Hardware acceleration uses a fixed-function encoder built into a GPU or SoC rather than a software encoder running on general-purpose CPU cores. NVIDIA's NVENC, Intel's Quick Sync, AMD's VCE and Apple's VideoToolbox are the common ones.
The speed difference is large — often an order of magnitude — and the quality difference is real in the other direction. At the same file size, a hardware encoder generally produces worse output than x264 at a reasonable preset, because it implements a subset of the tools a software encoder can search and cannot spend arbitrary time searching them. The gap has narrowed with each hardware generation but has not closed.
Choose hardware when:
- The job is live. Streaming and real-time capture have a hard deadline; the quality trade is not optional.
- Throughput matters more than bytes. Batch-converting a large library, or producing proxies for editing.
- Battery matters. Dedicated silicon is far more efficient per joule.
Choose software when:
- The output is final and will be stored, sent or published.
- The file has to fit a limit, where every percent of efficiency is a percent of quality you get to keep.
The measurements on this site were produced with software libx264 at preset medium, deliberately. A hardware encode at the same nominal settings would not reproduce them, and a comparison that mixed the two would not mean anything.