Equipment lifecycles and technology refresh

From The Business of Energy in U.S. Academic Research Computing
Jump to: navigation, search

Computation turns electricity into answers. Electrical power is needed to maintain data in memory, to spin hard drives, to send data across networks, and to perform digital calculations. These operational costs for computation are well understood, and part of the total cost of ownership calculations in the spreadsheet accompanying this project.


Impacts that are less often taken into account as part of the research computing lifecycle are those associated with the manufacturing of the many components that make up a modern research computer. Within a given computational node or blade, there are CPUs, memory, power supplies, networks, cables, electrical slots and connectors, circuit boards, voltage regulators, and other components. These are made from metals, plastics and glass.

A 2011 analysis by the National Resources Defense Council identified a number of substances that go into a modern computer or other device such a smartphone or tablet. Just the metals include:

  • Aluminum
  • Antimony
  • Arsenic
  • Barium
  • Beryllium
  • Cadmium
  • Chromium
  • Cobalt
  • Copper
  • Gallium
  • Gold
  • Iron
  • Lead
  • Mercury
  • Palladium
  • Platinum
  • Silver
  • Tin
  • Zinc

These metals are used together in ways that make them difficult to separate for recycling or material recovery, such as layering mercury within a plastic casing, with metallic conductors, for a LCD screen.

While a computer that might be manufactured in the US, China, or elsewhere, these raw materials and components manufactured from them come from around the world. Extraction of the components, involving mining, often uses tremendous quantities of energy and water, and might produce toxic waste or other pollutants. The US Environmental Protection Agency provides assessments of the pollution and other impacts of mining for items mined in the US, such as their copper report (EPA, 2014). Products mined primarily outside of the US might not be regulated or monitored. Recently, there has been increased awareness that rare earth minerals and heavy metals needed for technology products often come from conflict-laden regions, thus creating human and social costs. In 2014, Intel announced it would no longer use gold, tungsten, tin or tantalum from the Democratic Republic of Congo, as an initial step towards lessening the negative effects of mineral extraction (Intel, 2014).

Assessing Impact

In the near future, it seems likely that manufacturers of computing equipment will follow Intel's lead in pursuing a greener supply chain, with fewer negative human impacts. This is challenging, however, because even the most well-operated and regulated mineral extraction operation will have a large energy impact and will create wastes. In the meantime, personnel in research computing centers have the opportunity to assess materials used in creating our systems, the impact of the mineral extraction, and the energy utilized for manufacturing. When possible, end-of-life for these systems should also be assessed - particularly if components will end up in a landfill or other location where toxins could leach into the groundwater, or form gases that will vent into the atmosphere.

Simple Measures

In an ideal world, we would know all of the raw materials used to manufacturer our systems, along with their origins. In practice, only approximations are available. We would also like to know that discarded computers would be appropriately recycled, so that their components could be melted down or otherwise made available for reuse. This, too, is difficult to ascertain. In fact, recycling often occurs in the developing world, and can do social and environmental damage (e.g., Kluger, 2011). Currently, we need to make do with approximations and estimates.

An article by Williams (2004) is often still cited as relevant, and it seems there are no contemporary estimates by governments, manufacturers or industry groups on the energy it takes to produce the components for a computer. A 2009 article by Kris de Decker does a better job at identifying embedded energy in many of the components of computers, but without sources for all the numbers. We do know that producing microprocessors and similar microelectronics for computers is far more energy-intensive than other materials. Whereas producing steel from iron ore might take around 12 kiloWatt hours per kilogram, de Decker cites over 2000 kiloWatt hours per kilogram for production of electronic-grade silicon. Printed circuit boards, disk drives, and other major computer subsystems are manufactured from a variety of materials, ranging from steel, to copper, to silicon - often mounted or encased in polymers.

A 2011 article by Denga, Babbitt and Williams (the same Williams as the 2004 article) was based on complete disassembly of a laptop computer. For a 3.7kg laptop (8.1 pounds), they computed a range of 280 to 685 mega-Joules to produce it, with a resulting carbon-equivalent production of 16-41 kilograms. 1 megajoule (MJ) = 277.77 watt-hour (Wh), so an average estimate might be around 130,000 Wh for the laptop.

In a blade-type server (i.e., IBM BladeCenter, or Cray XC30 chassis), the weight of a node might be similar to a laptop. A standalone compute system used in a cluster might weigh 2 or 3 times more, due to the larger chassis, power supply, and other components such as video output components and a CD/DVD drive. In the blade-type server, power supplies and other components are still part of the system, but not necessarily part of the node itself.

In the accompanying spreadsheet, I have started with the idea that a blade-type server might be equal to two laptops from the Denga study, when all subsystems (including off-node subsystems) are taken into account. For a standalone-type node, I based my estimates on a weight of 2.5 laptops.

Moving Forward

The state of knowledge in 2014 is such that it's challenging to form an accurate estimate of the materials utilized to manufacture computers and their components. The sources of those materials are often unknown, as is their manufacturing process. As such, we can only estimate the impacts of building such systems. Similarly, we cannot know the extent to which discarded systems can be effectively recycled or reused (unless, of course, it is provided to another operator in a running state).

For total cost of ownership planning, it is incumbent on Center administrators to consider the overall impact of a system choice. This favors upgrading equipment over buying new equipment. It favors making purchase decisions with such upgrades in mind, and often allows major physical components (such as cabinets and system chassis) to be reused. Some of the largest computer consumers have arrived at similar conclusions, and in fact standardization of rack components for relatively lightweight computers is the emphasis of the OpenRack standardization effort (

Energy Roles in Procurement and Operation

Here are some ideas for how embedded energy, materials use, and lifecycle of systems might be taken into account for a system procurement decision:

  1. Require the vendor to describe power utilization under zero computational load. Ideally, any given component of the computer system should utilize zero power, and therefore produce zero heat, when not in use. Vendors employ a variety of techniques to shut down components, or reduce the power load, when not in use.
  2. Engage in a cost-risk analysis of the chilled water flow rate and temperature. Chilling and moving water is more efficient than air for removing waste heat, yet is still a substantial fraction (usually at least 10%) of a supercomputer’s energy requirements. Warmer water, and water moving less swiftly, will save energy but might result in earlier component failure due to overheating. Vendors must be asked to provide a projected failure curve, so a center can make an informed decision about set points for temperature and pressure where risks of component failures are tolerable, without expending excess energy on cooling and water pressure.
  3. Engage in a similar cost-risk analysis of the chilled airflow rate and temperature, including the set point for humidity. The new system may be the primary tenant in a room in the data center, thus that room’s environmental settings can be determined independently of other rooms where equipment might have different requirements.
  4. Include information about end-of-life, including reuse, return, and recycling.
  5. Include upgrade-ability as a criterion for the procurement decision. What components can be upgraded in place, and what ones require replacement?
  6. Specify compliance with RoHS and other environmental guidelines, even if they are optional for your location.


de Decker, Kris. (2009). "The monster footprint of digital technology." Low-tech Magazine. Online: Accessed March 28, 2014.

Denga, Liqiu, Callie W. Babbittb & Eric D. Williams. (2011). "Economic-balance hybrid LCA extended with uncertainty analysis: case study of a laptop computer." J. of Cleaner Production, 19(11): 1198–1206. DOI: 10.1016/j.jclepro.2011.03.004. Available online:

Environmental Protection Agency. (2014). "Technical Resource Document: Extraction and Beneficiation of Ores and Minerals: Copper." Online: Accessed March 28, 2014.

Intel. (2014). "Where Can You Find the World’s First Conflict-Free Processors? Look Inside.™ Online:

Kluger, Jeffrey. (2011). "The E-Waste Blight Grows More Dangerous Than Ever." Time. Online: Retrieved March 28, 2014.

National Resources Defense Council. (2011). "Your computer's lifetime journey." Online: Accessed March 28, 2014.

Williams, Eric D. (2004). "Revisiting energy used to manufacture a desktop computer: hybrid analysis combining process and economic input-output methods." Electronics and the Environment, 2004. Conference Record, pp. 80-85. Available online: