I Put Three AI Services on One 24 GB GPU. It Used 17.8. Here's What Fits.