The monthly cost for an A100-based instance will start at ... a discrete CPU and discrete GPU that communicate over PCIe. “This is Nvidia’s inference platform, one architecture for diverse ...