134 second inference on a 19 second video, 26 token prompt. RTX 3060. Or should/can I only increase my GPU?