---
name: undominated-gpu-footprint-calculator
description: Calculates required GPU VRAM, tensor parallelism sharding (TP=2, 4, 8), and minimum cluster configurations for hosting open-weight models.
license: MIT
metadata:
  author: Undominated.ai
  version: "1.0.0"
  tags: ["gpu", "vram", "sharding", "tensor-parallel", "architecture"]
---

# GPU Footprint & Sharding Architect

## Mission & Boundary
Calculates required GPU VRAM, tensor parallelism sharding (TP=2, 4, 8), and minimum cluster configurations for hosting open-weight models.

## When to Deploy
Deploy before provisioning GPU infrastructure or evaluating self-hosting vs serverless token endpoints.

## Operational Boundaries & Limitations
Estimates exclude framework runtime overheads (CUDA context, PyTorch activation memory). Always add 15-20% headroom.

## Execution Protocol
Analyze model parameters, attention layers, head dimensions, and quantization formats. Calculate exact weight memory and KV cache memory. Output verified node configuration.
