Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ

Bài kiểm tra về cấp nguồn và làm mát trong trung tâm dữ liệu AI

Giải thích việc cung cấp năng lượng cho trung tâm dữ liệu, loại bỏ nhiệt, làm mát bằng chất lỏng trực tiếp, cân bằng PUE và nước cho các hệ thống tăng tốc AI dày đặc.

Đường dẫn hướng dẫn liên quanCấp nguồn và làm mát trong trung tâm dữ liệu AI
Câu hỏi 1 của 8

Why can AI accelerator racks require different cooling designs than lower-density server racks?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. Why can AI accelerator racks require different cooling designs than lower-density server racks?
    • Because accelerators produce less heat than most storage equipment
    • Because rack dimensions eliminate the facility heat load
    • They can concentrate more electrical power and heat in a smaller space
    • Because accelerator racks do not need electrical distribution
  2. What does direct-to-chip liquid cooling do?
    • Circulates coolant through the entire server room instead of using air
    • Circulates coolant near hot components to carry heat into a facility loop
    • Removes all heat without a heat-rejection system
    • Exhausts server air directly outside without heat exchange
  3. What does PUE compare?
    • IT equipment energy divided by total facility energy
    • Water consumed per IT energy unit
    • Compute output per rack area
    • Total facility energy divided by IT equipment energy
  4. What does PUE not measure by itself?
    • Water consumption or carbon intensity
    • Total facility energy
    • IT equipment energy
    • The energy ratio between facility and IT equipment
  5. Why can cooling designs trade electricity against water use?
    • Lower cooling energy always means lower water use in every climate
    • Liquid cooling eliminates all facility water requirements
    • Cooling methods such as evaporation have different energy and water requirements
    • Power draw has no relationship to heat removal
  6. Which examples are part of a data center’s cooling load?
    • GPU compute cores and memory banks
    • Fans, pumps, chillers, and cooling towers
    • Training jobs and optimizer states
    • Storage systems and network interfaces
  7. What monitoring can help detect liquid-cooling risks?
    • Model validation accuracy and label quality
    • Batch size and data split consistency
    • The number of cloud regions used for training
    • Coolant flow, pressure, temperature, and leak sensors
  8. Why should facility planning use measured workload profiles?
    • Measured power and utilization can differ from nameplate maximums
    • GPU specifications fully determine site impact
    • A single benchmark predicts facility demand for every workload
    • Local climate has no effect on cooling design