GP is likely referring to network latency here. There's a tradeoff between smaller GPUs/etc at home that have no latency to use and beefier hardware in the cloud that have a minimum latency to use.