Core Insights - Nvidia's introduction of the LPU (Language Processing Unit) in its new chip architecture signifies a shift in the AI computing market from "general computing" to "specialized inference" [1][2] - The demand for PCBs (Printed Circuit Boards) is expected to increase significantly due to the LPU's requirements, leading to higher complexity in PCB manufacturing [1][3] Group 1: LPU Characteristics and Impact - LPU is designed for AI inference, particularly for low-latency real-time interactions, achieving a latency of around 100 milliseconds and outperforming H100 GPUs by approximately 10 times in inference speed [1][2] - The LPU's architecture, which includes high-speed on-chip SRAM with a bandwidth of up to 80TB/s, addresses memory bottlenecks, enhancing overall energy efficiency by about 10 times [1][2] Group 2: PCB Demand and Manufacturing Challenges - The implementation of LPU will require a significant increase in PCB area, as multiple LPUs will need to be interconnected due to the 230MB SRAM capacity limitation of a single LPU [3] - The materials used for PCBs will need to be upgraded to meet the higher signal transmission efficiency requirements, likely involving advanced materials such as 52-layer M9-grade copper-clad laminates [3] Group 3: Advanced Packaging and Equipment Demand - Nvidia's PD (Prefill-Decode) separation technology aims to optimize the deployment of GPUs and LPUs, addressing challenges related to server layout and wiring density [3] - The hybrid architecture of GPU and LPU will necessitate higher precision in packaging technology, increasing the demand for high-precision assembly equipment in PCB manufacturing [3] Group 4: Investment Opportunities - Companies involved in high-precision drilling and exposure processes, such as Dazhong CNC (301200.SZ), Inno Laser (301021.SZ), and Dier Laser (300776.SZ), are recommended for investment [4] - Firms specializing in high-precision assembly equipment, like Kaige Precision Machinery (301338.SZ) and Jintuo Co., Ltd. (300400.SZ), are also highlighted as potential investment targets [4] - High-end PCB drill manufacturers, including Dingtai High-Tech (301377.SZ) and World (688028.SH), are suggested for consideration [4]
光大证券:英伟达(NVDA.US)计划引入LPU方案 AI推理有望延伸至PCB设备领域