1 paper
Haoran Wu, Zeyu Cao, Yao Lai +15
Emerging agentic LLM workloads are driving rapidly growing demand on both memory capacity and bandwidth, with different phases of inference (e.g., prefill and decode) imposing dist…