GPU transcoding
GPU transcoding is the use of graphics processors (GPUs) to decode and re-encode video, rather than relying only on a computer's main processor. Modern GPUs include dedicated video engines that can handle many streams at once. This makes transcoding much faster, which is especially useful for live video and large upload volumes.
Why it matters
GPU transcoding lets a service process far more video per server, which shortens the wait between upload and playback and makes large live operations possible. It can lower cost per stream at high volumes. The tradeoffs are GPU prices, limited availability in the cloud and slightly lower compression efficiency than slow software encodes. Skipping GPUs entirely is also an option: FastPix encodes on-demand uploads behind its API, so there is no hardware to buy, reserve or keep busy.
How it works
- Compressed video is sent to the GPU's built-in decoder, which turns it back into raw frames.
- The frames stay in GPU memory, where they can be resized for each rendition.
- The GPU's dedicated encoder compresses each rendition, often several streams at the same time.
- Finished renditions are passed to a packager to become HLS or DASH segments.
Is GPU transcoding faster than CPU transcoding?
Yes, GPU transcoding is usually much faster than CPU transcoding, because GPUs have dedicated video decode and encode engines. A single GPU can handle many live streams at once. CPU encoders are slower but often reach better quality at the same bitrate, so many services use GPUs for live and fast turnaround, and CPUs for on-demand libraries where file size matters most.
Does GPU encoding reduce quality?
GPU encoding can produce slightly lower quality than a slow software encode at the same bitrate, but the gap is small on modern GPUs. Hardware encoders take fewer shortcuts than they once did. For live video, the difference is rarely noticeable to viewers. For on-demand catalogs, some teams still prefer software encoders to save delivery costs.
Where it's used
Game streaming platforms use GPUs to turn every streamer's feed into several quality levels in real time, and large short video apps use them to clear upload spikes in minutes.
By the numbers
- NVENC encodes video on a dedicated part of the GPU rather than on its general CUDA cores, according to its documented design. (Nvidia NVENC (Wikipedia))
