@meta
  v: 1
  route: /
  generated: 2026-06-10T05:09:59.291Z

@intent
  purpose:    Calculate VRAM budget for local LLM inference. Find a GPU that fits a model, or a model that fits a GPU.
  audience:   ai-engineer, ml-researcher, self-hoster, homelab-buyer
  capability: compute_budget, find_compatible_gpus, find_compatible_models, view_math, search_huggingface

@state
  formula: modelSizeGB = params * (bits / 8)
  voice: frontier-operations
  site_version: 0.1.0
  gpu_catalog_size: 42
  model_catalog_size: 30
  gpu_index: /gpu/rtx-4090
  model_index: /model/llama-3-1-70b

@actions
  - id: calculate
    method: GET
    href: /#calculator
  - id: view_math
    method: GET
    href: /the-math
  - id: browse_gpus
    method: GET
    href: /gpu/rtx-4090
  - id: browse_models
    method: GET
    href: /model/llama-3-1-70b

@context
  > vrambudget.com shows the math behind local LLM memory budgets. Every other "can I run this LLM" tool lists models without showing the formula. This one shows the formula, then lists models. The interactive calculator takes GPU + context + concurrency + safety headroom and returns how much VRAM is left for weights. The /the-math page explains every term. Per-GPU and per-model pages show the precomputed fit matrix.

@nav
  self:      /
  peers:     [/the-math]
  drilldown: /the-math