HBM Is Not All You Need: Efficient Disaggregated LLM Serving across Memory-heterogeneous Accelerators
## Episode Summary In this episode, we cover: - **HBM Is Not All You Need: Efficient Disaggregated LLM Serving across Memory-heterogeneous Accelerators** (arXiv) - **TraceLab: Characterizing Coding Agent Workloads for LLM Serving** (arXiv) - **Zhihe A210 octa-core RISC-V SoC with 12 TOPS NPU powers SoM-based development board - CNX Software** (google_riscv) - **Using a Performance Model to Implement a Superscalar CVA6** (riscv_news) - **Open-Source RISC-V Platform Trains Chip Designers From RTL To Silicon (ETH Z., lowRISC, U of Bologna)** (semiengineering) --- *Sponsored by LimitLess AI*
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.