All GPU jobs on Very Good FFmpeg now run on Nvidia L4 hardware. If you have ever seen a GPU job fail with an error that had nothing to do with your command, and then succeed on retry, this is the change that fixes it.
Nothing about the request changes. "machine": "nvidia" routes to an L4, and the same commands, pricing, and 6-hour runtime apply.
What changed
Previously, GPU jobs landed on whichever worker was free from a mixed pool of Nvidia RTX and A-series cards. That pool is gone. Every GPU worker is now an Nvidia L4 running the same driver image.
{
"input_files": { "input.mp4": "https://example.com/video.mp4" },
"output_files": ["output.mp4"],
"machine": "nvidia",
"ffmpeg_commands": ["-i inputs/input.mp4 -c:v h264_nvenc outputs/output.mp4"]
}The request above is identical to what you would have sent last month. See Hardware Selection for the full options.
Why we made the switch
Reliability. The mixed pool meant mixed driver setups, and a small share of GPU jobs failed outright with driver-level errors. The failures were not caused by the user's command, and the same job would usually succeed when retried onto a different worker. That is the worst kind of failure for an API: rare enough to be hard to reproduce, common enough that customers had to build retry logic around it.
Standardising on one data-center card with one driver image removes that class of failure entirely. Every GPU job now runs on identical hardware and identical drivers, so a command that works once works every time.
FFmpeg 8 on GPU workers
A happy side effect of rebuilding the GPU image from scratch: GPU workers now run FFmpeg 8. If your NVENC commands relied on a newer encoder option or filter that the old workers did not have, it should now work.
What existing users need to do
Nothing.
- Existing jobs, saved workflows, scripts, and n8n nodes that set
"machine": "nvidia"keep working unchanged. - GPU pricing has not changed.
- Retry logic you added to work around GPU flakiness can stay, but it should no longer trigger.
If you had tuned a command against a specific consumer card, it is worth a quick re-run on L4 to confirm the output matches what you expect.
Get started
New to the API? Grab a free API key (2 GB free, no credit card) and send your first GPU job. The Hardware Selection guide has a copy-paste example.