Avatar 🍥

KnightLi Blog

记录并分享日常

  1. Home
  2. About
  3. Archives
  4. Search
  5. Links
    1. Dark Mode

Archives

2026 383
2025 23
2024 5
2023 9
2022 33
2021 5
2020 8

Categories

AI Tools Technical Docs Hardware AI Industry Development Tools Operations Security Updates Business Analysis

Tags

Ai-Tools AI Agent AI Coding Developer Tools Codex Claude Code Local LLM Linux Openai MCP Ubuntu ChatGPT Claude Ollama Gemini Anthropic Llama.cpp NAS Open Source GPU AI Art Gemma 4 Google Prompts Windows DeepSeek GGUF LLM Python AI Models
Hardware

How to Choose a Diode: General, Fast Recovery, Schottky, Zener, LED, and TVS Explained

A quick diode selection guide covering general-purpose diodes, fast recovery diodes, Schottky diodes, Zener diodes, LEDs, and TVS diodes, with typical use cases for each.

2026-04-30
5 minute read
中文简体 中文繁體 日本語 Español
Development Tools

Getting Started with Compiling UEFI Programs: From uefi-simple to Your First .EFI

A beginner-friendly guide to compiling your first UEFI .EFI program: what a UEFI program is, why uefi-simple is a good starting point, which tools to prepare, and where beginners usually get stuck.

2026-04-30
6 minute read
中文简体 中文繁體 日本語 Español
Hardware

LGA1851 Z990/W980/Q970/Z970/B960/Z890/W880/Q870/B860/H810 Motherboard Lane Reference

A text-based summary of motherboard chipset lane configurations, covering CPU direct lanes, chipset expansion lanes, and common I/O resources across Intel and AMD consumer platforms, HEDT, Threadripper, and EPYC.

2026-04-30
23 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Claude.md Is Not Better When It Is Longer: How to Write Global Memory Files for AI Coding

A practical look at what global memory files such as Claude.md and AGENTS.md are for, where they often go wrong, and how to write them: fewer introductions, more durable constraints, and reusable workflows moved into skills or commands.

2026-04-29
8 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Codex Is Starting to Control the Computer. What Does That Mean for the Future?

An introduction to Codex's computer use capability, and an analysis of how this kind of Agent ability may affect workflows, software interaction, and the way ordinary users operate computers.

2026-04-29
5 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Why Does a Codex Skill Exist in the Directory but Still Not Show Up?

A troubleshooting note about a Codex skill that existed under ~/.codex/skills but failed to load because SKILL.md started with a UTF-8 BOM, preventing YAML front matter detection.

2026-04-29
3 minute read
中文简体 中文繁體 日本語 Español
AI Tools

What Is the Difference Between ~/.codex/skills and Project .codex/skills in Codex

A clear explanation of the difference between global `~/.codex/skills` and project-level `.codex/skills` in Codex, and why a skill can exist on disk but still not appear in the current session.

2026-04-29
3 minute read
中文简体 中文繁體 日本語 Español
Hardware

Why Are 16-Core Board+CPU Combos So Cheap? Is a Xeon D-1581 Integrated Board Actually Worth Buying?

A direct look at Xeon D-1581 integrated board+CPU combos: why they look so cheap, what they are actually good for, and the pros and cons people most often overlook.

2026-04-29
4 minute read
中文简体 中文繁體 日本語 Español
Development Tools

GoAccess Latest Build-from-Source Notes: From Source Install to Real-Time HTML Reports

A command-focused GoAccess setup note based on the latest official repository, covering source installation of the newest release on Ubuntu or Debian, version checks, HTML report generation, and real-time viewing.

2026-04-29
3 minute read
中文简体 中文繁體 日本語 Español
AI Tools

How to Choose Between GPT 5.5, Claude Opus 4.7, DeepSeek V4, and Qwen 3.6 Max

A direct guide to what GPT 5.5, Claude Opus 4.7, DeepSeek V4, and Qwen 3.6 Max each do well, where they still fall short, and how to choose between them.

2026-04-28
7 minute read
中文简体 中文繁體 日本語 Español
Hardware

Is the Core Ultra 5 230F Worth Buying? How It Compares with the 12400F, 13490F, and 7500F

A practical look at whether the Core Ultra 5 230F is worth buying right now by comparing it with common alternatives like the 12400F, 13490F, and 7500F, including its strengths, weaknesses, and the kinds of builds it suits best.

2026-04-28
7 minute read
中文简体 中文繁體 日本語 Español
AI Industry

Why Elon Musk and SpaceX Want the $60 Billion Option to Acquire Cursor

A breakdown of why Elon Musk and SpaceX are not buying Cursor outright today but instead taking a $60 billion acquisition option, with motives tied to compute, user distribution, valuation flexibility, Musk's AI strategy, and pre-IPO positioning.

2026-04-28
7 minute read
中文简体 中文繁體 日本語 Español
Hardware

How to Pick a GPU in April 2026: Which Models to Avoid and Which Ones Are More Worth Considering

A model-focused GPU buying guide for April 2026, covering which cards are less worth buying and which ones are more sensible picks, with an emphasis on the 5060 Ti, 5070, 5070 Ti, and a few older cards.

2026-04-27
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Ralph and Multi-Agent Collaboration: How to Keep AI Working Reliably Over Long Tasks

A practical look at the difference between the Ralph loop approach and multi-agent collaboration, and at the key design choices behind long-running AI workflows that stay stable.

2026-04-27
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

What Ralph Is: Turning Claude Code and Amp into a Repeatable Autonomous Development Loop

Based on the snarktank/ralph README, this article explains Ralph's core idea: letting Claude Code or Amp run one PRD story at a time in fresh context, while git, progress.txt, and prd.json preserve continuity across iterations.

2026-04-27
6 minute read
中文简体 中文繁體 日本語 Español
Hardware

How to Choose an Intel 800 Series Chipset: Feature Differences Between Z890, W880, Q870, B860, and H810

A breakdown of Intel 800 Series chipset segmentation, focusing on the differences between Z890, W880, Q870, B860, and H810 in expansion resources, overclocking permissions, ECC, vPro, USB4, and PCIe 5.0 support.

2026-04-27
7 minute read
中文简体 中文繁體 日本語 Español
Operations

Ubuntu 26.04 LTS GPU and Hardware Updates: CUDA, ROCm, DPC++, and More Platform Changes

A summary of the Ubuntu 26.04 LTS release notes related to GPU computing, AI software stacks, hardware support, and platform requirements, including DPC++, CUDA, ROCm, Intel GPUs, Raspberry Pi, RISC-V, and IBM Z.

2026-04-26
8 minute read
中文简体 中文繁體 日本語 Español
Operations

Ubuntu 26.04 LTS Released: Major Desktop Updates with GNOME 50 and Linux 7.0

A quick summary of the key updates in the official Ubuntu 26.04 LTS release notes, including GNOME 50, Linux kernel 7.0, Wayland, desktop app updates, hardware requirements, and upgrade paths.

2026-04-26
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

DeepSeek V4 Pro vs GPT-5.5: After Testing Frontend, Writing, and Coding, the Gap Feels Bigger Than Expected

After putting DeepSeek V4 Pro and GPT-5.5 into three high-frequency tasks—frontend development, writing, and coding—you quickly find that the real gap is not the first output, but stability, rework rate, and the experience of sustained collaboration.

2026-04-25
7 minute read
中文简体 中文繁體 日本語 Español
AI Tools

How to Split Tasks Between ChatGPT, Claude, and Gemini: Choosing for Daily Use, Coding, and Special Capabilities

This article breaks down how to divide work between ChatGPT, Claude, and Gemini, covering daily conversations, command-line programming, and special capability scenarios, along with the common mistakes people make with each.

2026-04-25
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Why LLM APIs Charge by Tokens: A Clear Guide to Input, Output, and Context Costs

This article explains why LLM APIs are billed by token, why input and output are priced separately, how long context and tool calls amplify cost, and how developers can estimate usage more accurately.

2026-04-25
8 minute read
中文简体 中文繁體 日本語 Español
AI Tools

DeepSeek-V4 Preview Released: 1M Context, Two Models, and API Migration Notes

Based on DeepSeek's official news page published on April 24, 2026, this article summarizes the key points of DeepSeek-V4 Preview, including V4-Pro, V4-Flash, 1M context, agent-focused optimizations, API model changes, and the retirement notice for older models.

2026-04-24
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

How to Fix Ollama Using CPU Instead of GPU

A practical troubleshooting guide for Ollama running on CPU instead of GPU, covering GPU detection, ROCm or CUDA setup, service restarts, VRAM limits, and common AMD compatibility issues.

2026-04-24
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

What Is NVIDIA nvbandwidth: How to Use This GPU Bandwidth Testing Tool

Based on the official NVIDIA/nvbandwidth repository and Releases page, this article explains what the GPU bandwidth testing tool does, what it depends on, how to use it, how multinode testing works, and what changed in v0.9.

2026-04-24
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

K-Nearest Neighbors for Beginners: Understanding Machine Learning Classification Through Neighbor Voting

A beginner-friendly explanation of the basic idea behind K-nearest neighbors: what K means, why nearby samples matter, how voting works, and where KNN is useful or limited.

2026-04-24
8 minute read
中文简体 中文繁體 日本語 Español
AI Tools

OpenAI Releases GPT-5.5: Stronger Agentic Coding, Knowledge Work, and Research

Based on OpenAI's GPT-5.5 announcement on April 23, 2026, this article summarizes the key updates around agentic coding, knowledge work, research, safety, API availability, and pricing.

2026-04-24
6 minute read
中文简体 中文繁體 日本語 Español
Hardware

How Intel's ATX 3.0 Design Guide Classifies PCIe Auxiliary Power Connectors for GPUs

Based on Intel's ATX 3.0 Multi Rail Desktop Platform Power Supply Design Guide, this article sorts out the roles, power ranges, and sideband signals of the common PCIe GPU auxiliary power connectors: 2x3, 2x4, and 12V-2x6.

2026-04-23
7 minute read
中文简体 中文繁體 日本語 Español
AI Tools

How to Choose Common Embedding Models: OpenAI vs BGE vs E5 vs GTE vs Jina

A practical comparison of common embedding models such as OpenAI, BGE, E5, GTE, and Jina, with a focus on how to choose for Chinese-language use cases.

2026-04-23
7 minute read
中文简体 中文繁體 日本語 Español
AI Tools

What image vectorization is: from pixel images to searchable, analyzable vector representations

A practical explanation of image vectorization: why images need to move from pixel representations to vector representations, how that process usually works, and what problems it actually solves in search, recommendation, recognition, and enterprise digital workflows.

2026-04-23
8 minute read
中文简体 中文繁體 日本語 Español
Technical Docs

What auto-editor does: cut silence automatically and export to Premiere or Resolve

A practical overview of what auto-editor is good at: making a first-pass rough cut by removing silence or low-motion sections automatically, then exporting to editors like Premiere, DaVinci Resolve, or Final Cut Pro, or rendering directly.

2026-04-23
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

AI Terms Explained: Agent, MCP, RAG, and Token in Plain Language

A plain-language guide to 10 common AI terms, including Agent, Skills, MCP, API, RAG, AIGC, and Token, to help beginners build a basic framework for understanding everyday AI discussions.

2026-04-23
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

How to Tune llama.cpp on 8GB VRAM: Why 32K Is Safer and 64K Needs KV Cache Quantization

A practical guide to tuning llama.cpp on 8GB VRAM: what 32K, 64K, and KV Cache mean, why 32K is often the safer balance point, why 64K depends more on cache quantization, and why blindly increasing CPU threads can make performance worse.

2026-04-23
7 minute read
中文简体 中文繁體 日本語 Español
Hardware

How to Check Whether a Tesla V100 Has ECC Errors

Use nvidia-smi to quickly inspect the ECC status of a Tesla V100 and determine which error counters should be 0 or N/A.

2026-04-23
2 minute read
中文简体 中文繁體 日本語 Español
Hardware

Is Tesla V100 Still Worth Buying: ECC Checks, Cooling Mods, and DIY Pitfalls

A practical guide to buying a Tesla V100: how to read production dates and visual clues, how to interpret ECC values, what signs suggest the card has been tampered with, and why DIY cooling and power setups fail so easily.

2026-04-23
9 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Claude Code's Four-Part Environment Setup: CLAUDE.md, Rules, Memory, and Hooks Explained

Why does environment setup matter more than prompts once you use Claude Code seriously? This article explains CLAUDE.md, Rules, Memory, and Hooks in one pass, and gives a practical order for getting started.

2026-04-23
11 minute read
中文简体 中文繁體 日本語 Español
AI Tools

llama.cpp GPU Performance Ranking: Full CUDA, ROCm, and Vulkan Scoreboards Explained with pp512 / tg128 / FA

Based on the visible scoreboard data in GitHub Discussions as of 2026-04-23, this article compiles the full llama.cpp GPU benchmark tables for CUDA, ROCm, and Vulkan, and explains what pp512, tg128, Q4_0, and FA actually mean.

2026-04-23
28 minute read
中文简体 中文繁體 日本語 Español
AI Tools

What the Common GPU Inference Benchmark Metrics Actually Mean: FA, pp512, tg128, and Q4_0

When reading GPU inference benchmarks, you often run into metrics like FA, pp512, tg128, Q4_0, and t/s. They all relate to performance, but they do not measure the same thing. This article breaks down what each of them actually means.

2026-04-23
8 minute read
中文简体 中文繁體 日本語 Español
Development Tools

How to Choose an Embedded Development Environment in 2026: Keil, STM32CubeIDE, VS Code, and AI Collaboration

In 2026, when AI-assisted coding has become common, how should embedded developers choose their environment? Instead of betting on a single IDE, a more practical answer is often to let Keil handle build and debugging while VS Code handles editing and AI collaboration.

2026-04-22
7 minute read
中文简体 中文繁體 日本語 Español
AI Tools

A Practical Guide to Common Tensor Formats in LLMs: FP32, FP16, BF16, TF32, and FP8

A practical introduction to the most common tensor formats used in large models: FP32, FP16, BF16, TF32, and FP8, including their bit layouts, trade-offs, and why they shape training and deployment behavior.

2026-04-22
7 minute read
中文简体 中文繁體 日本語 Español
Development Tools

How to Choose Among 8 Common Config File Formats: From INI, XML, JSON, YAML, TOML to Markdown

A practical comparison of 8 common config file formats, including INI, XML, JSON, YAML, TOML, Apache-style config, Protocol Buffers, and Markdown as it becomes newly relevant in the AI Agent era.

2026-04-22
10 minute read
中文简体 中文繁體 日本語 Español
AI Tools

A 16GB GPU Can Still Run 35B Models: VRAM Compression Strategies for MoE Models in LM Studio

A practical look at how a 16GB GPU can still run 35B-class MoE models: with the right architecture choice and LM Studio settings, 16GB VRAM does not necessarily cap you at 12B to 14B models.

2026-04-22
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Claude Code Multi-Agent Collaboration: How to Choose Between Subagents and Agent Teams

A practical breakdown of the differences between Subagents and Agent Teams in Claude Code, including when each one fits best in multi-agent workflows.

2026-04-22
8 minute read
中文简体 中文繁體 日本語 Español
AI Tools

GPT Image 2 Officially Launches: From Generating Images to Commercial Use

A summary of the most important capability shifts after the launch of GPT Image 2, including text rendering, photorealism, world knowledge, UI screenshot generation, and localized editing, along with how to use it in ChatGPT and the API plus prompt references for several common image-generation scenarios.

2026-04-22
9 minute read
中文简体 中文繁體 日本語 Español
AI Tools

nuwa-skill: Turning "distilling a person" from an idea into an executable workflow

alchaincyf/nuwa-skill is not just about imitating a famous person's tone. It turns the process of researching, extracting, and validating how someone thinks into a reusable Claude Code Skill.

2026-04-22
6 minute read
中文简体 中文繁體 日本語 Español
AI Tools

OpenAI Introduces ChatGPT Images 2.0: Image Generation Starts Moving Toward Deliverable Output

Based on OpenAI's ChatGPT Images 2.0 announcement on April 21, 2026, this article highlights the most important signals from the release: stronger text and layout control, better multilingual performance, broader stylistic coverage, and a shift from image generation as output to image generation as deliverable work.

2026-04-22
5 minute read
中文简体 中文繁體 日本語 Español
Hardware

12V-2x6 vs. 12VHPWR: Notes on GPU 16-Pin Power Connector Differences

A concise note on the main differences between the 12V-2x6 and 12VHPWR GPU 16-pin power connectors: cable compatibility, pin length, SENSE logic, H++ marking, and 600W output capability.

2026-04-19
4 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Karpathy's 65-Line CLAUDE.md: Helping AI Coding Avoid Three Common Mistakes

A summary of Karpathy's observations on AI coding, and how Forrest Cheung turned those problems into CLAUDE.md rules: think first, keep it simple, make precise changes, and work toward verifiable goals.

2026-04-19
6 minute read
中文简体 中文繁體 日本語 Español
Hardware

Core Ultra 9 285T ES Notes: Q4A7, a B860 Engineering Board, and the 35W Power Wall

Notes on the Core Ultra 9 285T ES sample Q4A7: platform, motherboard, power delivery, memory, performance, gaming results, and buying advice. The specs look tempting, but the 35W power wall, high DDR5 latency, scarce ES boards, and limited BIOS make it better suited to low-power tinkering than a gaming PC.

2026-04-19
9 minute read
中文简体 中文繁體 日本語 Español
AI Tools

Using Claude Code Quota More Efficiently: Models, Context, Caching, and /compact

A practical note on why Claude Code and Claude Pro/Max usage can run out quickly: model choice, 5-hour usage windows, long conversations, files and images, cache misses, CLAUDE.md, MCP, and skills, with habits around /compact, /clear, /context, and /status.

2026-04-19
8 minute read
中文简体 中文繁體 日本語 Español
AI Tools

rembg Project Notes: A Local Background Removal Tool

A practical look at danielgatis/rembg: what it is, how to install it, CLI usage, Python integration, HTTP server mode, Docker usage, model choices, and where it fits in local background removal workflows.

2026-04-19
7 minute read
中文简体 中文繁體 日本語 Español
1 2 3 4 5 6 7 8 9 10
© 2022 - 2026 KnightLi Blog
记录并分享
Built with Hugo
Theme Stack designed by Jimmy