What Is a Codec?

A codec, short for coder-decoder, is a method for compressing digital audio or video and then decompressing it again. The encoder side shrinks the data so it is cheaper to store and send. The decoder side rebuilds it for display, playback or analysis. The word can mean the standard itself, such as H.264, or a piece of software or hardware that implements it.

Compression matters because raw video is enormous. Uncompressed 1080p at 60 frames per second runs to roughly 3 Gb/s, which is why it needs SDI or HDMI cables rather than a network. An H.264 encoder can carry a similar picture in well under 10 Mb/s, a reduction of several hundred times.

Most video codecs are lossy: they throw away detail the eye is unlikely to miss, so the decoded picture is close to, but not identical to, the original. Lossless codecs keep every bit and compress far less, so they are kept for archiving and medical or scientific images.

Video encoding is the process of turning raw frames into a compressed stream. Modern codecs use the same broad steps:

Each codec generation trades more processing for a lower bitrate at the same quality. The figures below are typical, and real results depend on the content, resolution and encoder settings.

H.264 remains the safe default because almost every device can decode it. H.265 suits 4K and multi-camera systems where bandwidth or storage is tight. AV1 is growing in web streaming because it is royalty-free, and H.266 is still early, with limited hardware support.

Software encoders run on the processor and give the most control over quality, but they use a lot of CPU, especially at 4K. Hardware encoders are dedicated circuits in GPUs, processors and capture devices, such as Intel Quick Sync Video and NVIDIA NVENC. They encode several streams in real time at low power, which is what multi-camera and edge AI systems need.

Hardware support varies by chip, so it is worth checking before choosing a computer. For example, NVIDIA’s Jetson Orin Nano has no hardware video encoder, while other Jetson Orin modules do. Many capture cards, which bring video into a computer, pass raw frames to the host, but some encode on the card. Our range of M.2 video capture cards includes an HDMI card with 1080p30 hardware encoding, which takes that load off a small computer.

In streaming and AV-over-IP, the codec sets latency as well as bandwidth. Our hardware video encoders and decoders compress HDMI and SDI sources to H.264 or H.265 for IPTV, recording and remote viewing, while NDI encoders use NDI’s own low-latency compression, or H.264 and H.265 in the NDI HX versions, for live production networks.

For AI analysis, video usually has to be decoded back to raw frames before a model can use it, so decode capacity matters as much as encode. Feeding a model from a capture card avoids that round trip and keeps the full picture quality.

Latency is the other trade-off. Long groups of pictures and B-frames improve compression but add delay, so remote control, robotics and live monitoring links use low-latency settings, intra-only codecs such as MJPEG, or uncompressed video over SDI. Recording and distribution can accept more delay in exchange for smaller files.

We fit capture cards and set up hardware encoding in industrial and edge AI computers as standard, matching the codec, resolution and number of streams to each system before it ships.