⚡Data Center Power Delivery Model

Interactive model of AI data center power — from grid to chip. Designed for AI-safety legislators and policy makers.

Simulation Backend

DETECTING...

v1.0.0

Last simulation runtime: —

🧩Preset Configuration

Load a demonstrator setup to see key features of the power signature.

Preset Configurations

📖Glossary

GPU / Accelerator — A specialised chip for AI calculations. Modern training uses thousands working together.

VRM — Voltage Regulator Module — converts 48V to <1V for the GPU. Like a step-down adapter.

ORv3 Power Shelf — Standardised unit (Open Compute Project) that converts building AC power to 48V DC.

Busbar — A thick copper bar distributing DC power inside the rack — like the main pipe in a building.

Row PDU — Row Power Distribution Unit — distributes AC power from the busway to individual racks in a row.

PDB — Power Distribution Board — a PCB inside the tray that splits 48V into separate GPU and CPU power domains.

TDP — Thermal Design Power — the maximum sustained power a chip is designed to use.

SNR — Signal-to-Noise Ratio — how clearly a signal stands out. Higher = easier to detect.

Context: AI training consumes large amounts of power with distinctive signatures due to the clustered nature of the workloads. This model helps users to understand the power signature at different stages of the data center power delivery system, and how measureable different signals are at each stage. Power monitoring is a potential confidence building avenue for AI safety and verification. This model is designed to help policymakers understand the opportunities and limitations of power monitoring for AI systems.

Select a preset in the left panel, then click "Run Simulation" to generate the power flow diagram.