Kevin Di's Hatch

Kevin Di
Latest
Popular
Glasp
Hatch
Why LLM Speed Is No Longer a Software Problem, but a Physics Problem
glasp.co/hatch

Why LLM Speed Is No Longer a Software Problem, but a Physics Problem

Hatched on May 26, 2026 · 3 views

Glasp
Hatch
AI 时代真正的分界线,不是算力,而是如何把算力拆开
glasp.co/hatch

AI 时代真正的分界线,不是算力,而是如何把算力拆开

Hatched on May 25, 2026 · 4 views

Glasp
Hatch
为什么 AI 芯片的真正瓶颈,不在算力,而在把信息“送对地方”
glasp.co/hatch

为什么 AI 芯片的真正瓶颈,不在算力,而在把信息“送对地方”

Hatched on May 24, 2026 · 6 views

Glasp
Hatch
为什么系统设计里最危险的错误,是把“更快的连接”当成“更好的分工”
glasp.co/hatch

为什么系统设计里最危险的错误,是把“更快的连接”当成“更好的分工”

Hatched on May 23, 2026 · 3 views

Glasp
Hatch
算力不是越堆越强,真正的胜负在“可重构性”
glasp.co/hatch

算力不是越堆越强,真正的胜负在“可重构性”

Hatched on May 22, 2026 · 2 views

Glasp
Hatch
真正的性能优化,不是让模型更聪明,而是让它记住得更少
glasp.co/hatch

真正的性能优化,不是让模型更聪明,而是让它记住得更少

Hatched on May 21, 2026 · 7 views

Glasp
Hatch
为什么真正的芯片革命,先发生在“看不见的瓶颈”上
glasp.co/hatch

为什么真正的芯片革命,先发生在“看不见的瓶颈”上

Hatched on May 20, 2026 · 6 views

Glasp
Hatch
当算力变得更强,真正稀缺的其实是“可组合性”
glasp.co/hatch

当算力变得更强,真正稀缺的其实是“可组合性”

Hatched on May 19, 2026 · 4 views

Glasp
Hatch
真正的AI推理竞赛,不是算力竞赛,而是把数据搬得更快
glasp.co/hatch

真正的AI推理竞赛,不是算力竞赛,而是把数据搬得更快

Hatched on May 18, 2026 · 3 views

Glasp
Hatch
为什么 AI 的真正瓶颈不是算力,而是谁拥有“内存的主权”
glasp.co/hatch

为什么 AI 的真正瓶颈不是算力,而是谁拥有“内存的主权”

Hatched on May 17, 2026 · 8 views

Glasp
Hatch
算力的真正竞争,不在峰值,而在边界如何被重新定义
glasp.co/hatch

算力的真正竞争,不在峰值,而在边界如何被重新定义

Hatched on May 16, 2026 · 8 views

Glasp
Hatch
真正决定 AI 集群上限的,不是算力,而是互联的可重构性
glasp.co/hatch

真正决定 AI 集群上限的,不是算力,而是互联的可重构性

Hatched on May 15, 2026 · 4 views

Glasp
Hatch
语言模型真正的瓶颈,不是“会不会想”,而是“能不能把注意力放对地方”
glasp.co/hatch

语言模型真正的瓶颈,不是“会不会想”,而是“能不能把注意力放对地方”

Hatched on May 14, 2026 · 4 views

Glasp
Hatch
AI芯片真正的战场,不是算力,而是拓扑的消失
glasp.co/hatch

AI芯片真正的战场,不是算力,而是拓扑的消失

Hatched on May 13, 2026 · 6 views

Glasp
Hatch
LLM 推理的真正难题,不是算力,而是配平
glasp.co/hatch

LLM 推理的真正难题,不是算力,而是配平

Hatched on May 12, 2026 · 6 views

Glasp
Hatch
为什么更强的模型,常常先被带宽打败
glasp.co/hatch

为什么更强的模型,常常先被带宽打败

Hatched on May 11, 2026 · 5 views

Glasp
Hatch
AI推理的真正瓶颈,不在模型里,而在数据中心的最后一公里
glasp.co/hatch

AI推理的真正瓶颈,不在模型里,而在数据中心的最后一公里

Hatched on May 10, 2026 · 7 views

Glasp
Hatch
真正稀缺的不是芯片,而是把芯片组织成可交付系统的能力
glasp.co/hatch

真正稀缺的不是芯片,而是把芯片组织成可交付系统的能力

Hatched on May 9, 2026 · 5 views

Glasp
Hatch
为什么大模型的极限,不是算力,而是通信与等待
glasp.co/hatch

为什么大模型的极限,不是算力,而是通信与等待

Hatched on May 8, 2026 · 6 views

Glasp
Hatch
当算力帝国撞上文本出口:真正的瓶颈不在芯片,而在系统的最后一米
glasp.co/hatch

当算力帝国撞上文本出口:真正的瓶颈不在芯片,而在系统的最后一米

Hatched on May 7, 2026 · 6 views

Glasp
Hatch
Why Efficient AI Is Really a Story About Memory, Not Just Speed
glasp.co/hatch

Why Efficient AI Is Really a Story About Memory, Not Just Speed

Hatched on May 6, 2026 · 4 views

Glasp
Hatch
当 AI 芯片开始卖网络时,真正的战场已经不是算力
glasp.co/hatch

当 AI 芯片开始卖网络时,真正的战场已经不是算力

Hatched on May 5, 2026 · 5 views

Glasp
Hatch
为什么大模型推理的真正瓶颈,不是算力,而是“搬运记忆”
glasp.co/hatch

为什么大模型推理的真正瓶颈,不是算力,而是“搬运记忆”

Hatched on May 4, 2026 · 7 views

Glasp
Hatch
算力不是瓶颈,真正的瓶颈是“带宽与缓存的政治学”
glasp.co/hatch

算力不是瓶颈,真正的瓶颈是“带宽与缓存的政治学”

Hatched on May 3, 2026 · 5 views

Glasp
Hatch
真正决定 AI 胜负的,不是算力,而是把算力变成可流动的能力
glasp.co/hatch

真正决定 AI 胜负的,不是算力,而是把算力变成可流动的能力

Hatched on May 2, 2026 · 7 views

Glasp
Hatch
为什么大模型的性能瓶颈,最后都会变成调度问题
glasp.co/hatch

为什么大模型的性能瓶颈,最后都会变成调度问题

Hatched on May 1, 2026 · 4 views

Glasp
Hatch
GPU泡沫退潮之后,真正值钱的不是算力,而是可编程的稀缺性
glasp.co/hatch

GPU泡沫退潮之后,真正值钱的不是算力,而是可编程的稀缺性

Hatched on Apr 30, 2026 · 9 views

Glasp
Hatch
为什么最贵的系统故障,往往卡在最不起眼的转换环节
glasp.co/hatch

为什么最贵的系统故障,往往卡在最不起眼的转换环节

Hatched on Apr 29, 2026 · 5 views

Glasp
Hatch
Why the Fastest LLMs Win by Remembering Less
glasp.co/hatch

Why the Fastest LLMs Win by Remembering Less

Hatched on Apr 28, 2026 · 4 views

Glasp
Hatch
算力不是机器,是一张负债表:为什么真正值钱的是可持续的计算能力
glasp.co/hatch

算力不是机器,是一张负债表:为什么真正值钱的是可持续的计算能力

Hatched on Apr 27, 2026 · 8 views

Glasp
Hatch
从随机采样到万卡网络:大模型真正稀缺的不是算力,而是秩序
glasp.co/hatch

从随机采样到万卡网络:大模型真正稀缺的不是算力,而是秩序

Hatched on Apr 26, 2026 · 9 views

Glasp
Hatch
真正决定AI芯片胜负的,不是算力,而是“路由”
glasp.co/hatch

真正决定AI芯片胜负的,不是算力,而是“路由”

Hatched on Apr 25, 2026 · 10 views

Glasp
Hatch
真正的智能,不在算力峰值,而在把记忆放在离思考最近的地方
glasp.co/hatch

真正的智能,不在算力峰值,而在把记忆放在离思考最近的地方

Hatched on Apr 24, 2026 · 10 views

Glasp
Hatch
AI 训练的真正瓶颈,不是算力,而是每一次“搬运”
glasp.co/hatch

AI 训练的真正瓶颈,不是算力,而是每一次“搬运”

Hatched on Apr 23, 2026 · 8 views

Glasp
Hatch
为什么最好的算力,不是“更快”,而是“更贴近真实任务的尺寸”
glasp.co/hatch

为什么最好的算力,不是“更快”,而是“更贴近真实任务的尺寸”

Hatched on Apr 22, 2026 · 8 views

Glasp
Hatch
当并行不再只追求更快,而是先找到它真正该服务的任务
glasp.co/hatch

当并行不再只追求更快,而是先找到它真正该服务的任务

Hatched on Apr 21, 2026 · 5 views

Glasp
Hatch
AI 竞赛的真正瓶颈,不是算力,而是精度该放在哪里
glasp.co/hatch

AI 竞赛的真正瓶颈,不是算力,而是精度该放在哪里

Hatched on Apr 20, 2026 · 5 views

Glasp
Hatch
算力战争真正打的不是芯片,而是控制权
glasp.co/hatch

算力战争真正打的不是芯片,而是控制权

Hatched on Apr 19, 2026 · 3 views

Glasp
Hatch
当算力从稀缺变成拥堵,真正的瓶颈就不再是 GPU
glasp.co/hatch

当算力从稀缺变成拥堵,真正的瓶颈就不再是 GPU

Hatched on Apr 18, 2026 · 9 views

Glasp
Hatch
当云平台遇见高速网络:真正决定AI集群上限的不是算力,而是编排
glasp.co/hatch

当云平台遇见高速网络:真正决定AI集群上限的不是算力,而是编排

Hatched on Apr 17, 2026 · 7 views

Glasp
Hatch
当记忆成为瓶颈: 把序列拆成流水的时机与方法
glasp.co/hatch

当记忆成为瓶颈: 把序列拆成流水的时机与方法

Hatched on Apr 16, 2026 · 7 views

Glasp
Hatch
当显存追不上参数:把网络当作内存,重构大模型的计算拓扑
glasp.co/hatch

当显存追不上参数:把网络当作内存,重构大模型的计算拓扑

Hatched on Apr 15, 2026 · 7 views

Glasp
Hatch
当算法遇见场景: 找到机器学习工程的真正适配点
glasp.co/hatch

当算法遇见场景: 找到机器学习工程的真正适配点

Hatched on Apr 14, 2026 · 8 views

Glasp
Hatch
### Unlocking the Potential of Token-Level Pipeline Parallelism and NVLink Technology
glasp.co/hatch

### Unlocking the Potential of Token-Level Pipeline Parallelism and NVLink Technology

Hatched on Apr 13, 2026 · 5 views

Glasp
Hatch
# Accelerating the Development of Domestic AI Chips: Bridging the Gap with Advanced Models
glasp.co/hatch

# Accelerating the Development of Domestic AI Chips: Bridging the Gap with Advanced Models

Hatched on Apr 12, 2026 · 5 views

Glasp
Hatch
# The Future of Chip Technology: Navigating the Landscape of UCIE and NVLink
glasp.co/hatch

# The Future of Chip Technology: Navigating the Landscape of UCIE and NVLink

Hatched on Apr 11, 2026 · 6 views

Glasp
Hatch
Understanding Dynamic Inference and GPU Architecture: A Deep Dive into AI Technologies
glasp.co/hatch

Understanding Dynamic Inference and GPU Architecture: A Deep Dive into AI Technologies

Hatched on Apr 10, 2026 · 4 views

Glasp
Hatch
# The Race for High Bandwidth Memory: Transforming AI and Computing Performance
glasp.co/hatch

# The Race for High Bandwidth Memory: Transforming AI and Computing Performance

Hatched on Apr 9, 2026 · 7 views

Glasp
Hatch
### 未来内存和计算架构的变革:MRAM与CXL和GB200的潜力
glasp.co/hatch

### 未来内存和计算架构的变革:MRAM与CXL和GB200的潜力

Hatched on Apr 8, 2026 · 8 views

Glasp
Hatch
### Accelerating Transformer Models: Insights from Sparse Attention and NVLink Technologies
glasp.co/hatch

### Accelerating Transformer Models: Insights from Sparse Attention and NVLink Technologies

Hatched on Apr 7, 2026 · 5 views

Glasp
Hatch
The Rise of Next-Generation AI Chips: A New Era in Large Language Model Inference
glasp.co/hatch

The Rise of Next-Generation AI Chips: A New Era in Large Language Model Inference

Hatched on Apr 6, 2026 · 4 views

Glasp
Hatch
### The Evolution of Large Language Models: A Comparative Analysis and Optimization Strategies
glasp.co/hatch

### The Evolution of Large Language Models: A Comparative Analysis and Optimization Strategies

Hatched on Apr 5, 2026 · 8 views

Glasp
Hatch
Cerebras vs. NVIDIA: The Race for AI Chip Supremacy
glasp.co/hatch

Cerebras vs. NVIDIA: The Race for AI Chip Supremacy

Hatched on Apr 4, 2026 · 11 views

Glasp
Hatch
Navigating the Challenges of Blackwell: Insights into AIGC Network Optimization
glasp.co/hatch

Navigating the Challenges of Blackwell: Insights into AIGC Network Optimization

Hatched on Apr 3, 2026 · 4 views

Glasp
Hatch
### The Future of AI Processing: Unleashing the Power of Innovative Chip Technologies
glasp.co/hatch

### The Future of AI Processing: Unleashing the Power of Innovative Chip Technologies

Hatched on Apr 2, 2026 · 6 views

Glasp
Hatch
# Optimizing Performance in Machine Learning Inference: Insights into KVCache and Fully-Connected Layers
glasp.co/hatch

# Optimizing Performance in Machine Learning Inference: Insights into KVCache and Fully-Connected Layers

Hatched on Apr 1, 2026 · 4 views

Glasp
Hatch
### The Future of Computing: Innovations in CXL and AI Networking Solutions
glasp.co/hatch

### The Future of Computing: Innovations in CXL and AI Networking Solutions

Hatched on Mar 31, 2026 · 3 views

Glasp
Hatch
# Revolutionizing AI Hardware: The Future of Open Design and Unified Memory Architecture
glasp.co/hatch

# Revolutionizing AI Hardware: The Future of Open Design and Unified Memory Architecture

Hatched on Mar 30, 2026 · 6 views

Glasp
Hatch
### The Evolution of AI Chips: Innovations and Implications for the Future
glasp.co/hatch

### The Evolution of AI Chips: Innovations and Implications for the Future

Hatched on Mar 29, 2026 · 5 views

Glasp
Hatch
Pioneering the Future of AI: Insights from Chip Development and Performance Optimization
glasp.co/hatch

Pioneering the Future of AI: Insights from Chip Development and Performance Optimization

Hatched on Mar 28, 2026 · 6 views

Next >