AI HN来自 Hacker News 的 AI 新闻
EN
今天
3天
7天
30天
全部
29 · "H100"
每页
1
After the Bubble(tbray.org)
91 ·savant2·17 天前·82 评论
AI Chips
本文分析了生成式AI泡沫即将破裂的问题,重点指出GPU的脆弱性(如Llama 3训练期间Nvidia H100的故障)和高功耗成本是关键因素。文章提到,与过去的泡沫(铁路、互联网泡沫)破裂后留下有价值基础设施不同,由于GPU损耗快和能源成本高,生成式AI泡沫破裂后可能不会留下类似的长期价值,并指出特殊目的实体(SPVs)是大型科技公司在不增加资产负债表债务的情况下建设AI数据中心的财务手段。
2
Are we repeating the telecoms crash with AI datacenters?(martinalderson.com)
241 ·davedx·23 天前·194 评论
本文将AI数据中心的繁荣与2000年代电信业崩溃进行对比,强调供需动态差异。与电信业光纤容量的指数级提升不同,AI GPU的每瓦性能增长正在放缓,而GPU功耗(如Blackwell B200的1000-1200W TDP)却大幅上升,表明AI基础设施具有独特的基本面。
3
US startup Substrate announces chipmaking tool that it says will rival ASML(reuters.com)
89 ·outrun86·大约 2 个月前·71 评论
AI Chips
美国初创公司Substrate宣布推出一款芯片制造工具,声称可与ASML的产品竞争,而ASML的工具对于生产AI系统所用的高端半导体至关重要。
4
Diamond Thermal Conductivity: A New Era in Chip Cooling(spectrum.ieee.org)
78 ·rbanffy·2 个月前·43 评论
AI Chips
本文探讨了钻石热导率作为芯片冷却的新方案,这是包括高性能GPU等AI硬件在内的先进计算系统的关键需求。
5
America’s semiconductor boom [video](youtube.com)
191 ·zdw·2 个月前·136 评论
AI Chips
这篇报道聚焦美国的半导体热潮,强调其与AI的相关性,因为半导体在驱动AI计算和相关硬件系统方面发挥着关键作用。
6
Processing Strings 109x Faster Than Nvidia on H100(ashvardanian.com)
216 ·ashvardanian·3 个月前·26 评论
AI Chips
StringZilla v4是首个支持CUDA的SIMD优先字符串处理库版本,现已发布。它提供快速的编辑距离计算(500+ GigaCUPS),引入了基于AES的哈希和52位MinHash等新哈希函数,并支持信息检索、数据库和生物信息学等大规模工作负载。该库基于Apache 2.0开源协议,可通过pip安装。
7
Launch HN: RunRL (YC X25) – Reinforcement learning as a service(runrl.com)
71 ·ag8·3 个月前·22 评论
本文介绍了由YC支持的强化学习即服务平台RunRL。该平台允许用户使用自定义奖励函数优化AI模型,集成OpenAI、Anthropic等现有AI API,并获取H100 GPU等训练所需的计算资源。平台为研究者和开发者提供SDK,同时有自助和企业级定价选项。
8
Deploying DeepSeek on 96 H100 GPUs(lmsys.org)
285 ·GabrielBianconi·4 个月前·80 评论
AI ChipsOpen Source AI
LMSYS团队使用SGLang在96块H100 GPU(12个节点×8)上部署了DeepSeek大语言模型,采用预填充-解码分离和大规模专家并行技术。该实现达到了高吞吐量(对于2000 token输入,每个节点每秒处理52.3k输入token和22.3k输出token),性能与DeepSeek官方报告相当,成本仅为其API的五分之一,且完全开源并提供可复现的实验指导。
9
The Future of Compute: Nvidia's Crown Is Slipping(mohitdagarwal.substack.com)
144 ·wilson090·8 个月前·120 评论
AI Chips
💡 The story discusses Nvidia's slipping dominance in the compute market, which is directly linked to AI hardware (e.g., GPUs like H100 used for AI training/inference), aligning with the hardware category.
10
The Tiny Star Explosions Powering Moore's Law(spectrum.ieee.org)
131 ·mcharawi·10 个月前·13 评论
AI Chips
💡 The story discusses EUV light source, a critical technology for manufacturing advanced semiconductors (e.g., Nvidia H100) that power AI compute systems.
11
Huawei's Ascend 910C delivers 60% of Nvidia H100 inference performance(tomshardware.com)
118 ·sien·11 个月前·61 评论
AI Chips
💡 The story focuses on Huawei's Ascend 910C AI chip and its inference performance relative to Nvidia's H100, which directly falls under the hardware category covering AI chips and compute.
12
CUDA Moat Still Alive(semianalysis.com)
221 ·pella·大约 1 年前·172 评论
AI Chips
💡 The story discusses benchmarks of AI chips (Nvidia H100/H200, AMD MI300x) and CUDA's competitive advantage, which are core to AI hardware and compute.
13
Exploring inference memory saturation effect: H100 vs. MI300x(dstack.ai)
57 ·latchkey·大约 1 年前·12 评论
AI Chips
💡 The story explores inference memory saturation effects by comparing Nvidia H100 and AMD MI300x chips, which are critical AI hardware components for compute and inference tasks.
14
U.S. chip revival plan chooses sites(spectrum.ieee.org)
177 ·pseudolus·大约 1 年前·127 评论
AI Chips
💡 The story about the U.S. chip revival plan choosing sites relates to semiconductor manufacturing infrastructure, which is critical for AI compute (e.g., chips like Nvidia H100 used in AI data centers).
15
Ultraprecise method of aligning 3D semiconductor chips invented(techxplore.com)
170 ·thebeardisred·大约 1 年前·24 评论
AI ChipsAI Safety
💡 The story focuses on an ultraprecise method for aligning 3D semiconductor chips, which are key components in AI hardware like Nvidia's H100 (using 3D stacking). This directly relates to AI-related chips and compute, fitting the hardware category.
16
$2 H100s: How the GPU Rental Bubble Burst(latent.space)
403 ·swyx·大约 1 年前·279 评论
AI Chips
💡 The story focuses on the bursting of the GPU rental bubble involving Nvidia H100s, which are AI-specific compute chips, falling under the hardware category.
17
Taiwan is heading toward an energy crunch?(wired.com)
82 ·vunderba·大约 1 年前·119 评论
AI Chips
💡 The story focuses on Taiwan's energy crunch affecting its computer chip production, which includes AI-critical chips (e.g., Nvidia H100) essential for AI compute infrastructure, aligning with the hardware category covering chips and compute-related issues.
18
Japan on edge of EUV lithography chip-making revolution(asiatimes.com)
183 ·ksec·超过 1 年前·97 评论
AI Chips
💡 The story discusses Japan's progress in EUV lithography for chip-making, a critical technology for producing advanced AI chips (like Nvidia H100) essential for AI compute.
19
NVIDIA Transitions Fully Towards Open-Source Linux GPU Kernel Modules(developer.nvidia.com)
881 ·shaicoleman·超过 1 年前·254 评论
AI Chips
💡 The story discusses NVIDIA's shift to fully open-source Linux GPU kernel modules, which are integral to their AI-critical GPU hardware (e.g., H100) that powers AI training and inference.
20
Karpathy: Let's reproduce GPT-2 (1.6B): one 8XH100 node 24h $672 in llm.c(github.com)
182 ·alecco·超过 1 年前·58 评论
AI ChipsInference Optimization
💡 The story centers on the practical details of reproducing GPT-2 using llm.c, including required compute resources (8XH100 node), time taken (24h), and cost ($672)—all falling under AI training infrastructure.
21
So you want to rent an NVIDIA H100 cluster? 2024 Consumer Guide(photoroom.com)
297 ·ea016·超过 1 年前·139 评论
AI Chips
💡 The story focuses on renting NVIDIA H100 clusters, which are key AI hardware components for compute-intensive tasks. This falls under the hardware category covering chips and compute.
22
ASML Aims for Hyper-NA EUV, Shrinking Chip Limits(eetimes.com)
103 ·paulbaumgart·超过 1 年前·65 评论
AI Chips
💡 The story focuses on ASML's Hyper-NA EUV lithography technology, which is critical for manufacturing advanced semiconductors used in AI hardware (e.g., GPUs like Nvidia H100). This directly supports AI compute infrastructure.
23
AMD's MI300X Outperforms Nvidia's H100 for LLM Inference(blog.tensorwave.com)
280 ·fvv·超过 1 年前·264 评论
AI Chips
💡 The story discusses AMD's MI300X and Nvidia's H100, which are AI-specific chips designed for LLM inference, directly fitting the hardware category.
24
TSMC Has a Kill Switch in the event of a Chinese invasion(pcgamer.com)
56 ·cwwc·超过 1 年前·32 评论
AI Chips
💡 The story discusses TSMC's kill switch for its chip manufacturing machines, which are critical for producing AI chips like Nvidia H100. This falls under the hardware category as it relates to AI compute hardware production.
25
Electromigration Concerns Grow in Advanced Packages(semiengineering.com)
58 ·PaulHoule·超过 1 年前·40 评论
AI Chips
💡 The story discusses electromigration concerns in advanced semiconductor packages, which are critical components of high-performance AI chips (e.g., Nvidia H100) used for AI compute tasks.
26
TSMC's debacle in the desert: Missed deadlines and tension among coworkers(restofworld.org)
91 ·impish9208·超过 1 年前·83 评论
AI Chips
💡 The story discusses operational issues at TSMC's Arizona chip manufacturing plant, which produces chips essential for AI compute systems (e.g., Nvidia's H100), directly linking to AI hardware production.
27
Intel Gaudi2 chips outperform Nvidia H100 on diffusion transformers(stability.ai)
146 ·memossy·将近 2 年前·60 评论
AI ChipsImage Generation
💡 The story compares the performance of Intel Gaudi2 and Nvidia H100 chips (AI-specific hardware) on diffusion transformers, which falls under the hardware category covering AI chips and compute.
28
Inside the miracle of modern chip manufacturing(ig.ft.com)
125 ·pseudolus·将近 2 年前·23 评论
AI Chips
💡 The story focuses on modern chip manufacturing, which likely includes AI-specific chips like Nvidia H100 or AMD MI300, fitting the hardware category about AI chips and compute.
29
TSMC is having more luck building in Japan than in America(economist.com)
192 ·helsinkiandrew·将近 2 年前·297 评论
AI Chips
💡 The story focuses on TSMC's fab construction efforts, and TSMC is a critical manufacturer of chips used in AI accelerators (e.g., Nvidia H100), aligning with the hardware category for AI-related chips and compute.
📅周报
Hacker News|Powered by Doubao