society […] (countable) The sum total of all voluntary interrelations between individuals. — https://en.wiktionary.org/wiki/society (retrieved on 2026-08-30) (I’m going to use the shorthand “I received” to mean “I was a participant in a conversation in which…

JSON is a standard format for data interchange. It is effectively a tiny subset of JavaScript made of objects and arrays. It looks as follows {"key":1, "text":[1.0,2.0]}. Many programming languages include a JSON library in their standard libraries: C#, Go…

In-memory compute has been an attractive proposition for many years because compute within a memory chip can exploit its higher internal bandwidth. Additionally, in-memory compute avoids the long latency path between DRAM and traditional compute cores. At…

最近尝试用 LLM 进行一些知识的学习，也会在博客记录一些学习的过程，但发现抛开代码后， LLM 的回答的用词有时候非常奇怪，甚至有些不符合人类的表达习惯。 原理 我们以 DeepSeek v4 Flash 为例，来看看它的模型结构，其他 Transformer 模型的结构也大同小异： flowchart TB IN(["input_ids"]) --> EMB["embed 输入词表&LTbr/>129,280 × 4,096"] EMB --> BLK["Block × 43&LTbr/>MLA 稀疏注意力…

Airbus A330-200 Prestige with registration VQ-BIG (ex VP-BHD) airframe details and operator history including first flight and delivery dates, seat configurations, engines, fleet numbers and names

The size of a processor's page-table entries directly limits how much physical memory that proc [...]

This website contains a preview of how Intel intends to document and present the Intel® Architecture in the future. Our goal is a specification that is more accurate and accessible. We would like to receive feedback on how well we meet this goal before we…

Hello you fine Internet folks,At Hot Chips, Intel talked a bit more about their upcoming Datacenter GPU codenamed Crescent Island. Crescent Island’s party piece is the ultra-high memory capacity curtesy of the LPDDR5X memory subsystem. Making Room for Whales…

背景 最近拿到了一台 4 卡 RTX 4090 的机器，但用的是一代至强可扩展（Skylake）+ 技嘉 C621-WD12 主板。我们直接把显卡装上虽然可以使用，但是缺少了 P2P 功能对 LLM 推理跑 TP 不友好，因此我进行了一番折腾。 Resizable Bar 首先，这个主板的 BIOS 默认甚至打不开 Above 4GB Decoding，这意味着所有 PCIe Bar 都会被分配在 Sub 4GB 物理地址空间，而我们一张卡的显存都有 24G，那自然不可能被分配出来。 后来让 5.6 Sol…

Fujitsu’s A64FX was a pivotal core. For the Arm ecosystem, it introduced SVE. SVE continues to define Arm’s vector execution strategy to this day, with newer extensions like SME adopting a similar vector length agnostic approach. For Fujitsu, A64FX marked…

Apple debuted M6 in the new Mac mini and M5 Ultra in the new Mac Studio, providing an extraordinary leap in performance and AI capabilities.

Hello you fine Internet folks, Intel unveiled their upcoming Diamond Rapids Server CPU at Computex where they said that Diamond Rapids has twice the memory bandwidth, PCIe Gen 6, and 50% more cores than Granite Rapids. However, it looks like in the 3 or so…