flozi00TechHub
Technical articles and worked examples on servers, GPUs, AI infrastructure and modern IT systems.
A starting point for curious minds
Understand technology. One step at a time.
From the first token to a GPU cluster: every technical article explains its central idea with its own interactive diagram. Play, pause and explore the blocks. Start with attention, KV cache and quantization, then keep learning throughout the hub.
Start the visual learning path →AI basics · Model architectures · GPUs · Serving · Linux
LLM inference: Start here
LLM VRAM Calculator: GPU Memory & Inference Estimates
Estimate LLM VRAM, KV-cache memory and theoretical token throughput for Qwen, DeepSeek and GLM. Compare GPUs, precision and batch sizes.
LLM VRAM Requirements: Weights, Quantization & KV Cache
How much VRAM does an LLM need? Calculate weights, quantization and KV-cache memory step by step, with a Qwen3-VL-32B example and a VRAM calculator.
NVIDIA B200 vs GB200: Differences & MLPerf Benchmarks
B200 vs GB200 explained: compare system design, NVLink domains and published MLPerf training results, with the limits of performance comparisons.
KV Cache Explained: Memory Formula, GQA, MQA & MLA
Calculate KV-cache memory per token and request. Compare GQA, MQA and MLA, understand PagedAttention, and work through GPU capacity examples.
What question do you want to answer?
Reading paths for VRAM, GPU selection, serving, KV cache, architectures and Linux explain which guide helps with each task.
Explore topics and reading paths →Browse by topic
Find model families, AI fundamentals, inference, hardware and calculators in one menu.
Find Anything Fast
Search all articles, model analyses and guides — specs, configurations and benchmarks included.
in the header — just start typing
Explore by Topic
View the topic directory →Learning & fundamentals
9 articlesUnderstand AI step by step: visual explanations, attention, KV cache and quantization.
Show 6 more articles
- KV Cache Explained: Memory Formula, GQA, MQA & MLA
- KV-Cache Glossary: GQA, MLA, TTFT and Eviction
- LLM Quantization: FP8, INT4, GGUF & VRAM Requirements
- LLM VRAM Requirements: Weights, Quantization & KV Cache
- NVFP4 vs INT4: A Visual Guide to 4-Bit Quantization
- Visual Tech Learning: From AI Basics to Systems
AI models
33 articlesModel families, architectures and practical applications, from Nemotron and Qwen to speech, images and embeddings.
Show 30 more articles
- DeepSeek V4 Pro: Self-Hosting VRAM and Cost Math
- DeepSeek V4.1 Flash: KV-Cache Compression
- EmbeddingGemma 2 Architecture: Soft Tokens, PLE and MRL
- Fibo 2.0 Architecture: Gist Tokens and Editing Costs
- GLM-5.3 MTP on B300: Acceptance Math and NCCL NVLS
- GPU for Qwen3 Inference: VRAM, Bandwidth & Model Fit
- Granite Speech 5.0 TurboCTC Architecture
- Inside GLM-5.3-Flash: KDA, Sparse MLA, mHC, and 8-of-288 MoE
- Kimi K3 Architecture: KDA, MLA and LatentMoE
- KITE: two-tower MoE and KV cache (research)
- Kolibri 1 Architecture: Hybrid Attention, MoE and FP8
- LFM2.5-24B-A2B: Routing and Cache Architecture
- Ling-3.1-flash: KDA, MLA and Actual Cache Layouts
- Liquid d1-3B: Shared-State Branch Inference
- Mellum2.1: MoE, Sliding Attention and Selective YaRN
- MI350P vLLM Optimization: Kernels and Measurements
- Muse Spark Open: Residual Gates and Normalized Experts
- Nemotron 3.5 Bolt: LatentMoE, IQ2 and Memory Budgets
- Nemotron 3.5 Embed: compatible multimodal retrieval
- Nemotron 3.5 Super VL: Visual Tokens and Video Pruning
- Nemotron 3.5 Transcribe: Architecture and Dataset Auditing
- Nemotron E4B DFlash: Block Speculation and Serving Math
- OlmoEarth v1.3: pooler, Matryoshka student, Searchlight
- Parakeet TDT vs Encoder–Decoder ASR: Production Trade-offs
- pplx-embed-v2: 9B documents, 0.6B queries
- Qwen-Image 2.1: Attention, KV Cache and Turbo Math
- Qwen3-Next-80B-A3B: Architecture and Serving Limits
- Qwen3.8 Flash Next: GDN, Attention and N-gram Memory
- Step-5-Preview: Sparse GQA, Indexer and KV Memory
- TypeSafe Jev: What the Public Evidence Supports
Inference & optimization
57 articlesServing, memory, quantization, GPU kernels and costs for operating AI systems.
Show 54 more articles
- Beam 1.0: NoPE, Softmax Scaling and KV Memory
- BF16 vs FP16: Why Greedy LLM Outputs Can Change
- BOOST: Host Memory and HBM for LLM Decode
- CXL-SSD Prefix Caching: Why KV Chunk Awareness Matters
- DeepSeek V4 Pro: Self-Hosting VRAM and Cost Math
- DeepSeek V4.1 Flash: KV-Cache Compression
- Expert Parallelism: The Router's Dilemma
- FlashLoop: Looped Transformers and KV-Cache Savings
- GLM-5.3 MTP on B300: Acceptance Math and NCCL NVLS
- HBF for KV Cache: Capacity vs Latency
- HBM Capacity and KV-Cache Economics
- KernelOPT: Agentic Triton Search and Verification
- KITE: two-tower MoE and KV cache (research)
- Kolibri 1 Architecture: Hybrid Attention, MoE and FP8
- KREX: Reliable Kernel Timing on Shared GPUs
- KV Cache Explained Visually: Prefill, Decode and Memory
- KV Cache Explained: Memory Formula, GQA, MQA & MLA
- KV Cache Memory Growth Explorer
- KV-Cache Billing: Token Counts vs Serving Time
- KV-Cache Compression: Tensor Rank and Error
- KV-Cache Glossary: GQA, MLA, TTFT and Eviction
- KVSET: Size an LLM Prefix Cache with LRU Traces
- LFM2.5-24B-A2B: Routing and Cache Architecture
- Ling-3.1-flash: KDA, MLA and Actual Cache Layouts
- llama.cpp vs Ollama vs vLLM: Local LLM Serving Compared
- LLM Inference Economics: Costs & Cloud Break-Even
- LLM Inference Performance: Roofline Math & GPU Bandwidth
- LLM Prefix Caching: LRU Eviction and Its Limits
- LLM Quantization: FP8, INT4, GGUF & VRAM Requirements
- LLM Quantization: Separate Prefill and Decode Formats
- LLM Serving Engines 2026: Upgrades and Version Pinning
- LLM VRAM Calculator: GPU Memory & Inference Estimates
- LLM VRAM Requirements: Weights, Quantization & KV Cache
- LoRA and QLoRA Fine-Tuning: The Actual Memory Math
- Mellum2.1: MoE, Sliding Attention and Selective YaRN
- MI350P vLLM Optimization: Kernels and Measurements
- Nemotron 3.5 Bolt: LatentMoE, IQ2 and Memory Budgets
- Nemotron 3.5 Embed: compatible multimodal retrieval
- Nemotron 3.5 Super VL: Visual Tokens and Video Pruning
- Nemotron E4B DFlash: Block Speculation and Serving Math
- NVFP4 vs INT4: A Visual Guide to 4-Bit Quantization
- NVIDIA and Hugging Face: Serving Dependencies
- OlmoEarth v1.3: pooler, Matryoshka student, Searchlight
- Pipeline Parallelism: Pumping Data Upstream
- Prompt-Cache Prices: September 2026 Comparison
- PTQ: Estimate Quantization Error Before Deployment
- Qwen-Image 2.1: Attention, KV Cache and Turbo Math
- Random Attention vs Score-Based KV Eviction
- Risk-Controlled KV Eviction and Full-Cache Fallback
- Step-5-Preview: Sparse GQA, Indexer and KV Memory
- Tensor Parallelism: Slicing the Silicon
- UNISON: Scheduling KV Cache for Paused Agents
- Visual Tech Learning: From AI Basics to Systems
- vLLM vs SGLang: Serving, Caching & When to Use Each
Embeddings & RAG
6 articlesEmbedding models, vector search, retrieval quality and memory budgets.
Show 3 more articles
Hardware & systems
17 articlesCompare GPUs, memory and networking, and set up Linux, drivers and containers.
Show 14 more articles
- CXL-SSD Prefix Caching: Why KV Chunk Awareness Matters
- CXMT at 10% of DRAM Revenue: What the Figures Show
- GPU Buying Guide for LLM Inference (2026)
- GPU for Qwen3 Inference: VRAM, Bandwidth & Model Fit
- HBF for KV Cache: Capacity vs Latency
- HBM Capacity and KV-Cache Economics
- MLPerf Inference: Hardware vs Software Gains
- NVIDIA B200 vs GB200: Differences & MLPerf Benchmarks
- NVIDIA Error 802 on Ubuntu: Diagnose Before Trying nokaslr
- NVIDIA GPU Containers on Ubuntu 24.04
- NVIDIA Groq 3 LPX: SRAM, Bandwidth and LLM Decode
- NVIDIA NVLink: Bandwidth, NVSwitch & GPU Topology
- Scale-Up vs Scale-Out: NVLink and All-Reduce
- Snapdragon X2 on Linux: Local LLM Memory Limits
Benchmarks & evaluation
7 articlesAssess measurements, model evaluations and datasets, and interpret results.
Show 4 more articles
AI regulation & transparency
5 articlesEU AI Act, data protection, disclosure and the technical limits of watermarking.
Show 2 more articles
Calculators & tools
3 articlesCalculate VRAM, KV-cache growth and inference costs interactively.
Latest Articles
Nemotron 3.5 Transcribe: From Recorded Speech to Dataset Audits
How Nemotron 3.5 Transcribe combines speech and speaker perception with a hybrid MoE language model, and how to use it to audit existing audio–transcript datasets without mistaking model agreement for ground truth.
Attention Explained Visually
Follow queries, keys, values and softmax through a small attention example, with step-by-step diagrams and plain-language explanations.
How LLMs Actually Work — An Animated Walkthrough
From token to probability distribution: embeddings, attention, training and sampling — explained without the equation jump, with animation.
The KV Cache: Bit-Exact Memory Math, GQA vs MQA vs MLA, and PagedAttention
Why the key-value cache eats your GPU memory: the exact per-token formula, MHA vs MQA vs GQA vs MLA with computed examples, PagedAttention block math, RadixAttention prefix reuse, FP8 KV cache, and a capacity worked example — grounded in the primary papers.
The KV Cache Explained Visually
See what a language model saves in its KV cache, how prompt processing differs from generation, and why longer context uses more memory.