Hardware encoding and software encoding

Hardware and software video encoders create a new compressed representation using different execution paths and supported configuration, rate-control and coding-tool choices.

An encoder creates a new compressed video representation from prepared pictures. A hardware encoder uses an exposed hardware encoding path; a software encoder implements its selected coding work through CPU execution.

The decision concerns the output representation. It is separate from which decoder reconstructed the input.

Choose the output before comparing implementations

The encoder must fit the target codec, dimensions, bit depth, chroma format and supported coding tools. Its rate-control and latency choices shape how it spends bits and buffers pictures.

Choice What it describes
Codec/profile Target compressed-video configuration
Picture dimensions and cadence Prepared pictures to be represented
Rate control How the encoder manages bitrate or quality objectives
Preset/tuning Implementation-specific processing and coding choices
Reordering and lookahead Available analysis and picture-order behavior

An Emby H.264-to-H.264 conversion still uses another encode when it changes the representation. Matching codec family names at the input and output does not establish stream copying.

Equal bitrate does not mean equal picture quality

Encoders can choose different predictions, quantization and other coding decisions for the same pictures. Their supported tools and preset meanings also differ.

A useful comparison holds the programme interval and target constraints together:

Preserve Why it matters
Same prepared source pictures Avoids comparing different scaling or tone mapping
Same output codec and picture configuration Avoids spending bits on a different representation
Comparable achieved output rate Nominal settings can produce different actual rates
Declared latency/analysis constraints Some choices need additional buffering
Identified implementation and preset Preset names are not universal quality units

The result should be assessed on representative content, including motion and texture. Hardware execution is not a predetermined quality score, and a “slow” label from one software encoder is not a matching mode in another encoder.

Output rate and duration describe stored payload

For an original example with a known average encoded-video rate of 5,000,000 bit/s over 1800 seconds:

5,000,000 × 1800 / 8 = 1,125,000,000 bytes

That is 1.125 GB of video payload in decimal units. Audio, packaging and transport add other bytes.

The calculation assumes the achieved average rate. A maximum-rate constraint or a quality-target setting does not automatically make the output average equal to that number.

Accepted input and completed output are different events

An encoder can accept pictures into a queue before output appears. Analysis, frame reordering and asynchronous hardware completion can separate those events.

The application must retrieve generated bitstream data and retain input storage for as long as the encoder needs it. At the end, draining obtains remaining output that may have been delayed.

See Also

References