NCA-AIIO - AI Infrastructure (40% of the exam) - Section 2.3
Identify key concepts and high-level specifications for power and cooling.
Identify the key power and cooling concepts for GPU-dense infrastructure, including total power draw per rack, power usage effectiveness (PUE), and the difference between air cooling and liquid cooling approaches such as direct liquid cooling (DLC). Recognise the thermal design power (TDP) requirements of modern accelerated servers and how they constrain data centre facility planning.
Practice question for this objective
lock_openFree sampleAI Infrastructuremedium
A facilities team reports that their data centre has a Power Usage Effectiveness (PUE) value of 1.6. A modernisation project reduces cooling and power distribution overhead, bringing PUE to 1.15. What does this change indicate about the proportion of energy reaching IT equipment?
AThe ratio of cooling overhead to total IT load has increased, meaning more energy is now consumed by infrastructure than by compute.
BTotal facility power consumption has doubled because efficient cooling systems require more redundant components.
CThe data centre now operates at a higher ambient temperature, which lowers compute performance to compensate for reduced cooling capacity.
DA greater share of total facility power now reaches IT equipment, as overhead losses from cooling and distribution have been reduced.check_circle Correct
Interpret Power Usage Effectiveness as a measure of data centre energy efficiency and explain what a lower PUE value means for IT workloads. PUE is defined as total facility power consumption divided by the power consumed solely by IT equipment. A PUE of 1.0 would represent a theoretical ideal where no energy is lost to cooling, lighting, power conversion, or other infrastructure. A PUE of 1.6 means 60 cents of every dollar spent on electricity goes to overhead; a PUE of 1.15 means only 15 cents does. For GPU-dense deployments where compute power draw is very high, improving PUE has a large absolute impact on operating cost and carbon footprint.
Why A is wrong: A PUE moving closer to 1.0 always means overhead is decreasing relative to IT load, not increasing; a value of 1.15 leaves less overhead than 1.6.
Why B is wrong: Improving PUE does not imply doubling total power; it means a greater fraction of consumed power reaches useful IT work rather than being lost to overhead.
Why C is wrong: A lower PUE reflects more efficient energy use, not a trade-off against compute performance; modern liquid cooling can achieve low PUE values while keeping components well within thermal limits.
Why D is correct: PUE equals total facility power divided by IT equipment power; a PUE of 1.15 means only 15% of total energy is consumed by overhead, compared with 60% overhead at a PUE of 1.6. Reducing PUE toward 1.0 means progressively more energy is doing useful compute work.
Answers that look right on this material and are not. Each one is a distractor from a different question in the NCA-AIIO bank for this domain.
Direct-to-chip cooling submerges entire boards in dielectric fluid, while immersion cooling attaches cold plates to individual processors.
errorWhy it is wrong: This reverses the two approaches; attaching cold plates to processors is the defining feature of direct-to-chip, not immersion cooling.
GPU servers require more physical rack units than CPU-only servers, spreading heat across a wider area that air handlers cannot reach.
errorWhy it is wrong: GPU servers are often denser in rack height, not necessarily taller, and heat concentration rather than physical spread is the limiting factor for air cooling.
Increasing the volume of cold-aisle containment and raising computer-room air conditioner fan speeds to push more chilled air through the hot racks.
errorWhy it is wrong: Air-based cooling, even with containment and higher fan speeds, reaches practical limits at densities above roughly 20-25 kW per rack. At 30 kW or more, air cooling cannot reliably extract heat fast enough without excessive airflow velocity and noise.