JD Ads Cuts AI Inference Latency 15% by Rethinking the CPU SideHOT
JD's advertising arm says the real bottleneck in generative ad recommendation sits on general-purpose servers, not GPUs. Working with Huawei's Kunpeng team, it collapsed four cache tiers into two and rebuilt its RPC layer, cutting end-to-end inference latency 15% ahead of the 618 shopping festival.





