A new open-source tool called ffmpeg-over-ip solves a long-standing problem for developers who need GPU-accelerated video transcoding without direct hardware access. Created by GitHub user steelbrain, the tool lets applications use remote GPU power through a simple TCP connection instead of complex hardware passthrough setups.

Traditional GPU transcoding requires painful infrastructure: Docker containers need specific NVIDIA runtime flags and driver alignment, virtual machines demand PCIe passthrough that locks GPUs to single VMs, and remote setups rely on shared filesystems with endless path mapping headaches. The tool eliminates these barriers by running a patched ffmpeg server on any machine with a GPU, while client applications connect through a single TCP port.

The architecture works by having the client binary pretend to be ffmpeg, forwarding commands to the server which tunnels all file I/O back through the connection. Files never touch the server's storage. Pre-built binaries support hardware acceleration across platforms including NVENC, QSV, VAAPI, AMF, and VideoToolbox. The project includes comprehensive documentation covering quick start guides, Docker integration, configuration options, and troubleshooting.