Nadi
Nadi runs big, long-context AI jobs on a GPU too small to hold them, by moving the model's working memory into ordinary RAM. Same answers, on the card you already have.
Research, not announcements — nothing here has a date until it has a build.
The current public research notes.