Design Day. Peak PUE. Utilization. ROI.

Design Day

Data center design is predicated on theoretical worst-case conditions commonly referred to as a “design day.” Environmental conditions outside the data center are the primary variables influencing design day analysis. Design engineers use computational fluid dynamics (CFD) models to simulate the condenser inlet temperatures that the site may experience, factoring in variables such as:

·         Historical weather data, with allowance for anticipated warming trends.

·         Equipment spacing and hot air recirculation.

·         Prevailing winds.

·         Hot exhaust air discharge from generators.

·         Physical barriers to air flow.

·         Solar loads.

For critical facilities, design day analysis represents a Murphy’s Law approach to engineering – it assumes that anything that can go wrong will go wrong, and all at the same time. If a facility is engineered to withstand design day conditions, it can theoretically navigate any other foreseeable set of conditions it may encounter. This level of reliability is demanded by all stakeholders that entrust their data to the facility operator.

Cooling v IT Loads

The computational (IT) load and cooling load typically account for 85-95%+ of the power consumption at a data center. The IT load is the revenue driver for the facility. All other electrical loads, including cooling, are costs associated with delivering that IT capacity. Therefore, facilities want to have the highest possible ratio of computational load to everything else. This ratio is called the Power Utilization Effectiveness (PUE). It is the common measure of data center efficiency. PUE is calculated as follows:

The goal is to have a PUE as close to 1 as possible. Once a site is operational, PUE can be tracked to gauge operational efficiency.

For the purposes of this article, we are going to specifically focus on PUE during peak design day conditions – something referred to as Peak PUE. Peak PUE determines the upper limit of how much critical IT load the system can support during worst-case scenario (“design day”) conditions, given a certain amount of available power.

Underutilization

It is a reality of mission critical infrastructure that resiliency comes at a high price. Redundancy is expensive. Backup power is expensive. Uninterruptible power supply (UPS) systems are expensive. The capital investment to engineer and build a large, modern, redundant, resilient data center can run into the billions.

The chart below illustrates an annual distribution of ambient temperatures at Luke AFB in Arizona. Each cell value shows the number of hours during which the site will typically experience temperatures in that range during a given month. NOTE: CFD factors such as hot air recirculation and hot exhaust air from generators are not illustrated in this graphic. It is not unusual for CFD simulations of high-density data center sites to model condenser inlet temperatures in the 150°F range.

The entire site must be engineered to withstand entering air temperatures that may only occur for a handful of hours each year, if at all.  That condition determines the Peak PUE of the system. The inherent inefficiency is that very expensive capital equipment – and the electrical capacity needed to operate it – are substantially underutilized for most of the rest of the year:

This is essentially the same challenge faced by electrical grids everywhere. Grid operators must invest in enough generation capacity to serve the needs of the entire grid during the highest extremes of demand – on the hottest days of the summer in most places. But that capacity and investment is significantly underutilized for much of the rest of the year. Electrical utilities attempt to mitigate this inherent inefficiency in two ways:

1.      Demand and capacity charges to disincentivize consumption during peak demand periods.

2.      Incentive programs to encourage the implementation of demand reduction strategies and technologies.

Similarly, strategies and technologies exist that allow data centers to cut the peak off their demand profile (lower peak PUE) and mitigate the underutilization of expensive capital equipment. Adiabatic precooling technology reduces condenser inlet air temperatures by as much as 40°F, insulating the system from summertime temperature extremes. For design engineers and operators, the uncontrolled variable of ambient air temperature becomes a controllable design parameter which allows for the site to be engineered with minimal seasonal underutilization:

The value of fuller utilization

Lowering Peak PUE and reducing seasonal underutilization creates value for data center owners in one of two ways:

1.      Lowering capital expenditure by enabling a reduction in the size and/or count of cooling and backup power equipment, or

2.      Increasing revenue-generating IT load given a limited amount of available power.

Power availability is the single biggest constraint facing the data center industry. Any power that is allocated to non-IT electrical loads is power than cannot be allocated to IT loads. As such, relatively small gains in lowering Peak PUE translate into significant financial gains if it means that the power can be reallocated to supporting IT loads.

Consider the following scenario for a data center site with 300 MW of total available power:

What would be the implications of incorporating a technology into the design that lowered Peak PUE by just 0.1 (or 6-7%)?

For this scenario where there is 300 MW of total available power, it would mean that 14 MW of power could be re-allocated from non-IT load to revenue-generating IT load.

The natural follow-on question is: “What is the value of 14 MW of IT load?”

Like any other commercial building lease, there is substantial variability in data center lease rates according to geography, service level, etc. With that acknowledged, $175,000 per megawatt per month is a reasonable average based on reported lease rates during 2024. Here is the math using those assumptions:

It is worth highlighting that this represents incremental revenue, generated from the exact same capital equipment configuration (with the addition of a precooling technology). It is also worth pointing out that this primary CAPEX return on investment is multiplied by an operational return on investment that comes from ongoing savings on energy and demand charges. With many electrical utilities enacting policies that charge data centers for unutilized capacity, this important value proposition has never been more significant.

Next
Next

Enhanced Heat Rejection for High Density Data Center Design