# VRAMLab

> Architecture-aware open-weight LLM GPU & Apple Silicon memory sizing workbench across 28 verified Transformer architectures (Dense & Mixture-of-Experts weights, Multi-Head / Grouped-Query / DeepSeek Multi-head Latent / Hybrid Sliding-Window KV cache math, zero-upload local .gguf & config.json binary header inspection, vLLM / SGLang / Ollama / Kubernetes GPU serving configs, and 30-day self-hosted GPU vs Cloud API FinOps break-even modeling).

## Core Interactive GPU & LLM Memory Workbenches

- [LLM VRAM, KV Cache & Throughput Calculator](https://vram.codeap.com/): 0ms client-side memory breakdown across Dense & MoE weights, MLA/GQA/SWA KV cache, Apple Silicon sysctl iogpu.wired_limit_mb ceilings, and bandwidth-bound tok/s speed.
- [Open-Weight Model VRAM Directory (28 Models)](https://vram.codeap.com/models): Multi-precision (FP16, FP8, Q8_0, Q6_K, Q5_K_M, Q4_K_M, AWQ INT4, IQ4_XS) and multi-context (8K, 32K, 128K) sizing guides.
- [GPU & Apple Silicon Capacity Guides](https://vram.codeap.com/hardware): Runnable LLM matrices and bandwidth ceilings across RTX 3090, RTX 4090, RTX 5090, Apple Silicon 64GB/128GB/192GB, A100 80GB, and H100 80GB.
- [Local .gguf & config.json Binary Header Inspector](https://vram.codeap.com/tools/gguf-inspector): Zero-upload browser File.slice() + DataView GGUF v2/v3 metadata extractor with instant VRAM projection.
- [vLLM, SGLang, Ollama & Kubernetes Config Compiler](https://vram.codeap.com/tools/serving-config-generator): OOM-safe CLI flags, Modelfiles, Docker Compose GPU stacks, and Kubernetes Deployment manifests.
- [Self-Hosted GPU vs Cloud API Break-Even Calculator](https://vram.codeap.com/tools/gpu-vs-api-breakeven): 30-day FinOps break-even daily request solver with prompt-cache discount and continuous-batching saturation checks.

## Methodology, Governance & Sibling Studios

- [Architecture & Bandwidth Methodology](https://vram.codeap.com/about): First-principles bpw tables, MHA/GQA/MLA/SWA KV equations, and unified memory ceiling rules.
- [Privacy Policy & Zero-Upload Guarantee](https://vram.codeap.com/privacy): 100% local tab execution and browser storage controls.
- [XML Sitemap](https://vram.codeap.com/sitemap.xml): Complete crawlable URL index across 9 locales.
- [CodeAP Studio](https://codeap.com/llms.txt): Unified engineering workbench portal.
- [DevToken](https://devtoken.codeap.com/llms.txt): Live LLM API token pricing ledger, 429 rate-limit sizer, and MCP/Zod generators.
- [LLMEOL](https://eol.codeap.com/llms.txt): Multi-vendor LLM deprecation tracker and CI config auditor.
- [ToMarkdown](https://markdown.codeap.com/llms.txt): Zero-upload PDF/DOCX/XLSX/HTML-to-Markdown & RAG chunker.

## Localized Editions
- [VRAMLab — Español](https://vram.codeap.com/es): Español (es) edition.
- [VRAMLab — Français](https://vram.codeap.com/fr): Français (fr) edition.
- [VRAMLab — Deutsch](https://vram.codeap.com/de): Deutsch (de) edition.
- [VRAMLab — Italiano](https://vram.codeap.com/it): Italiano (it) edition.
- [VRAMLab — Português (Brasil)](https://vram.codeap.com/pt-br): Português (Brasil) (pt-BR) edition.
- [VRAMLab — 日本語](https://vram.codeap.com/ja): 日本語 (ja) edition.
- [VRAMLab — 한국어](https://vram.codeap.com/ko): 한국어 (ko) edition.
- [VRAMLab — 简体中文](https://vram.codeap.com/zh-cn): 简体中文 (zh-CN) edition.