AMAZINGINDEX.COM 日报快照
41.1
VOL. 2026.08
2026.08.30
← 返回 2026.08.30 日报
日报快照 · Daily Snapshot
NO. 011

三星把算力塞进内存芯片

#ARTICLE HackerNews 2026.08.30
推荐指数 85.0 NO. 011 · 2026.08.30
发布2026/08/29Score222Comments77

三星在LPDDR5X内存中集成MAC计算单元,无需修改内存控制器即可实现存内计算。这对AI推理的内存带宽瓶颈是实质性突破,边缘端大模型部署可能因此提速。

PIM最大的工程陷阱是内存和计算的热耦合问题,三星选择LPDDR5X而非HBM路线,说明他们优先瞄准手机、汽车等功耗敏感场景而非数据中心。这和其他玩家如SK海力士的HBM-PIM形成差异化。

值得警惕的是,存内计算的编程模型至今没有统一标准,三星保留标准内存控制器接口是聪明妥协,但开发者仍需面对数据划分和同步的新复杂度。做端侧推理框架的团队应该开始评估PIM-aware的算子调度策略。

意见分歧 71 条评论

核心争论:PIM技术能否突破LLM内存带宽瓶颈,还是缺乏杀手级应用和算法适配

reliabilityguy

Interesting that Samsung still pursues PIM. IIRC they had a paper in ISCA21 or 22 where they showed HBM2 module with PIM, which back then impressed me quite a lot. That being said, I am not sure what’s the killer application for this technology, and without such application adoption is unlikely.

xyzzy123

As I understand it, the killer app is llms. You could run MACs directly in RAM, offloading a lot of work from CPU and cutting down on insane (external) memory bandwidth required. Imagine (this is a fantasy pitch but potentially achievable for some use cases) wanting to run a larger llm and all you h

embedding-shape

> Imagine (this is a fantasy pitch but potentially achievable for some use cases) wanting to run a larger llm and all you have to do is buy more RAM so it fits. Isn't this how it works today already? Granted you wanted to run it on RAM rather than VRAM.

替代方案: 常规DRAM外部访问SRAM加速器HBM2-PIM标准x86内存控制器rep macb指令方案
查看原文 →