GPU video transcoding offloads encode/decode operations to dedicated hardware blocks (NVENC and NVDEC on NVIDIA), freeing CPU resources for other work. Modern NVIDIA GPUs ship with 1-3 NVENC units (depending on tier) and a single NVDEC, supporting H.264, HEVC, and AV1 encoding on the latest generations.
For live streaming, transcoding pipelines, and 1:N transcoding services, NVENC throughput per dollar matters more than VRAM. A single RTX 4090 can typically handle 8-12 concurrent 1080p60 H.264 streams. Server applications using the unlocked NVENC limits can do more.
For batch post-production transcoding (Premiere Pro, DaVinci Resolve, FFmpeg pipelines), VRAM matters for working with high-resolution footage and effects, but raw NVENC count drives throughput. AV1 hardware encoding on RTX 40-series and newer offers significantly better compression at the same bitrate than HEVC.