Sign in with
Sign up | Sign in

Power Savings: Advanced Configuration And Power Interface

Power Supply Reference: Consumption, Savings, And More
By

As power-management techniques continued to develop, maintaining the complex information states necessary to implement more advanced functions became increasingly difficult for the BIOS. Therefore, another standard was developed by Intel, Microsoft, and Toshiba. Called Advanced Configuration and Power Interface (ACPI), this standard was designed to implement power-management functions in the OS. Microsoft Windows 98 and later automatically use ACPI if ACPI functions are found in the system BIOS. The need to update system BIOSs for ACPI support is one reason many computer vendors have recommended performing a BIOS update before installing Windows 98 or later on older systems.

ACPI was initially released in 1996 and first appeared in the Phoenix BIOS around that time. ACPI became a requirement for the Intel/Microsoft “PC’97” logo certification in 1996, which caused developers to work on integrating ACPI into system designs around that time. Intel included ACPI support in chipsets starting with the PIIX4E South Bridge in April 1998, and ACPI support was included in Windows starting with the release of Windows 98 (June 25, 1998) as part of what Microsoft called its “OnNow” initiative. By the time Windows 2000 came out (February 17, 2000), ACPI had universally replaced APM on new systems. ACPI 4.0a was released in April 2010, and the ACPI 5.0 Specification is currently under development. The official ACPI specifications can be downloaded from www.acpi.info.

Placing power management under the control of the OS enables a greater interaction with applications. For example, a program can indicate to the OS which of its activities are crucial, forcing an immediate activation of the hard drive, and which can be delayed until the next time the drive is activated for some other reason. For example, a word processor may be set to automatically save files in the background, which an OS using ACPI can then delay until the drive is activated for some other reason, resulting in fewer random spin-ups of the drive.

ACPI goes far beyond the previous standard, APM, which consisted mainly of processor, hard disk, and display control. ACPI controls not only power but also all the Plug and Play (PnP) hardware configuration throughout the system. With ACPI, system configuration (PnP) and power-management configuration are no longer controlled via the BIOS Setup; they are instead controlled entirely within the OS.

ACPI enables the system to automatically turn internal peripherals on and off (such as CD-ROM drives, network cards, hard disk drives, and modems) as well as external devices such as printers, monitors, or any devices connected to serial, parallel, USB, video, or other ports in the system. ACPI technology also enables peripherals to turn on or wake up the system. For example, a telephone answering machine application can request that it be able to respond to answer the telephone within 1 second. Not only is this possible, but if the user subsequently presses the power or sleep button, the system only goes into the deepest sleep state that is consistent with the ability to meet the telephone answering application’s request.

ACPI enables system designers to implement a range of power-management features that are compatible with various hardware designs while using the same OS driver. ACPI also uses the Plug and Play BIOS data structures and takes control over the Plug and Play interface, providing an OS–independent interface for configuration and control.

ACPI defines several system states and substates. There are four Global System states, labeled from G0 through G3, with G0 being the fully operational state and G3 being mechanically turned off. Global System states are immediately obvious to the user of the system and apply to the entire system as a whole. Within the G0 state, there are four CPU Power states (C0–C3) and four Device Power states (D0–D3) for each device. Within the C0 CPU Power state, there are up to 16 CPU Performance states (P0–P15).

Device Power states are states for individual devices when the system is in the G0 (Working) state. The device states may or may not be visible to the user. For example, it may be obvious when a hard disk has stopped or when the monitor is off; however, it may not be obvious that a modem or other device has been shut down. The Device Power states are somewhat generic; many devices do not have all four Power states defined.

Within the G1 Global Sleep state are four Sleep states (S1–S4). The G2 Global Soft Off state is also known as the S5 Sleep state, in which case the system is powered off but still has standby power. Finally, G3 is the Mechanical Off state, where all power is disconnected from the system.

The following list shows the definitions and nested relationship of the various Global, CPU/Device Power, and Sleep states:

G0 Working—This is the normal working state in which the system is running and fully operational. Within this state, the Processor and Device Power states apply. The Device Power states are defined as follows:

  • G0/D0 Fully-On—The device is fully active.
  • G0/D1—Depends on the device; uses less power than D0.
  • G0/D2—Depends on the device; uses less power than D1.
  • G0/D3 Off—The device is powered off (except for wakeup logic).


The Processor Power states are defined as follows:

  • G0/C0 CPU On—Normal processor operation.
  • G0/C1 CPU Halted—The processor is halted.
  • G0/C2 CPU Stopped—The clock has been stopped.
  • G0/C3 CPU/Cache Stopped—The clock has been stopped and cache snoops are ignored.


G1 Sleeping
—The system appears to be off but is actually in one of four Sleep states—up to full hibernation. How quickly the system can return to G0 depends on which of the Sleep states the system has selected. In any of these Sleep states, system context and status are saved such that they can be fully restored. The Sleep states available in the Global G1 state are defined as follows:

  • G1/S1 Halt—A low-latency idle state. The CPU is halted; however, system context and status are fully retained.
  • G1/S2 Halt-Reset—Similar to the S1 sleeping state except that the CPU and cache context is lost, and the CPU is reset upon wakeup.
  • G1/S3 Suspend to RAM—All system context is lost except memory. The hardware maintains memory context. The CPU is reset and restores some CPU and L2 context upon wakeup.
  • G1/S4 Suspend to Disk (Hibernation)—The system context and status (RAM contents) have been saved to nonvolatile storage—usually the hard disk. This is also known as Hibernation. To return to G0 (Working) state, you must press the power button, and the system will restart, loading the saved context and status from where they were previously saved (normally the hard disk). Returning from G2/S5 to G0 requires a considerable amount of latency (time).


G2/S5 Soft Off
—This is the normal power-off state that occurs after you select Shutdown or press the power button to turn the system off. The system and all devices are essentially powered off; however, the system is still plugged in and standby power is coming from the power supply to the motherboard, allowing the system to wake up (power on) if commanded by an external device. No hardware context or status is saved. The system must be fully rebooted to return to the G0 (working) state.

G3 Mechanical Off—Power is completely removed from the system. In most cases this means the system must be unplugged or the power turned off via a power strip. This is the only state in which it is safe to disassemble the system. Except for the CMOS/clock circuitry, power consumption is completely zero.

In normal use, a system alternates between the G0 (Working) and G1 (Sleeping) states. In the G1 (Working) state, individual devices and processors can be power-managed via the Device Power (D1–D3) and Processor Power (C1–C3) states. Any device that is selectively turned off can be quickly powered on in a short amount of time, from virtually instantaneous to only a few seconds (such as a hard disk spinning up).

When the system is idle (no keyboard or mouse input) for a preset period, the system enters the Global G1 (Sleeping) state, which means also selecting one of the S1–S4 sleep states. In these states, the system appears to be off, but all system context and status are saved, enabling the system to return to exactly where it left off, with varying amounts of latency. For example, returning to the G0 (Working) state from the G1/S4 (Hibernation) state requires more time than when returning from the G1/S3 (Suspend) state.

When the user presses the power button to turn the system off or selects Shutdown via the OS, the system enters the G2/S5 (Soft Off) state. In this state, no context is saved, and the system is completely off except for standby power. Fully disconnecting AC or battery power causes the system to be in the Global G3 (Mechanical Off) state, which is the only state in which the system should be disassembled.

During the system setup and boot process, ACPI performs a series of checks and tests to see whether the system hardware and BIOS support ACPI. If support is not detected or is found to be faulty, the system typically reverts to standard Advanced Power Management control, which is referred to as legacy power management under ACPI. Virtually all ACPI problems are the result of partial or incomplete ACPI implementations or incompatibilities in either the BIOS or device drivers. If you encounter any of these errors, contact your motherboard manufacturer for an updated BIOS or the device manufacturers for updated drivers.

Display all 35 comments.
This thread is closed for comments
  • 1 Hide
    de5_Roy , January 11, 2012 4:07 AM
    very informative!
  • 0 Hide
    palladin9479 , January 11, 2012 4:47 AM
    Holy cow. Thanks for that Asus PSU link. I now know what's causing my system instability.

    AMD Phenom II x4 980BE OC'd
    4 x 4GB DDR3-1600 memory
    2x NVidia GTX-580 SLI'd
    4x SATA HDD's
    1x SATA DVDRW
    7x FANs (Water cooled system)

    Comes to 1150W recommended. I have a Corsair HX-1000 1000W PSU.
  • 0 Hide
    sincreator , January 11, 2012 4:57 AM
    Still running a Thermaltake 750w toughpower here. Been 5/6 years now. Man this PSU has seen some upgrades. lol. I'll probally buy another toughpower/Corsair sometime in the near future.(If this one ever dies. lol)
  • -4 Hide
    Dacatak , January 11, 2012 5:41 AM
    Still using the same Enermax Liberty 500W from 2006 for my new Sandy Bridge upgrade with GTX 560Ti.
    The only reason you'd need more than 500W is if you need to power more than one GPU.

    Of course, as stated in the article, not all 500W PSUs are equal. The Enermax Liberty was among the best 500W PSUs in its day, and its quality is still exceptional even by today's standards.
    It has dual 12V rails with 22A on each with a combined output of 32A total. Most of the dual-rail 500W PSUs sold nowadays max out at 18A per rail.

    The Enermax was definitely ahead of its time, and in general, PSUs sold directly by their manufacturer (OEMs such as Enermax, FSP, Kingwin, Seasonic) tend to be of superior quality than those sold by third-party rebranders (Antec, OCZ, Thermaltake, Corsair, etc.).
  • 7 Hide
    cumi2k4 , January 11, 2012 8:22 AM
    Was wondering about power cycling and thermal shock... The article said that thermal shock from powering on & off can cause deterioration in a system. You suggest S3 (Suspend to Ram), but does this also cause thermal shock to the system when resuming from sleep mode?
  • 2 Hide
    lordvj , January 11, 2012 11:59 AM
    ^ this. was wondering the same thing
  • 2 Hide
    jaquith , January 11, 2012 2:37 PM
    Great article and thanks, it'll 'hopefully' make my job easier in the Forum and stop the silly arguments I have recommending PSU's. I really wish folks would stop skimping on their PSU's on nice systems.

    Another important point that folks have a tendency to forget is 'electrolytic capacitor aging' which over time takes their once 650W and after a year or so reduces it to 520W~500W aka Capacitor Aging.

    Great PSU Sizer -> http://www.thermaltake.outervision.com/
    Peak:
    100% CPU Utilization (TDP)
    100% System Load
    30%~35% for Capacitor Aging
  • 0 Hide
    zak_mckraken , January 11, 2012 2:40 PM
    @cumi2k4 and lordvj : We can only assume it does cause a thermal shock, since only the RAM retains power in S3 mode. The other unpowered components thus cool down during stand by mode, like a regular shutdown.

    Very informative article by the way!
  • 1 Hide
    TeraMedia , January 11, 2012 2:40 PM
    @palladin9479:

    Yeah, me too! I had significantly underbudgeted power for fans (9), ODD/HDDs (8) and USB devices (3), and was going nuts trying to figure out why the system was unstable at times. I thought I had a bad MoBo, or HDDs, or GPU, or ??!?!@#$? Now I know.
  • 1 Hide
    xenol , January 11, 2012 3:04 PM
    I'm kind of suspect about the ASUS power supply link. It tells me for my old system, I should get a 600W power supply but I ran a 500W on it for years without problem.
  • 0 Hide
    Onus , January 11, 2012 3:35 PM
    The statement about third party rebranders depends on who the OEM is. If Seasonic or Delta makes it (e.g. most Antec units), it is going to be a good PSU. Many Corsair and XFX are made by Seasonic too. Channel Well, Sirtec, and some others have some units that aren't so great.

    I found the article of some interest (and will revisit the sleep settings on my own system), but some of it was also years out of date. That's probably hard to avoid on a writing project of this magnitude.
  • 5 Hide
    chaz_music , January 11, 2012 3:38 PM
    Good collection of interesting PSU topics. I especially liked the ACPI information. I have several comments and suggestions to change in the article though. I work in the PSU industry and can shed some light on a few issues.

    On efficiency, most people leave out the fact that we tend to use air conditioning here in the USA a good part of the year. Here in the mid Atlantic, we tend to use A/C for about ~ 7 months annually. This adds a thermal penalty to any heat that you dump into the office/home air during those months. With most A/C systems, the cost to remove 1W of heat is an additional 0.5W of A/C power (50% overhead). Taking the above numbers and some rounding, I use an overhead rating of 30% total for any heat dumped into my home / office. So take your power loss numbers and multiply by 1.30 to get the total cost impact to your wallet. This also should be done for using CFL and LED lighting. They are not allowed to use A/C cost in their advertising, so the public does not get to see the true possible savings.

    There are several types of UPS systems that you should write about. The one you outlined is called a double conversion unit, which is always processing the power to give a clean regulated sine wave output. These are the least efficient and most expensive though. Double conversion is always taking the AC input, making DC, and using a PWM inverter to make regulated AC again for the output. Double conversion efficiencies are typically around 88-90% efficient, so this can impact you total system efficiency and operational costs. A cheaper UPS is the standby type, which allows the raw utility power to go straight to the load with some light duty surge clamping in between. When the input power voltage goes out of bounds, there is a switch over that is usually around 4-8msec which is faster than the PSU hold up time of 20msec. Since normal operation is straight pass through, the usual efficiency is close to 100% (minus the UPS internal power needs and charging). Note though that some UPS systems are crap and can use upwards to 100W just being plugged in.

    I did not follow your discussion on the alarm buzzer indicating overcharge, which should never happen in any UPS. Most modern UPS system implement a battery test to make sure that the battery capacity and internal resistance is able to hold up the load. If the battery fails, they set off the buzzer. In almost all UPS systems, a buzzer alarm is critical - something is wrong. Some UPS systems also monitor the ground feed continuity and will alarm if the input feed ground starts to float making the UPS and the load unsafe to touch.

    The UPS output waveforms are not all sine wave. Often the double conversion types are sine wave, adding to their cost. The standby UPS systems are usually step wave which is also called quasi-sine which is marketing term for step wave (to confuse the buyer). Most PC loads and monitors work fine with step wave (and are even more efficient on step wave!), although some PFC PSUs have problems. Magnetic loads can have real heartburn with step wave (motors, transformers) due to high losses and non-sinusoid voltage waveform effects.

    Ferroresonant transformers are good voltage regulators, but the way they work is very lossy. A good ferro will only run around 90% efficiency. If your load is attached to a ferro, you are adding another power loss in your system. In my opinion, you are better off spend a few more dollars and getting a UPS (which there are ferro types still out there also).

    There is no mention of oversizing your PSU also. Many HTPC and SOHO/home server needs are on 24/7 so power usage and efficiency are paramount to the cost of use / ownership. If you install an oversized PSU, you are taking a efficiency hit (for most brands) that increases your energy usage. The 80 Plus standards do not test below 20% load, so the efficiency of most PSU designs drop off quickly below 20% load. I have seen several that are below 50% with 10% loading. A good analogy on oversizing that I have used before is thinking about car engines. You cannot get a V8 car engine to run as efficiently as a 4 cylinder due to the physics (more friction/mass, etc.). That same effect occurs in a PSU. Larger magnetics, power devices, and other overhead lowers the efficiency at low power. proper sizing can save a good bit of money. Just don't get it too small, especially thinking about system start up (HDD spin up, fans, CPU local PSUs ramping up, etc.).

    You comment on thermal shock is great, but there are many other factors to consider in reliability. Spinning down any HDD and fan loads reduces bearing wear for those mechanical parts. But keeping the main motherboard PCB powered and some operation continuing also helps with reliability. The minor amount of heat that is generated helps keep the PCB dry (PCB material is hydroscopic!), which one major part of the high voltage area in a PSU failing after a long storage (like right after purchase) causing a DOA. And as others pointed out in the comments, allowing the system to go into a sleep state will also cause a cool down thermal shock. The biggest problem with thermal shock is that it break solder joints and helps break bond wires/connections in ICs. It also speeds up electrolytic cap leaking and shortens the life. Does anyone remember the motherboard cap failure from a few years ago?

    The absolute largest cause of computer failures is caused by ESD damage. The data from companies that keep statistics on this unanimously show this as a fact, but the PC enthusiast industry does not work to educate the end users of this well at all. In the electronics industry as a whole, ESD accounts for nearly 55-60% of all failures! This includes component suppliers, etc. So if you want a great topic for a future article, tackle ESD. It is real and it is very costly when ignored. Ever had a PC part that was DOA, i.e., that just "did not work at all" when powered up the first time and would not work at all? Good chance it was ESD.

    Thanks for the article.
  • 3 Hide
    george21546 , January 11, 2012 3:52 PM
    Buy a power meter kill-o-watt comes to mind. Cost 15-20 and will tell you amps, watts, power factor and cycles per second. Best of all it will measure watts over time so you can check how much your system is using in each of it's states. I like to oversize power supplies by 25% unless upgrades are planned.
  • 0 Hide
    chris maple , January 11, 2012 6:48 PM
    The low ends of the ranges shown are too high. Discrete video cards are available that use less than 10 watts, same for hard drives. Motherboards rarely exceed 25 watts.
    My system has an Intel Core I7-870, discrete video card, 2x2G RAM, 2 1Tbyte hard drives, an SSD and a DVD burner. It usually runs at 70 watts and has never exceeded 200 watts driven hard.
  • 0 Hide
    ethaniel , January 11, 2012 8:49 PM
    I thought it was only -5% tolerance for the -/+12v rail. Good data.
  • 0 Hide
    BlackHawk91 , January 11, 2012 10:22 PM
    Would enabling the S3 sleep mode interfere with OC settings and/or performance?
  • 0 Hide
    hardcore_gamer , January 12, 2012 12:12 PM
    palladin9479Holy cow. Thanks for that Asus PSU link. I now know what's causing my system instability.AMD Phenom II x4 980BE OC'd4 x 4GB DDR3-1600 memory2x NVidia GTX-580 SLI'd4x SATA ......


    You have a serious bottleneck there bro ;) . Time to upgrade the CPU.
  • 0 Hide
    g-unit1111 , January 12, 2012 10:50 PM
    palladin9479Holy cow. Thanks for that Asus PSU link. I now know what's causing my system instability.AMD Phenom II x4 980BE OC'd4 x 4GB DDR3-1600 memory2x NVidia GTX-580 SLI'd4x SATA HDD's1x SATA DVDRW7x FANs (Water cooled system)Comes to 1150W recommended. I have a Corsair HX-1000 1000W PSU.


    Yeah... that floored me as well, mine is 900 minimum.

    1 x AMD Phenom II X6 1055T OC'd
    2 x Geforce GTX 550TI
    4 x 4GB DDR3
    1 x SSD
    2 x HD
    2 x DVD-RW
    5 x CPU fans (double heat sink)

    I know now what's causing most of my heat issues is that I'm running an underpowered PSU (Corsair 750). I will definitely make this my next upgrade.

    And that thing about putting systems to sleep, I'll do that more often.
  • 0 Hide
    palladin9479 , January 12, 2012 11:57 PM
    Remember that ASUS link is calculating the approximate maximum power draw possible on your system. Basically with everything going full blast which doesn't happen too often.

    PSU's in general start to get stressed once their over 80% of their rated output. Prolonged stress can cause components to wear out much earlier then before. This is why a PSU may be fine for awhile but then start to have random issues six months or more after installation. I just didn't think I was burning that much juice, but now it seems I am.
  • 0 Hide
    A Bad Day , January 13, 2012 2:51 AM
    Just a question, is it worth watercooling a PSU? I know it would boost efficiency and allow it to put out higher watt than specified, but is it worth it?
Display more comments