Supposedly double the model size so 14gb. RTX 4090 might be able to handle it. You can use lambdalabs to rent a server gpu for one of the larger models.