Converting a channel takes a lot of processor. A server with a video card (GPU) can do it on the card instead, for a fraction of the processor. Minimus uses the card for transcoded channels and for created channels that convert their titles.
Which cards
- Minimus uses NVIDIA cards (NVENC), and Intel and AMD graphics (VA-API, or Quick Sync on Intel).
- So far, encoding on the card has been tested on AMD cards. NVIDIA and Intel cards are used when a server has one, but are not proven yet: after turning one on, check the picture and the server's processor use.
- With no card that works, the processor encodes, as on a server without a card.
How it works
- Each server tests its video card when it starts: it encodes one test frame with each kind of encoder, and reports what worked.
- Open the server's page (Edit Server). Under Node, Video card shows it: for example AMD (VA-API), NVIDIA (NVENC) or Intel (Quick Sync). None: the processor encodes means no card worked; Not reported yet means the server hasn't reported since it was updated.
- In Settings → Streaming, Encode on the video card when a server has one is on by default. Turn it off to encode everything on the processor.
If the card fails on a created channel's title, that server goes back to the processor on its own, until it restarts.
Setting up a server with a card
- The install and update scripts give the streaming software access to the card when the server has one.
- NVIDIA cards also need NVIDIA's driver and its container toolkit installed on the server, so Docker can use the card. Intel and AMD need the usual graphics driver.
- After adding a card to an existing server, update the server so it gets access to it.
Good to know
- Sending a channel as it is (Original quality) never uses the card or much processor: the card only matters for channels you convert.
- What uses the card: Quality set to Convert on a channel, and created channels with a logo, a banner or titles in another format.
Updated 12 November 2026