1080p60 covers most cameras, while 4K60 needs HDMI 2.0 or 12G-SDI inputs and a wider PCIe link.
-
//
Channels. How many sources the card captures at once, and whether every channel runs at full resolution and frame rate together.
-
//
Latency. How quickly frames reach software. It matters for live analysis, robot control and surgical displays.
-
//
Drivers and SDK. Support for the operating system you deploy, including Linux on NVIDIA Jetson, and for the software you plan to use.
Analogue vs digital capture
Digital sources such as HDMI, SDI and DisplayPort already carry the picture as data, so the card receives it, checks it and passes it on with no loss. Analogue sources, such as composite video from older cameras or HD-TVI, AHD and HD-CVI from coax CCTV, have to be digitised first. The card samples the waveform, separates colour from brightness and builds each frame, and the quality depends on the decoder chip as much as on the camera.
Analogue capture is still common on vehicles, in retrofits and in defence systems with long service lives. Our analogue TVI capture cards bring HD-TVI cameras into a computer over existing coax, from single-input USB devices to 16-input PCIe cards.
Capture cards in industrial and edge AI systems
Capture is often the first link between existing video equipment and AI software. A hospital can add AI analysis to an endoscopy tower by capturing its SDI or HDMI output, without touching the medical device itself. A broadcaster can feed camera signals into an encoder or replay system. A defence or security system can bring SDI, HD-TVI and HDMI feeds from different cameras into one computer for recording and object detection.
Low-latency cards write frames straight into GPU memory, so the AI model starts work without an extra copy. SDI inputs use the BNC connector, and our guide to SDI capture for edge AI covers broadcast-grade inputs in depth. The endoscopy video capture case study shows one in use on a medical cart.
Capture card vs frame grabber
The terms overlap. A capture card usually takes standard video signals such as HDMI, SDI or composite. A frame grabber traditionally takes machine vision camera interfaces, such as Camera Link, CoaXPress, GMSL2 or multiple GigE Vision ports, and adds triggering and precise timing for inspection systems. Many makers now use the two names interchangeably for analogue and multi-channel cards too. Our PCIe video capture cards range from single-input HDMI cards to 16-channel analogue and multi-input SDI boards, and our M.2 video capture cards fit compact edge AI computers.
Quick answers on capture cards
Capture cards: common questions
What is a capture card used for?
Getting live video into a computer. Uses range from streaming and recording to medical imaging, broadcast encoding, machine vision and AI video analytics.
Do you need a capture card to stream?
Only for an external video source. Streaming from the computer’s own screen or a USB webcam needs none. A camera, console or second computer with an HDMI or SDI output needs one.
Does a capture card affect video quality?
Not if it captures uncompressed. Cards that pass raw frames keep the full source quality. Cards that compress on board trade some quality for lower bandwidth and storage.
Can a capture card record several cameras?
Yes, multi-channel cards can. Industrial PCIe cards take four, eight or 16 inputs and capture them all at once.
Do capture cards work on Linux?
Many do. Industrial cards usually ship Linux drivers and an SDK, including builds for NVIDIA Jetson. Check support for your kernel version before buying.
We fit capture cards in industrial and edge AI computers as standard, install the drivers and validate each input with your sources before the system ships.