If you receive errors when attempting to view this white paper, please install the latest version of
"In today's "always on, always available" world where businesses can't stop and downtime is measured in dollars, American Power Conversion (APC)
provides protection against some of the leading causes of downtime, data loss and hardware damage: power problems
Source : APC
Making Large UPS Systems More Efficient
UPS Systems is also known as :
ups systems manufactures,
uninterruptible power supply systems,
ups for industrial users,
modern ups systems,
ups systems provides,
uninterruptible power supply,
ups backup power,
uninterruptible power supplies ups,
ups battery system,
apc ups systems,
ups surge protection,
ups system filters,
compare ups backup power systems,industrial ups systems,
modern ups systems data centers,
ups systems provides data centers,
including ups data centers,
uninterruptible power supply data centers,
ups backup power data centers,
uninterruptible power supplies ups data centers,
ups battery system data centers,
apc ups systems data centers,
ups surge protection data centers,
ups system filters data centers,
landing systems data centers.
As energy resources become more scarce and more expensive, electrical efficiency is
becoming a more important performance factor in the specification and selection of large
UPS systems. There are three subtle but significant factors that can materially affect a
company's cost of operating a UPS system and particularly the electrical bill. Unfortunately,
the people who specify systems often fail to recognize these factors, which leads to
increased costs to the owner because operational efficiencies are not correctly considered.
This paper discusses the common errors and misunderstandings in evaluating UPS
efficiency. UPS efficiency curves are explained, compared, and their cost implications
The traditional approach to the specification and selection of UPS systems has focused almost solely on
system reliability, as represented by the mean time between failure (MTBF) provided by manufacturers and
consulting engineers. Two issues are now conspiring to move efficiency, as much as reliability, to the
forefront in UPS evaluation: (1) a focus on total cost of ownership (TCO) over the lifetime of the system, and
(2) public and private environmental initiatives, as exemplified by "green building" certification programs and
demand-side management programs offered by utility companies.
There are two major contributors to UPS inefficiency: the inherent losses of the UPS modules themselves,
and how the system is implemented (i.e. right-sizing, redundancy). Oftentimes, when specifying UPS
systems, the only efficiency value considered is the best case value published by manufacturers. This is
misleading and will be explained further.
A hypothetical example is perhaps the best way to demonstrate how this practice can have a material effect
on a company's electrical expense. Consider two 1 MW UPS systems from two different manufacturers.
UPS system 1 and UPS system 2 have identical published efficiencies (93% at full load), are operated in a
2N architecture, use an electrical cost of $0.10 / kW hr, and support a 300 kW load. Many would argue that
there would be no difference in the annual electrical cost of operating these two systems. This is a flawed
statement except for emergency or maintenance scenarios, UPSs are never operated at a 100% load level
in a 2N configuration since each side of the "N" has to be capable of supporting the full load if one side fails.
Therefore, the maximum design load on each UPS in normal operation cannot exceed 50%. In reality 2N
systems rarely achieve even 50% load on each system. Some field surveys indicate that 2N data centers
operate at 20-40% of their 2N capacity.1 For this example, a typical 30% load is assumed, where each UPS
supports 150 kW. Each UPS in system 1 incurs an annual electrical cost of $10,470 in power losses vs.
$28,322 for each UPS in system 2. Since there are two UPSs in each system, the electrical losses are
doubled to $20,940 and $56,644 per year, respectively. These UPS losses manifest themselves as heat
which must be removed by the cooling system. Assuming each kW of heat requires 400 watts for the cooling
system to remove it, an additional $8,376 vs. $22,651 per year is required2. In this example, a typical data
center lifespan of 10 years, results in a total cost of UPS system losses of $293,165 vs. $793,021 as shown
in Table 1. So, how is it that the electrical losses between two seemingly identical UPS systems can differ
by almost a factor of three?
The answer lies in the efficiency curves of both UPS systems and how they are sized against the load. An
improvement of 5 percentage points in the efficiency of a single UPS can result in an electrical cost
reduction between 18% and 84% depending on how much load is on the UPS. This is illustrated later using
two UPS designs currently on the market.
In order to meet today's efficiency and environmental demands, UPS manufacturers can utilize three factors
to improve the efficiency of large UPS: technology, topology, and modularity. Together these factors can
reduce the electrical UPS losses in the form of heat energy (kW). This paper explains the efficiency curve
and will discuss common errors made in evaluating UPS efficiency. It will show how technology, topology,
and modularity allow manufacturers to improve UPS efficiency. For a discussion on full data center
efficiency see APC White Paper #113, "Electrical Efficiency Modeling for Data Centers".
UPS Efficiency Curve
If there is only one UPS efficiency number listed on a UPS data sheet, it is almost certainly quoted at 100%
load (rated load) and at various other favorable system states such as fully charged batteries, nominal UPS
input voltage, and optional input transformers and filters disconnected or not installed. The fact is that most
UPS manufacturers quote UPS efficiency at 100% load because it represents the very best efficiency the
UPS will attain. Unfortunately, very few customers will ever reap the benefits of this efficiency
because they will never reach 100% load. Specifying a UPS based on its nameplate efficiency is like
buying a car that gets maximum fuel efficiency on the highway and using it for city driving. A better way to
specify a UPS is to use the efficiency at approximately 30% load which tends to be the average load most
medium to large scale data centers operate at. To do this one must first understand what a UPS efficiency
curve is and how it is created.
Figure 1 shows the basic shape of a UPS efficiency
curve. The highest point on the curve corresponds with
the highest efficiency (Y axis) and the highest load level
(X axis). In this curve, the maximum UPS efficiency is
93%. In order to specify a UPS at a realistic load level,
the customer must find or test the UPS efficiency at a
common load level such as 30%, which on this curve is
89%. In cases where a data center uses redundant
UPSs (2N), the efficiency drops even more due to the
fact that the load is split across both UPSs which would
bring the efficiency down to 82%. This redundancy effect
is discussed later in the paper.
In the figure, the green bars represent all the power going to the IT loads while the red bars represent
internal UPS losses that define the efficiency curve in Figure 1. If a UPS had perfect efficiency, all the power
supplied to the UPS would be delivered to the data center loads resulting in completely green bars (no
losses) for all load levels. In this case, the efficiency "curve" would look like a horizontal line (100% for all
loads). However, as indicated by the red bars, some of the input power is used directly by the UPS. There
are three types of UPS losses: "no-load" losses, "proportional" losses and "square-law" losses.
At 0% load, all the input power is used by the UPS, hence the name "no-load" losses. This may also be
called other names such as tare, constant, fixed, shunt, and parallel. These losses are independent of load
and are attributed to powering such things as transformers, capacitors, logic boards, and communication
cards. No-load losses can represent over 40% of all UPS losses and are by far the largest opportunity for
improving UPS efficiency. This is discussed in more detail in the appendix.
As more load is added to a UPS, a larger amount of power must be "processed" by various components in
its power path. For example, the switching losses from transistors, and the resistance losses from capacitors
and inductors, all add to proportional losses.
As more load is added to a UPS, the electrical current running through its components increases. This
causes losses in the UPS with the square of the current sometimes referred to as "I-squared R" losses. The
power losses dissipated in the form of heat are proportional to the square of the current. Square-law losses
become significant (1-4%) at higher UPS loads.
The very nature of comparing the efficiencies of two or more UPSs means that only their losses (the red
bars in Figure 2) are evaluated. An efficiency curve alone can tell a great deal about a UPS including
quantifying its proportional, no-load, and square-law losses across all load levels. Plotting these three types
of losses relative to UPS load percentage will produce a power loss graph similar to that of Figure 3. Notice
how the no-load loss remains constant through the entire load spectrum while the proportional loss ramps
up as more IT equipment is plugged into the UPS.
Common mistakes made in specifying UPS systems
It is very easy for those who specify UPS systems to dismiss the efficiency improvement of one UPS over
another. Table 2 lists the various reasons and why they are flawed.
A business pays for the electricity that is measured by the utility meter ' This is the ultimate benchmark for
the specification of any equipment. This is why manufacturers' efficiency curve data should be based on
realistic customer installations. Furthermore, the design of a data center power system should comprehend
the efficiency impact over the entire power train and not just the UPS. A case in point is the removal of input
filters to increase measured UPS efficiency. UPSs by their very nature produce harmonics or unwanted
currents that increase heat losses in upstream wiring and transformers thereby decreasing efficiency. UPS
input filters minimize these adverse affects by attenuating the harmonic component of the alternating
current. By removing input filters to increase measured UPS efficiency, a manufacturer has in essence
moved the heat losses and their associated electrical cost further upstream. Ultimately the end user
unknowingly pays an efficiency cost penalty more than the 0.5 to 1 percentage points at full load. This is
because the UPS is typically loaded at about 30% where the filter's fixed losses weigh more heavily. For
example, at $0.10 / kW hour, assume 1 MW UPS at 30% load has a best case efficiency of 89%. If a filter is
added and drops the efficiency by 3 percentage points at that load level, the annual electrical cost increases
from $32,481 to $42,781, an increase of nearly 32%.
Perhaps the most effective method of specifying a UPS for efficiency is to request an efficiency curve from
the manufacturer that will completely describe the energy saving benefits of one UPS over another. Note
that the curve should come with input and output power data so that by using a simple spreadsheet, the
energy savings can be calculated from 0% to 100% load and every point in between. It is important that
the manufacturer's curve be based on a configuration similar to what is being specified. The
appendix of this paper provides an in depth discussion on UPS efficiency comparisons by investigating
various scenarios. The following section describes how manufacturers can improve UPS efficiency by using
various levers of design.
Improving Large UPS Efficiency
There are three significant losses that a manufacturer can lower in order to improve UPS efficiency; no-load
losses, proportional losses, and square-law losses. To do so, manufacturers have three points of leverage at
their disposal; technology, topology, and modularity. By understanding how these factors impact
efficiency, specifying engineers can better identify UPS systems that will significantly decrease the electrical
cost of operating them.
The word technology tends to overlap with topology and modularity but in this paper its meaning is
constricted to describe only the building blocks of a UPS which include the hardware and software.
Switching technology: IGBTs replace SCRs
Large solid state ("static") UPS systems work by converting alternating current (AC) to direct current (DC)
and DC to AC. Part of this process of power conversion is rapid on-off switching which leads to power losses
in the form of heat across the switch due to its inherent electrical resistance. In fact, even when a switch is
open, there is always some small amount of heat loss due to leakage current. This is analogous to the heat
generated when a rope (current) is pulled quickly through a person's hands (switch). When the rope is held
tightly (switch closed) heat is generated, when the rope is held loosely (switch open) very little heat is
Originally, the switching process was accomplished by silicon-controlled rectifiers (SCRs) which had highpower
/ high-voltage switching capabilities. SCRs were standard UPS components until the mid 1990s and
are still in use today in some older designs. They were relatively inexpensive and easy to design around, but
had serious drawbacks: worst was they tended to fail "short," which produced a short circuit at the most
critical point of the UPS ' the DC bus. Protective circuits and devices had to be added to protect the DC bus
from this failure mode ' which, in turn, lead to even more components that could (and would) fail. SCRs are
easy to turn on (a 1-2 volt signal to the gate will do it) but difficult to turn off (a reverse-bias voltage spike is
necessary). Transistors do not have this problem ' they require less power to turn on and off. Essentially
they are "on" when the gate signal is present, and "off" when it is not ' but until the mid 1990s, they were
limited in current-handling capabilities. This was solved when isolated gate bipolar transistors (IGBTs) were
introduced. Capable of higher speeds and higher power handling, IGBTs enabled the power conversion
process to be operated in a "high frequency pulse-width-modulation (PWM)" mode. High frequency PWM
reduces the size of filter components required which leads to further efficiency improvements.
Controls: DSP replacing Analog
Many manufacturers today are moving from analog controls to digital signal processing (DSP) controls. This
is analogous to switching from a traditional watch with gears and hands to a digital watch with a battery and
liquid crystal display (LCD). DSP controls are much more intelligent, can operate at much faster rates, and
therefore make many more decisions that help to improve efficiency. DSP controls also reduce the number
of components compared to analog circuits.
More advanced DSP controls can improve efficiency through intelligent adaptive switching, where the main
high frequency power switches can maintain output voltage precision with fewer loss-prone switching
transitions. For lighter loads, the reduction in switching transitions using DSP can be up to 50%, resulting in
significant improvements in efficiency. In addition, DSP controls require much less power than prior
generation controls, which allows a substantial reduction in no-load losses.
IGBT and DSP technologies are major technological improvements which have led to increased UPS
efficiency in the most recent generation of UPS products.
UPS topology basically defines how their power components are internally connected. Manufacturers can
use topology as a tool to reduce the losses for a particular application or size range. There are two principal
topologies used in large UPS systems: Double conversion on-line and Delta conversion on-line. In the case
of high-power UPS systems (over 200 kVA); a recent publication by the US Electrical Power Research
Institute found that delta conversion on-line UPS topology currently offers the greatest efficiency3 (Figure 4).
The effect of topology on UPS efficiency is explained in the following paragraphs.
In the case of delta conversion on-line systems, efficiency is improved mainly by reducing no-load losses,
but also by a reduction in square-law losses. By using the input transformer in a series arrangement, the
UPS input current and UPS output voltage can be fully regulated and controlled without having to convert all
incoming power to DC and back to AC again, as is done in a double conversion on-line system. This is
shown in Figure 5. Note that the output voltage in the delta conversion on-line UPS is fully regenerated by
the output inverter and isolated from the utility supply just like it is in a double conversion on-line UPS.
Another example of how topology reduces no-load losses is by eliminating the input filter associated with
double-conversion topology. Traditional double conversion UPSs generate high input harmonic current (from
9% to 30% total harmonic distortion) and low power factor (0.9 to 0.8). For this reason, an input filter is
added to double conversion designs, which increases the power factor, and minimizes harmonics or
unwanted current that increases heat losses in upstream wiring and transformers. Note, however, that
adding this input filter interferes with engine generator voltage regulation. By drawing sinusoidal current,
delta conversion topology generates negligible input harmonic current (less than 3%) with a unity power
factor, thus eliminating the need for an input filter altogether. Fore more discussion on differences in UPS
topologies, see APC White Paper #1 "The Different Types of UPS Systems".
Delta conversion is a good illustration of how topology can be used by manufacturers to increase UPS
efficiency and drive up energy savings, with no compromise in electrical performance. The following
comparison helps illustrate this savings.
Quantifying the topology effect
1N Topology Comparison ' Delta Conversion vs. Double Conversion
Configuration "A" is a 1 MW delta conversion on-line UPS. Configuration "B" is a 1 MW double conversion
on-line UPS. Figure 6 shows the efficiency curves, as a function of percent load, for each UPS. In both
cases, the load is assumed to be 300 kW. The efficiency of configuration "A" at 30% load is 94.9% versus
88.7% for configuration "B". This represents a difference of 6.2 percentage points in efficiency, which is a
significant cost savings over the life of the UPS.
Table 3 illustrates a 58% costs savings associated with the delta conversion topology of configuration "A"
versus the double-conversion topology of configuration "B". It should be obvious that the largest contributor
to cost for either UPS comes from the no-load losses which represent about 60% of all losses.
The costs presented in Table 3 nearly double when the same UPSs are analyzed as a redundant 2N
architecture (system plus system). The following comparison illustrates this.
2N Topology Comparison ' Delta Conversion vs. Double Conversion
Configuration "A" consists of redundant (2N) 1 MW delta conversion on-line UPS systems. Configuration "B"
consists of redundant (2N) 1 MW double conversion on-line UPS systems. The load is again assumed to be
300 kW. This means that each UPS is now loaded to only 15% since the two UPSs in each configuration
carry half of the load in normal operation. Table 4 describes the cost breakdown of this 2N scenario. Note
that for any particular UPS, although the square-law losses are halved in a 2N architecture, it doesn't offset
the doubling of the no-load losses since these kW losses are independent of the load.
Modularity is the third lever manufacturers can use to decrease energy waste. As illustrated in the efficiency
curve of Figure 5, the closer a UPS operates to its full load capacity, the more efficient it will be. Modularity
allows users to size the UPS system as closely to the load as practicable (in other words, it allows the UPS
to operate as far right on the curve as possible). A highly effective way to match capacity to load is easily
illustrated by a familiar piece of equipment in the data center ' the blade server (Figure 7).
The blade server's architecture illustrates two key design attributes that can be used to advantage in UPS
systems: it is modular, and it is scalable.
A blade server is modular in that a customer buys the frame for the blade servers and then installs standard
"blades" in the frame to achieve the amount of processing required for the application. As more blades are
inserted into the frame, it becomes a more powerful computing device. This yields a "scalable" system that
can be sized depending on computing needs.
Now, imagine a UPS system that uses modular power components in the same way. For example, suppose
a UPS chassis was capable of 1 MW of power output and as the load increased on the UPS system,
standardized power modules could be added to the system to match the desired output capacity. The UPS
could scale from 200 kW up to 1 MW in incremental steps as additional power capacity is needed. The result
is that overspending in capital is avoided ' you only buy the power components you need ' and the UPS is
working at a higher load level because the capacity of the system is more closely matched to the actual
load, which results in higher electrical efficiency. The following comparison helps illustrate this right-sizing
efficiency benefit for the same 300 kW load used in the previous examples.
Quantifying the modularity effect
1N Modularity Comparison ' Right-sized UPS vs. Oversized UPS
Configuration "A" is a 1 MW scalable delta conversion on-line UPS that is right-sized with (2) 200 kW
modules (400 kW). Configuration "B" is the same exact UPS, but oversized to 1 MW with (5) 200 kW
modules. The efficiency curve for this comparison is illustrated in Figure 8.
The graph illustrates the two points on the curve where this comparison takes place (75% load and 30%
load for configuration A & B respectively). These two points correspond to efficiencies of 96.9% and 94.9%
respectively. Table 5 illustrates the breakdown of the efficiency cost analysis for each case. While
proportional losses are equivalent, the no-load losses for the oversized UPS are 2.5 times greater than the
right-sized UPS. However, the efficiency gain of right-sizing is slightly reduced by the increase of square-law
losses which are 2.5 times greater than the oversized UPS. This is because square-law losses are more
pronounced at higher loads.
The following comparison illustrates how these savings increase further when the designs are redundant.
2N Modularity Comparison ' Right-sized UPS vs. Oversized UPS
Configuration "A" is a 2N (system plus system) 1 MW scalable delta conversion on-line UPS system that is
right-sized with (2) 200 kW modules (400 kW) in each UPS. Configuration "B" is identical to configuration "A"
except that each UPS is oversized to 1 MW with (5) 200 kW modules. Table 7 illustrates the breakdown of
the efficiency cost analysis for each case. The interesting thing to note is the that the proportional and noload
loss ratios between both UPSs are identical to the 1N modularity comparison yet the 10 year cost
savings jumps to 53%. Again, the square-law losses are the reason for this net decrease because they
represent a smaller percentage of total losses at lower loads.
Quantifying the effect of topology and modularity
The efficiency benefits of topology and modularity should be evident based on the previous series of
comparisons. But how much more could efficiency improve by combining the benefits of both modularity and
topology? The following set of comparisons quantifies this improvement.
1N Topology and Modularity Comparison ' Delta conversion Right-sized UPS vs. Double-conversion Oversized UPS
Configuration "A" is a 1 MW delta conversion on-line UPS that is scalable and right-sized with (2) 200 kW
modules (400 kW). Configuration "B" is a 1 MW double conversion on-line UPS that is non-scalable and
therefore oversized. In both cases, the load is assumed to be 300 kW. The efficiency of configuration "A" at
30% load is 96.9% versus 88.7% for configuration "B", a difference of 8.2 percentage points.
Table 7 shows a 75% savings in the cost of inefficiency by using the scalable right-sized delta conversion
UPS instead of the non-scalable oversized double conversion UPS. In this 1N architecture the total energy
cost of configuration "A" is almost four times that of configuration "B". Furthermore, the no-load losses for
configuration "A" are now reduced to 39% of all losses, nearly half the 60% for configuration "B". Figure 9
illustrates the breakdown of electrical costs due to the various losses in a 1N architecture.
The costs presented in Table 7 nearly double when configurations "A" and "B" are analyzed as a redundant
2N architecture (system plus system). In a 2N architecture the total energy cost of configuration "B" is almost
five times that of configuration "A" as shown in Table 8. In looking at Figure 9 and Figure 10, it is clear that
the cost impact of no-load losses is greater than all the others. Note that for any particular UPS, although the
square-law losses are halved in a 2N architecture, it doesn't offset the doubling of the no-load losses since
these losses represent the largest loss at almost all load levels.
From these comparisons it is clear that increasing UPS efficiency can be accomplished in two ways: by
opting for a UPS topology with a higher efficiency and by right-sizing a UPS system. In these examples
opting for a higher efficiency topology unmistakably results in the largest efficiency gain. However, this gain
requires the purchase of a new UPS which is only realistic in cases where the existing UPS has exceeded
its useful life. Alternatively, if right-sizing the UPS system were chosen as a way to increase efficiency, it
could result in the purchase of a new UPS but not always. If multiple UPS systems existed, right-sizing could
occur by migrating loads over to one or more UPS systems making it possible to turn off the unloaded
systems. This right-sizing method is also applied to air conditioning units in oversized data centers.
Figure 11 shows an example of a modular 1 MW UPS scalable in 200 kW increments. The net result is that
total cost of ownership (TCO) goes down because there are savings in the capital needed up front, and in
the expense of operating the system on a day-to-day basis.
In addition to higher electrical efficiency from
scaling UPS capacity to match the load, modular
UPS design has other attributes that contribute
significantly to availability, agility, and total cost
of ownership. For more about the advantages of
modular design, see APC White Paper #116,
Standardization and Modularity in Network-
Critical Physical Infrastructure.
Increased efficiency has secondary rewards
beyond a direct reduction in power consumption.
For example in the U.S., the Energy Policy Act of
2005 offers tax incentives for improving the energy efficiency of commercial buildings.5 Similarly, under the
Enhanced Capital Allowances (ECA) scheme, companies in the U.K. can write off 100% of capital spent on
qualifying energy efficiency equipment in the first tax year6. In some geographic areas (including many areas
of the United States), utility companies offer incentives to high-efficiency designs through demand-side
management (DSM) programs targeted at reducing overall utility demand. In such programs, efficient users
may have their electric rate reduced, or the power company may subsidize the capital cost of more efficient
technologies. These benefits further reduce TCO for power-savvy data center owners.
In order to confidently specify energy efficient UPSs, all UPS efficiency measurements must be taken under
similar conditions by different vendors and administered and approved by 3rd party test agencies. Recently,
the Lawrence Berkeley National Laboratory (LBNL) published a report on UPSs as part of their "High-
Performance High-Tech Buildings" project focused on improving energy efficiency in data centers as well as
cleanrooms and laboratories7. In this report an energy efficiency and power quality labeling scheme is
proposed for various types of UPS systems as a way of encouraging the use of higher efficiency UPSs.
There are also "green building" designations for high-efficiency designs, which single out efficient data
centers as members of a movement that is gaining high credibility in the marketplace. Companies are
finding "green" designations to be a corporate plus in their marketing messages, one they can achieve with
the added benefit of lowering operating costs. Everyone wins ' the company, their customers (through lower
product costs) and the environment. The green designation will draw increasing market recognition and
importance as energy resources become scarcer and more expensive.
Data centers consume a significant amount of power ' a fact largely ignored by the market and by
corporations. As total cost of ownership becomes a key decision factor, the differentiating value becomes
the efficiency of the systems. UPS technologies continue to evolve toward greater electrical efficiency. It is
important to remember that the true measure of success (assuming reliability standards are maintained) is
the actual efficiency that is achieved, not the details of the internal technology that accomplishes it. New
technologies may be invented, old technologies may be improved ' but from the user's perspective, it is the
efficiency curve that tells the story and, when combined with cost of equipment, provides actionable
information. If all systems are equally reliable, as most are, the sound business decision is to employ the
most efficient system possible. Contributing to a "green" corporate image, increasing agility, and simplifying
service requirements via modular design are additional benefits that underscore the soundness of that
APC White Paper #1 The Different Types of UPS Systems
APC White Paper #17 Understanding Power Factor, Crest Factor, and Surge Factor
APC White Paper #75 Comparing UPS System Design Configurations
APC White Paper #78 Mean Time Between Failure: Explanation and Standards
APC White Paper #113 Electrical Efficiency Modeling of Data Centers
APC White Paper #116 Standardization and Modularity in Network-Critical Physical Infrastructure
About the Author:
Victor Avelar is a Senior Research Analyst at APC. He is responsible for data center design and
operations research, and consults with clients on risk assessment and design practices to optimize the
availability of their data center environments. Victor holds a Bachelor's degree in Mechanical Engineering
from Rensselaer Polytechnic Institute and an MBA from Babson College. He is a member of AFCOM and
the American Society for Quality.