The most cost effective GPU server for AI image generation balances VRAM capacity, network quality, and total opera