March 20, 2026
Popcorn, pixels, and hot takes
Video Encoding and Decoding with Vulkan Compute Shaders in FFmpeg
GPU vs CPU cage match: editors cheer, skeptics yell “overkill”
TLDR: FFmpeg now uses Vulkan compute to keep video encoding/decoding fully on the GPU, promising smoother pro workflows without special chips. The comments are split: editors hype the no‑CPU handoff speed boost, while skeptics argue CPUs already handle 4K fine and warn about crashes and real‑world streaming paths.
FFmpeg just flipped the table by using Vulkan compute to speed up video encoding and decoding on regular graphics cards — no special video chip required. The devs say it keeps everything on the GPU to dodge the slow “back-and-forth” with the CPU, a move meant to help pros wrangle massive footage without buying a spaceship workstation. The crowd? Absolutely divided.
On one side, hype is loud. Commenters like hirako2000 call full-on GPU workflows a big deal for editing, especially for heavyweight formats (think archival film scans and studio codecs). Others point to Vulkan tricks that let today’s GPUs crunch more pieces in parallel than those classic codecs ever imagined. Even FFmpeg fans are dreaming of buttery scrubbing in Blender and Resolve without exporting warehouses of TIFFs.
Then the skeptics arrive with air horns. doctorpangloss claims old CPUs already crush 4K in real time and argues that streaming paths (hello WebRTC) hit the CPU anyway. sylware warns that hardware decoding can crash when the data’s messy, saying software still wins when things get weird. Jokes fly about “liquid-cooled 100-core boxes” doubling as space heaters, while meme-lords declare: “Delete the CPU?” “Delete the GPU?” — choose your fighter.
Result: an all-out GPU vs CPU showdown. Supporters say this finally fixes the hybrid mess; skeptics say it’s flashy tech for problems the CPU solved yesterday. Grab popcorn; this thread is rendering in drama at 120 fps.
Key Points
- •FFmpeg is using Vulkan Compute to run video encoding and decoding entirely on GPUs, targeting professional-grade workloads.
- •This compute approach complements Vulkan Video’s fixed-function hardware by accelerating formats and workflows not covered by dedicated video engines.
- •Codec pipelines mix parallel and serial stages, creating challenges for GPU acceleration; JPEG illustrates these constraints.
- •Hybrid CPU–GPU offloading often fails due to memory transfer latency, with examples including dav1d and x264’s OpenCL path showing no net gains.
- •The article concludes compute-based implementations must be fully GPU-resident, avoiding CPU hand-offs, and notes codecs define minimal parallel units (slices/blocks).