HBM
高带宽内存（High Bandwidth Memory，HBM）是一种基于三维堆叠技术的高性能内存标准，通过将多层DRAM芯片垂直堆叠并以硅通孔（TSV）互联，实现极高的数据传输带宽和能效比。HBM主要用于对内存带宽要求极高的场景，包括人工智能训练、高性能计算及图形处理等领域，是现代GPU和AI加速芯片的重要组成部分。
核心事实
时间轴 (近 90 天)
现代存储金字塔中GPU显存(HBM)带宽可达数TB/s但消费级容量通常8-24GB，CPU内存(DDR5)带宽约50-100GB/s，NVMe固态硬盘带宽约5-14GB/s
HBM通过TSV（硅通孔）垂直互连多层DRAM，HBM3为8-12层，TSV是直径约5-10微米的铜柱
Traditional GPU memory relies primarily on HBM, which achieves high bandwidth through 3D stacking and ultra-wide buses but has limited capacity, high expense, and significant power consumption
Storage-class memory exhibits notable disadvantages in access latency and bandwidth compared to HBM's multi-terabyte-per-second bandwidth
SRAM访问延迟约1ns，HBM约100ns，SRAM延迟比HBM低约10-50倍
英伟达H100采用的HBM3内存带宽高达3.35TB/s
轻薄本内存优先选LPDDR5X 7500MHz及以上且容量≥32GB，核显本尤其吃内存带宽，16GB双通道下核显性能比32GB低约15%，且32GB才能满足Windows+浏览器多标签+虚拟机等场景
HBM芯片目前主要由SK海力士和三星供应，产能扩张速度远跟不上需求增长
全部知识事实 (8)
英伟达H100采用的HBM3内存带宽高达3.35TB/s
85%待验证Traditional GPU memory relies primarily on HBM, which achieves high bandwidth through 3D stacking and ultra-wide buses but has limited capacity, high expense, and significant power consumption
90%待验证Storage-class memory exhibits notable disadvantages in access latency and bandwidth compared to HBM's multi-terabyte-per-second bandwidth
85%待验证HBM芯片目前主要由SK海力士和三星供应，产能扩张速度远跟不上需求增长
85%部分验证轻薄本内存优先选LPDDR5X 7500MHz及以上且容量≥32GB，核显本尤其吃内存带宽，16GB双通道下核显性能比32GB低约15%，且32GB才能满足Windows+浏览器多标签+虚拟机等场景
65%待验证现代存储金字塔中GPU显存(HBM)带宽可达数TB/s但消费级容量通常8-24GB，CPU内存(DDR5)带宽约50-100GB/s，NVMe固态硬盘带宽约5-14GB/s
50%待验证HBM通过TSV（硅通孔）垂直互连多层DRAM，HBM3为8-12层，TSV是直径约5-10微米的铜柱
50%待验证SRAM访问延迟约1ns，HBM约100ns，SRAM延迟比HBM低约10-50倍
50%