3 ms·

16GiB RAM with 8-bit quantization. This is a slightly scaled up SD3 Large model (38 layers -> 60 layers).