惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

K
Kaspersky official blog
Engineering at Meta
Engineering at Meta
D
DataBreaches.Net
Stack Overflow Blog
Stack Overflow Blog
Microsoft Security Blog
Microsoft Security Blog
Y
Y Combinator Blog
B
Blog RSS Feed
GbyAI
GbyAI
P
Proofpoint News Feed
aimingoo的专栏
aimingoo的专栏
MyScale Blog
MyScale Blog
D
Docker
阮一峰的网络日志
阮一峰的网络日志
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Recorded Future
Recorded Future
美团技术团队
The Register - Security
The Register - Security
V
Visual Studio Blog
H
Hackread – Cybersecurity News, Data Breaches, AI and More
T
Tailwind CSS Blog
爱范儿
爱范儿
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
T
The Blog of Author Tim Ferriss
博客园 - 司徒正美
量子位
B
Blog
F
Fortinet All Blogs
Martin Fowler
Martin Fowler
博客园 - 【当耐特】
MongoDB | Blog
MongoDB | Blog
A
About on SuperTechFans
I
InfoQ
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
有赞技术团队
有赞技术团队
雷峰网
雷峰网
大猫的无限游戏
大猫的无限游戏
J
Java Code Geeks
L
LangChain Blog
Latest news
Latest news
S
SegmentFault 最新的问题
Exploit-DB.com RSS Feed
Exploit-DB.com RSS Feed
Blog — PlanetScale
Blog — PlanetScale
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
Cisco Talos Blog
Cisco Talos Blog
F
Full Disclosure
C
Cisco Blogs
D
Darknet – Hacking Tools, Hacker News & Cyber Security
W
WeLiveSecurity
T
Tenable Blog
T
Tor Project blog

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant Common SOC 2 Failures (Real World) Stop Vibe-Checking Your AI App: A Practical Guide to Evals How to Use SonarQube and SonarScanner Locally to Level Up Your Code Quality Your Next To-Do App Is Dead — I Replaced Mine with an OpenClaw AI Sign a Nostr event in 60 lines of Python using coincurve — no nostr-sdk, no nbxplorer, no rust toolchain ITGC Audit Explained Like You’re in Big 4 Patch Tuesday abril 2026: Microsoft parcha 163 vulnerabilidades y un zero-day en SharePoint Stop scraping everything: a better way to track competitor price changes Listing on MCPize + the Official MCP Registry while routing payments OUTSIDE the marketplace — how I kept 100% of my x402 revenue Building an AI-Powered Risk Intelligence System Using Serverless Architecture Why We Ripped Function Overloading Out of Our AI Toolchain Testing AI-Generated Code: How to Actually Know If It Works SaaS Churn Is Killing Your Business. Here Is What to Do About It (Without a Support Team) The Speed of AI Is No Longer Linear - And Self-Improving Models Are Why How to Implement RBAC for MCP Tools: A Practical Guide for Engineering Teams From Standard Quote to Persuasive Proposal: AI Automation for Arborists I built a CLI that scaffolds complete multi-tenant SaaS apps Axios CVE-2025–62718: The Silent SSRF Bug That Could Be Hiding in Your Node.js App Right Now The dashboard that ended our friendship Data Pipelines Explained Simply (and How to Build Them with Python) The Hidden Cost of AI Systems Nobody Talks About. undefined vs undeclared, and how typeof behaves Switching from file-based jobs to NATS/Kafka in Rust without changing code io_uring Adventures: Rust Servers That Love Syscalls Why Agentic AI is Killing the Traditional Database The POUR principles of web accessibility for developers and designers Quantum Neural Network 3D — A Deep Dive into Interactive WebGL Visualization How To Install Caveman In Codex On macOS And Windows Automation Pipeline Reliability: Why Your Workflow Breaks When Nobody Is Watching I Built an 'Open World' AI Coding Agent — It Works From ANY Folder From Freelancing to Product: A Tech Service Company's SaaS Transformation China's AI Giants: Adding Tencent Hunyuan & ByteDance Doubao to AI University (74 Providers) On the Vibe Coders and Their Lies clerk: Auto-Summarize Your Claude Code Sessions AI Weekly — 2026/04/10–04/17 | The Model Lockdown Is Here, but the Toolchain Is the Real Battleground AI 週報 — 2026/04/10–2026/04/17 模型封鎖潮來了,但工具鏈才是真戰場 Maybe this is how Open-Source apps are born... 🚀 Fine-Tune LLMs with LoRA and QLoRA: 2026 Guide tRPC v11 + Next.js App Router: End-to-End Type Safety Without the Boilerplate ShadCN UI in 2026: Why I Stopped Installing Component Libraries and Started Owning My Components SaaS Billing in React Server Components: Stripe + Supabase Without a Single `useEffect` Join our DEV Weekend Challenge — $1,000 in Prizes Across TEN winners! Submissions Due April 20 at 6:59 AM UTC. Implementing FSRS Spaced Repetition in Flutter + Supabase — Adding Memory Science to an AI Learning App "I Texted My Localhost From the Train — Claude Code Fixed the Bug Before I Got Home" I Built a Sales Prep AI and It Went Deeper Than Expected Design to Code #2: One JSON, Eleven Outputs Solving the 100M-Row Problem: A Summary Table Pattern for High-Volume Push Notification Logs Flutter Web With Wasm: What Actually Changes For Developers I Built 50 Royalty-Free Soundtracks for My Side Project in a Weekend Using AI Music Generation The Vibe Coding Security Checklist: 7 Things to Check Before You Ship Stop Letting Googlebot Guess Fix Your React App's SEO Right Desconstruindo o Streaming do LinkedIn: Como Criar um Engine de Extração de Vídeo de Alta Performance com HLS e FFmpeg (EDA Part-1) EDA (Exploratory Data Analysis) Explained With Real Life — Why Looking at Your Data Is the Most Important Step in Machine Learning Brand Relationship Management at Scale: Our 4-Touch Outreach System for 200+ Brands Why String.fromEnvironment() Might Return an Empty String in Dart JGuardrails 1.0.0 — Hardening Java LLM Apps Against Jailbreaks, Toxicity, and Prompt Injection Plan and Schedule a Full Week of Threads Content From One Claude Conversation Coding Cat Oran Ep3, Five Tables Changed Everything Updated: BFF Pattern I'm done watching freelancers get buried by 200 proposals. So I'm building the alternative. This is my first post BFS Algorithm in Java Step by Step Tutorial with Examples Tracking LLM Pricing Monthly: An Open Dataset for 22 AI Models How We Measure Content ROI on a Comparison Site: Revenue Attribution Without Perfect Data Introducing Nova AI Ops: The AI-Native Operating System for SRE Teams I built a free desktop video downloader for Windows — Grabbit How Talkie OCR Helps Vision-Impaired & Dyslexic Users Read the World Around Them VRCFaceTracking安装和iPhone面捕配置教程,有bug Even CrowdStrike Can't See Your Agents The Automation Gold Rush: What n8n Workflows and Claude Are Opening Up for Developers Right Now
DiffusionGemma 26B 挑戰 GH200 效能極限
JH5 · 2026-06-19 · via DEV Community

1180 tok/s 的地表極速是什麼概念?在 256 tokens 的輸出下,運算只要 0.22 秒就瞬間結束,這表示 DiffusionGemma 26B 在 NVIDIA GH200 上跑 vLLM 的速度,整整比 M2 Max 快了 80 倍!

延續系列第一篇在 M2 Max 96GB (MLX) 篇 中探討地端 Agent「無限 Token 自由」的實驗,當時 Standard 4-bit 雖然擠出了 31.6 tok/s 的不錯峰值,但面對長 Context(上下文)與多用戶併發請求時,Mac 的排隊機制與記憶體頻寬依然顯得力不從心。

為了追求 Production等級部署,我們將戰場移到魔王級的硬體—— NVIDIA GH200 (Grace Hopper),當強大的 Diffusion 架構遇上 vLLM 優化,不僅 32,653 tokens 的 Context 直接逼近極限打滿,併發吞吐量也是狂飆猛飆,雖然上面還是舊的HBM3,但是效果體感上還是滿讓人滿意的。

測試環境:vLLM + GH200 480GB

項目 內容
GPU NVIDIA GH200 480GB(單顆 Hopper + Grace CPU NVLink-C2C)
HBM3 95.6 GB 可用
CPU 72 核 ARM Neoverse(Grace)
系統 Rocky Linux 9.7 (aarch64)
框架 vLLM 0.22.1rc1(容器:vllm/vllm-openai:gemma)
模型 nvidia/diffusiongemma-26B-A4B-it-NVFP4
部署參數 --dtype auto --max-model-len 32768 --gpu-memory-utilization 0.60 --max-num-seqs 4 --attention-backend TRITON_ATTN

為了簡化部署的架構,我是讓 vLLM 跑在 Podman 容器裡,GPU memory utilization 嘗試幾次後,後來設 0.60 不是為了省記憶體,是因為更高會在 warmup 階段因為 CUDA graphs 配置沒留空間而 OOM,KV cache 能用的都是剩下的。

Generation Throughput:1180 tok/s 是什麼概念

生成速度的測試我先維持了跟 Mac 一樣的 prompt 和參數:

Output 長度 速度 延遲
64 tokens 260 tok/s 0.25s
128 tokens 519 tok/s 0.25s
256 tokens 887~1180 tok/s 0.22~0.29s
512 tokens 936~1053 tok/s 0.49~0.55s
1024 tokens 1011 tok/s 1.01s

256 tokens 最快(1180 tok/s),因為剛好 fit 一個 canvas,不過更有趣的是 1024 tokens 只花了 1.01 秒,代表多個 canvas 的 parallel processing 在 Hopper 的架構上卻 scale 的更好。

跟 M2 Max 對比(同為 256 tokens):

平台 速度 差距
M2 Max (MLX) 14.7 tok/s 1x
GH200 (vLLM) 1180 tok/s 80x

這部分應該是 vLLM 的 Hopper GPU 針對 diffusion 架構做了 TRITON_ATTN backend 和專屬 denoising kernel 優化。

Context Scaling:32K 全滿可用

GH200 真正有優勢的地方在這,M2 Max 到 16K 就喘呼呼了,而GH200 直接推到 32K 上限:

Context 長度 速度 延遲
5K tokens 104 tok/s 0.61s
10K tokens 66 tok/s 0.97s
20K tokens 39 tok/s 1.63s
32K tokens 4.0 tok/s 15.9s

前 20K 的延遲都在 2 秒以內,對實際開發應用來說完全可接受,但是到 32K 的時候掉到 4 tok/s,原因是 KV cache 接近用盡 + diffusion intermediate states 競爭記憶體頻寬,不過我實際上接到 opencode CLI 上使用,體感還是滿好的。

併發吞吐量:vLLM 的優勢

跟 M2 Max MLX server 不同,vLLM 有真正的 batching 和 concurrent kernel execution:

情境 GH200 M2 Max
Sequential 平均延遲 0.06~0.12s 1.5s
Concurrent 4 總吞吐量 256 tok/s 1.4 tok/s

在併發測試時 GH200 跑出 256 tok/s(wall 0.39s),而M2 Max 只有 1.4 tok/s,不止數據上差距 180 倍,實際上接到 opencode使用時心情也是差了好幾十倍。

GH200 的定位

DiffusionGemma 26B 在 GH200 上運作好棒棒,但還是有一件事要留意,目前 91 GB 模型佔用對 GH200 的 96 GB HBM3 來說太滿了,剩下 5 GB 的 headroom 在短 context 時沒問題,但如果需要同時處理大量長 context 請求,記憶體會是瓶頸或是造成其他的問題。

好在GH200 還有 480 GB 的 coherent memory 可以透過 NVLink-C2C 存取,但 vLLM 預設不會拿來放 model weights,這邊的 bandwidth 跟 HBM3 比起來還是差了不少,但是還是有一些社群上大神提供的參數還可以再測試讓記憶體再有餘裕一點。

接下來

DiffusionGemma 26B 在 GH200 搭配 vLLM 下的表現堪稱恐怖,在短 Context 靠著極致頻寬與 TRITON_ATTN 後端優化可以無壓力秒殺,但是遇到多用戶、長 Context 的極端高併發場景,剩餘 5 GB 的 KV Cache 空間就會迅速面臨撞牆瓶頸。

目前是先架在公司內多人連線測試中,也陸續還在調整一些參數來讓大家的 Agent token更有餘裕,後續如果還有更優化版本,再分享上來。

下一篇我們會拿 NVIDIA Blackwell GB10 的 128GB 統一記憶體來看看DiffusionGemma 26B 在長序列滿載時是否能展現更完美的完全體型態。


  • 模型:nvidia/diffusiongemma-26B-A4B-it-NVFP4
  • 框架:vLLM 0.22.1rc1(容器:vllm/vllm-openai:gemma
  • GPU:GH200 480GB(Grace Hopper)