UPS Module Selection for Large Scale Data Centers

In specifying uninterruptible power supply (UPS) systems for the large-scale mission-critical facility, the first step is to determine the amount of load the UPS system will serve. Once load is determined, it is essential to determine the amount of redundancy that is required. System redundancy can be developed through the servers in your critical environment, typically referred to as “g...


In specifying uninterruptible power supply (UPS) systems for the large-scale mission-critical facility, the first step is to determine the amount of load the UPS system will serve. Once load is determined, it is essential to determine the amount of redundancy that is required.

System redundancy can be developed through the servers in your critical environment, typically referred to as “geo redundancy.” This refers to a system that allows for another server to pick up the work of a failed server on the data center floor. In this type of scenario, additional redundancy in a paralleled UPS system may not be required. In this case the system is referred to as an “N” redundant system.

If additional UPS redundancy is required or desired to serve the critical loads, the system can be set up in many different scenarios, which include but are not limited to the following:

N+1, N+2, N + Reserve Bus and 2(N+1)…

In larger systems, the sheer amount of load and required redundancy typically drives the need for a paralleled UPS system, which can typically be paralleled using two different methods: isolated redundant and parallel redundant.

Parallel redundant modules are the preferred method of distribution primarily because the isolated redundant systems rely on a 100% step load on the inactive UPS module. When the system is designed at N+1 redundancy, each of four paralleled modules will share 75% of the load at full loading. If and when one module fails, the remaining three UPS modules will jump up from 75% to 100% loading. Additionally, the isolated bus redundant UPS paralleling systems are inherently less efficient than the parallel redundant based on the fact that one of the modules will be energized and have no load on it a majority of the time. This is when the UPS modules are least efficient.

In addition to the determination of the amount of load to be served, the topology of the UPS system and the level of redundancy, the selection of the inverter technology is critical. I have selected a single generic UPS model for the basis of this discussion to keep the comparisons as simple as possible. The example below illustrates these criteria and the pros and cons of two different technologies available.


Let's say that each data center area serves 2,500 kW of critical load in an “N” redundant fashion. In this case, one option would be to use four 750-kVA / 675-kW UPS modules coupled (paralleled) with a piece of UPS paralleling switchgear. Additionally, three 1,000-kVA / 900-kW UPS modules could be utilized. In an “N” redundant system, these systems would equal the same amount of critical kW:

4 x 675 kW = 2,700 kW

3 x 900 kW = 2,700 kW

If each data center area was designed to N+1 redundancy, then the total capacity of the four 750-kVA/675-kW UPS modules would actually be 3 x 675 kW= 2,025 kW. In the case of N+1 redundancy, three 1,000-kVA/900-kW UPS modules would equate to only 2 x 900 kVA or 1,800 kW.

But this would not be an equitable comparison. The 750-kVA/675-kW units would provide for about 13% more load in this configuration. In each case the modules are rated at 0.9 power factor, and therefore, a similar comparison of the kVA rating could be achieved. Additionally, in the typical data center, the actual loads will probably be in the 0.95 power factor range. Therefore, the full rating of the kVA will never be utilized before all of the kWs of the UPS modules are used.

In this case with this specific load density and topology, three larger 1,000-kVA/900-kW UPS modules could provide the same load as four smaller 750-kVA/675-kW UPS units.

This is actually the break point as to where silicon controlled rectifier (SCR) inverters are utilized in lieu of isolated gate bipolar transistor (IGBT) inverters. Based on our research, UPS vendors currently do not provide UPS inverters in the IGBT format over the 750-kVA/675-kW range. To determine the suitability of the systems, a deeper look at the different technologies and a determination the pros and cons is required.


The following are the main selection criteria to evaluate the suitability of the size of the UPS modules, topology and the technology option for the inverter:

•. The potential savings in infrastructure and associated savings in labor (cost and time) and materials.

•. The physical size issues associated with the larger UPS modules.

•. The efficiency issues between the UPS modules.

•. The heat rejection of the UPS units.

•. UPS module noise and comparison between the two units.

•. Any difference in technology and associated reliability.

In the example above, the 1,000-kVA UPS modules are 178 in. wide. The 750-kVA modules with the 12-pulse rectifier and filter are 120 in. wide. In addition, the depth of the units will increase from 39 in. to 44 in.

The 1,000-kVA units in this example are approximately 150% wider and about 5 in. or 12% deeper. On the other hand, there would only be three of the larger modules in lieu of four of the smaller units. The total area of the 1,000-kVA/900-kW units is 7,832 sq. in. The total for three units would be 23,496 sq. in. The 750-kVA/675-kW units have an area of 4,680 sq. in. Four total units would occupy 18,720 sq. in. Just from a straight size, not including code required clearances, the three 1,000-kVA units take up about 25% more room than the four 750-kVA units. The electrical design engineer would have to look at the actual configuration of the UPS room, as well as code-required clearances, to see if one option would make more sense than another from the standpoint of physical size.

This physical size analysis does not include the physical size of the batteries associated with the systems. An “apples to apples” comparison of the physical size requirements of the batteries would be based on the number of minutes required at beginning of life and at 80% of life. This actual comparison of the size of the battery strings is not included in the scope of this article.

Figures 1 (p. 19) and 2 (at left) are sample efficiency curves for a 750-kVA/675-kW UPS module and a 1,000-kVA/900-kW UPS module. The 1,000-kVA efficiency curve is based on the standard design with a 12-pulse rectifier, isolation transformer and filter. The 750-kVA UPS module is based on a 12-pulse rectifier with isolation transformer and input filter. As you can see, the efficiency curves are almost identical (the curves appear to be based on a different scale, but the efficiencies at various loading percentages are very close). In this example, the systems will be based on an “N” design, and at full load, they will probably be in the 75% to 100% range of loading; in both systems they are at about 92% to 93% efficient for this loading.

Efficiency ratings are based on “non-linear” loads per published information for the UPS models under consideration. The non-linear content is not specifically mentioned in the literature but should be a good “real-life” representation of the comparison of efficiency between the two units.

Finally, the amount of heat rejection from the UPS units will dictate the amount of cooling required in the electrical rooms. This will also translate into total efficiency of the system. Based on the manufacturer's data, each of the 1,000-kVA units dissipates 231,203 BTU/h at full load. Each of the 750-kVA units dissipates 181,400 BTU/h at full load. Based on a conceptual layout of four 750-kVA units, the total heat dissipation at full load is 725,600 BTU/h.The three 1,000-kVA unit configuration would dissipate 693,609 BTU/h. This is a reduction of about 32,000 BTU/h per UPS system if we utilize the 1,000-kVA units. This represents less than a 5% reduction of heat dissipation for the larger units, but could represent life-cycle cost savings involved in cooling the UPS room over the life of the data center.


This analysis is a summary only and does not cover every difference between the two UPS module options. This is just an example of the thought process that should go into selecting the components within a critical facility. With any system, the total potential cost savings of one type of system or another needs to be evaluated for both the labor and material and will be specific to every project. The final decision should be made based on initial savings, total reliability and life cycle costs.

Two final consideration: If there are open-rack wet-cell batteries, then the input isolation transformer should limit AC fault current from the rectifier on the DC bus. Most large-scale data centers use open-rack, wet-cell battery systems. A UPS with a 12-pulse rectifier with input isolation transformer and 4% input harmonic filter really is the optimal choice for large-scale, highly reliable data centers or other critical environments.

And there is the issue of noise. The commutation circuits of the SCRs in the 1,000-kVA UPS units add perhaps 5-6 dBA of audible noise over the IGBTs in the 750-kVA UPS units during UPS operation, even when the UPS is lightly loaded. The audible noise of the 750-kVA unit is between 65-72 db, with 72 db for the 750-kVA unit (typical). The 1,000-kVA unit is documented at 75 db. The noise gap between the two systems closes slightly when one considers that three sources add 4.8 db and four sources adds 6 db to the total amount of noise.

Comparing UPS Inverters

The major difference between the two UPS models in our example is that the 1000-kVA/900-kW UPS system uses SCRs in the inverter section, while the 750-kVA/675-kW UPS module uses IGBTs.Many UPS manufacturers have changed their inverters from SCR to IGBT technology over the past five to ten years. For instance, the entire Liebert Series 600 line previously utilized SCRs, but all sizes except for the 1,000-kVA were changed to IGBTs around the year 2000. The IGBTs could not carry the current required by the 1,000-kVA/900-kW UPS modules. At the time, it was determined that paralleled IGBTs would be required for the 1,000-kVA/900-kW inverter. It is my understanding that the conversion to paralleled IGBTs has not been completed based on several reasons, one of which is over the concern over reliability of such a paralleled IGBT system. Until larger IGBTs are commonly available, you probably will not see the large 1,000-kVA/900-kW UPS units with IGBT inverters.

The 750-kVA/675-kW UPS module design is available in more configurations than the 1,000-kVA/900-kW unit. So to compare “apples to apples” for the various criteria mentioned above, the configuration for the 750-kVA/675-kW UPS module and the 1,000-kVA/900-kW UPS module should be the same. The 1,000-kVA is what Liebert considers as a classic “big iron” UPS design with 12-pulse rectifier with input isolation transformer and 4% input harmonic filter as the only available configuration. Liebert builds the 750-kVA model with either a 6-pulse or 12-pulse rectifier—with or without filter, with or without input isolation transformer—to allow them to be competitive with manufacturers who don't offer a 12-pulse.But the 1,000-kVA/900-kW is only available with the 12-pulse and with an input isolation transformer.

The 12%%MDASSML%%pulse rectifier is a more generator-friendly design than the 6-pulse. A 6-pulse rectifier will typically have about 30% total harmonic distortion (THD), rich in 5th and 7th harmonics. A 12-pulse rectifier will have less than one-half of the total harmonic distortion (THD) that we see in a 6-pulse rectifier and is rich in 11th and 13th harmonics.

An input filter is typically provided to meet IEEE 519, the THD allowed at the point of common coupling with the utility. In addition, UPS manufacturers can provide a passive filter to reduce the THD seen by the generator. In some situations, at low loading on the UPS, the static filter can provide an excessive capacitive component that will be sent back to the generator. Unlike a utility source, a generator cannot absorb the voltage rise caused by the capacitance in the system. This problem can be eliminated by evaluating the system and providing a smaller filter, sized to provide no excitation at the lowest possible load. In addition, with control circuitry, the filter can be completely removed under generator power or when the system is loaded such that over excitation can occur.

No comments
Consulting-Specifying Engineer's Product of the Year (POY) contest is the premier award for new products in the HVAC, fire, electrical, and...
Consulting-Specifying Engineer magazine is dedicated to encouraging and recognizing the most talented young individuals...
The MEP Giants program lists the top mechanical, electrical, plumbing, and fire protection engineering firms in the United States.
Economics of HVAC systems; NFPA 110-2016; Designing and choosing modular data centers
Combined heat and power; Assessing replacement of electrical systems; Energy codes and lighting; Salary Survey; Fan efficiency
Commissioning lighting control systems; 2016 Commissioning Giants; Design high-efficiency hot water systems for hospitals; Evaluating condensation and condensate
Tying a microgrid to the smart grid; Paralleling generator systems; Previewing NEC 2017 changes
Driving motor efficiency; Preventing Arc Flash in mission critical facilities; Integrating alternative power and existing electrical systems
Putting COPS into context; Designing medium-voltage electrical systems; Planning and designing resilient, efficient data centers; The nine steps of designing generator fuel systems
As brand protection manager for Eaton’s Electrical Sector, Tom Grace oversees counterfeit awareness...
Amara Rozgus is chief editor and content manager of Consulting-Specifier Engineer magazine.
IEEE power industry experts bring their combined experience in the electrical power industry...
Michael Heinsdorf, P.E., LEED AP, CDT is an Engineering Specification Writer at ARCOM MasterSpec.
click me