AI Ecosystem CourseBy Queensly Austin
Module 4

Data-Center Physical Infrastructure

Introduction Available · Lessons In Development

Follow two paths: power in and heat out

A room of accelerators requires several supporting systems. What keeps them working reliably? The building is one part of the answer. Electrical equipment, cooling, controls, networks, maintenance, and operating procedures all have a role. A dependable service also needs tested software, networking, monitoring and operating procedures.

We will follow two paths: electricity from the campus connection to a rack, and heat from a processor to the surrounding environment. Then we will examine redundancy, network routes, operating readiness, and capacity claims. Revisit Module 3 for power and energy, and Module 2 for workload requirements.

Your project is a 100 MW AI Data Center Architecture. Define what the capacity measure includes, allocate IT capacity, trace cooling, identify where failures can spread, and specify how a customer would accept the service. This is a learning exercise. Construction requires professional engineering, equipment specifications, codes, and site data.

Planned lessons

  1. Facility Load Is Larger Than Accelerator Load
  2. The Substation Is Part Of The Delivery Path
  3. Switchgear Controls And Protects Circuits
  4. UPS And Backup Generation Serve Different Time Horizons
  5. Distribution Reaches The Rack
  6. Computers Produce Heat That Must Move Somewhere
  7. Air Cooling Depends On Controlled Airflow
  8. Direct-To-Chip Liquid Cooling And Cold Plates
  9. CDUs And Heat Exchangers Connect Loops
  10. Chillers, Cooling Towers, And Dry Coolers Reject Heat
  11. Rack Density Changes The Building Problem
  12. N, N+1, And 2N Need A Stated Boundary
  13. Failure Domains Connect Physical And Software Design
  14. Fiber Routes, Geography, And Latency
  15. Construction, Energization, And Commissioning
  16. Customer-Ready Capacity And Phase Acceptance
  17. Operations And Retirement Are Part Of The Lifecycle