英伟达将推出新推理芯片

Core Viewpoint - The recent significant drop in the KOSPI index, attributed to global risk aversion due to geopolitical tensions and concerns over Nvidia's new SRAM-based inference chip, may have been misinterpreted regarding its impact on existing memory technologies [1][2]. Group 1: Market Reaction - The KOSPI index experienced a decline of over 10% in two days, marking the largest drop since 2008, primarily driven by global risk aversion following military actions involving Iran [1]. - A strong rebound was observed in the Korean stock market, with the KOSPI index increasing by 11%, and major tech companies like Samsung Electronics and SK Hynix saw significant gains of 13% and 15%, respectively [1]. Group 2: SRAM Chip Analysis - Reports indicated that Nvidia is developing a new inference chip using Groq's SRAM architecture, which raised concerns about reduced demand for existing memory types like HBM [1]. - KIS analysts clarified that the perception that low-cost SRAM chips would diminish the use of HBM reflects a misunderstanding of memory technologies, as SRAM has a larger unit area and higher cost per bit compared to DRAM [2]. Group 3: Memory Technology Diversification - SRAM architecture is not a replacement for DRAM but offers distinct advantages for specific applications requiring ultra-low latency, such as AI edge applications and certain data center workloads [5]. - The adoption of SRAM-based architectures is expected to diversify memory hierarchies within AI infrastructure, with HBM and DRAM continuing to serve as primary memory for large-scale model training and general inference servers [5][6]. - The overall potential market for the memory industry is anticipated to expand as the memory hierarchy becomes increasingly multi-tiered, incorporating SRAM, HBM, and DRAM [6].