Using your graphics card
Your Sols answer much sooner when their model runs on a graphics card. This page covers what the card does for the models, and how to tell which model fits yours. Setting up a card for the whole server is in Graphics cards.
Card or processor
The models work on the processor alone, but slowly. On four processor cores, Angel 1 takes minutes to turn a chat into memories. On a graphics card it takes seconds.
What counts is the card's own memory. A model that fits in it runs fastest. One that doesn't still runs, with what doesn't fit on the processor, but slowly. Graphics built into the processor have no memory of their own, and run models little faster than the processor does.
Which model fits
| Card memory | Model it suits |
|---|---|
| 20 GB or more | Seraphim 1 |
| 8 GB or more | Dominion 1 |
| 4 GB or more | Angel 1 |
| Less, or no card | Angel 1 |
Settings › Models suggests a model from this, under Best for this server, with the reason in a sentence. Each model's card also warns you when your graphics card has less memory than that model needs. See Picking a model.
When the card's memory can't be read
can't always tell how much memory a card has. An NVIDIA card's memory, for one, can't be read without NVIDIA's nvidia-smi tool. Then Settings › Models suggests Dominion 1 as a safe guess, says a bigger model may fit too, and shows no warnings on the model cards. Check the card's memory before you choose a bigger model.
To see what found, open Settings › About and look at Graphics. It lists each card, and its memory when it can be read.
Sharing the card
Your server runs one model at a time, so only the model in use takes room on the card. Other things can share the card with it, which leaves less room for the model. Graphics cards covers what else can use it.