Overview of Power Management for 3rd generation Ultrabook Platform, Haswell

Discussion in 'Laptops' started by IntelUser2000, Apr 24, 2012.

  1. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    This post will attempt to explain the power management features in the upcoming Ultrabook platform based on Haswell. It will be divided into few sections with each post dedicated per section.

    1. Brief introduction to power management and history
    2. Role of the Ecosystem, and its limitations
    3. Windows 8, the software key to ultra low power
    4. Turbo Mode and Configurable TDP Explained
    5. Steps taken to achieve 2013 power management
    6. And Finally, Haswell

    1. Brief introduction to power management and its history


    Power management in current mobile systems are fruits of multiple generations of development. From the very simple catastrophic throttling systems to more advanced frequency/voltage scaling techniques, vast range of techniques are used to keep power usage and thermals at check for today's high performance mobile devices like Ultrabooks.

    P-States: P-states refer to various frequency steps going from LFM(Low Frequency Mode) all the way to Base. It doesn't include Turbo Mode.
    C-States: C-states refer to various power states when in idle. Transitioning between C-states take time. The lower power the C-state(increasing numbers from C0 to Cx), the longer it takes to transition in and out of the state.

    Notable chips for power management

    Pentium III – The mobile 0.18 micron Pentium IIIs offered its first SpeedStep technology for Notebooks. Code-named Geyserville, It allowed CPU to run at lower frequency when in battery mode.

    Second generation of Geyserville in Pentium III-M and Pentium 4-M allowed switching between two states when in battery depending on load. Marketing started to call it Enhanced SpeedStep.

    K6-III – PowerNow! Technology for AMD was much more advanced than earlier versions of SpeedStep. It allowed various frequency grades depending on the workload to optimize performance vs power. For Intel chips, the version of Geyserville offered in Pentium M chips offered multiple speed grades, finally catching up to AMD.

    Pentium M – Extended clock gating to execution units like ALUs and FPUs. Core 2 further extended clock gating on logic units

    Nehalem – Introduced Power Gating to allow affected circuits to power down to completely turn off. Turbo Mode introduced this generation. Turbo Mode works in tandem with Power Gating to vary frequency depending on power usage and amount of cores active. Power Control Unit, or PCU(based on a 486-class MPU) is put on die to improve accuracy and performance for thermal/power monitoring. Dynamic Acceleration in Penryn chips tried to achieve similar effect, but didn't work well in practice.

    Sandy Bridge – Turbo Mode 2.0 allows the CPU to use thermal headroom to further increase clock speed potential. Overall platform idle power better managed compared to Nehalem architecture chips.

    Ivy Bridge – Introduced Configurable TDP, allowing manufacturers to change the fixed TDP using a trigger. For example, better cooling would allow higher cTDPup to work, and a Tablet mode might need to use cTDPdown to meet thermal requirements.

    C0: Active core execution instructions
    C1/C2: Stops the core from executing instructions, lower voltage. C2 further lowers voltage. Doesn't exist in all processors. Core clock is turned off.
    C3: Flushes L1 cache, lower voltage. Doesn't exist on all CPUs.
    C4: Partially flushes the L2 cache and turns off the PLL. Introduced with Pentium M.
    C6(Deep Power Down): Introduced with Penryn, flushes and turns off L1 and L2 caches, clock and PLL. Lowers voltage down significantly. With Nehalem, power gating allows voltage to go to zero
    C7: Extension of C6 to work for the L3 caches

    Various concepts to keep power and thermals in check were introduced at a platform level

    -Display Power Management Technology, or DPST/Refresh Rate Switching that allows dynamic refresh rate changes to lower display power
    -Zero Power ODD to lower power used by Optical drives,
    -Super LFM to lower P-state power as low as possible
    -DIPM and HIPM(Device and Host Initiated Power Management) on SSDs to lower idle power
     
    #1 IntelUser2000, Apr 24, 2012
    Last edited: Apr 24, 2012
  2. Loading...

    Similar Threads - Overview Power Management Forum Date
    Replaced laptop battery and power cord, still no power? Laptops Oct 9, 2018
    Need suggestion input on powerful laptop for coding/dev Laptops Aug 13, 2018
    [Solved]laptop screen goes wonky but only on battery power. Any advice? Laptops Jun 27, 2018
    Laptop no power Laptops Feb 17, 2018
    Direct email to Dell product managers? Laptops Mar 28, 2017

  3. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    2. Role of the Ecosystem, and its limitations

    While countless power management technologies surround the modern laptop system, the steps they take are merely evolutionary. To enable a vision of future Ultrabooks(and in a more limited fashion, regular Laptops) to be a truly portable device like Tablets, more through and radical approach is needed.

    The Problem with the PC ecosystem:

    -The PC ecosystem has been evolved around a very open, flexible standard. It's relatively easy to be part of the PC ecosystem. The computer giants like Dell started by simply assembling a computer system using commonly available parts. There was no incentive to really change the computing space, for the most part.
    -Being open and flexible has its disadvantages. Changing standards are difficult and take time. In Cellphones and Tablets, proprietary hardware and software allows most optimal performance and power management. No need to go gather everyone, no conflicts, just the way you want it.

    Imagine how hard it is to get everyone towards one standard(especially a field as big as the computer industry!). The CPU vendors want it one way, the memory vendors want to do it the another way, all can get in the way. It's like trying to get an entire city to march uniformly like well-trained soldiers.

    As you will get to realize, getting everyone and everything to work together is a necessity towards creating a low power system. Merely researching for deeper and deeper sleep states aren't the answer. The key is for everyone to work together towards one goal. That's the real challenge. In case you guys didn't know, CPU isn't the problem as illustrated here: [​IMG]

    In fact, Intel said the same at Fall 2011 IDF:
    [​IMG]

    By the time everything is ready, Intel and its partners would have been working at it for 5 years...
     
    #2 IntelUser2000, Apr 24, 2012
    Last edited: May 7, 2012
  4. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    3. Windows 8, the software key to ultra low power

    The upcoming Windows OS from Microsoft, Windows 8, is a big part in allowing ultra low power to happen. When you search for Windows 8, you will get to hear about something called

    Connected Standby.
    This video gives you a great overview of Connected Standby for Windows 8: http://channel9.msdn.com/Events/BUILD/BUILD2011/HW-456T

    Connected Standby is similar to a feature in Smartphones/Tablets where its in a really low idle power but still allows incoming calls to be received and feeds(like from Twitter) to be updated. Updating requires periodically waking up the system and go back to sleep quickly when done.

    Microsoft's requirement for Connected Standby: Less than 5% battery use over 16 hour. 5% of battery use over 16 hour using conventional 56WHr battery equals to 175mW. It has to be LESS than that.

    My sig's Atom Z520 UMPC lasts only 10 hours with the 24WHr battery with the display off. That's 2.4W. There exist a hard limit of idle power for x86 platforms. Ultrabook platforms use 3-4W with display off and display adds another 3W.

    The key to achieving ultra low idle power is in modifying how Interrupts work.

    On current Windows 7 based Sandy Bridge/Ivy Bridge(and other x86 chips) platforms, the OS has a timer to periodically wake up the CPU(and everything else), using Interrupts. By making every part of the system work together and coalesce all the interrupts, the amount of time the CPU and the system can be in idle can increase. Because idle time is longer, it can further save power by transitioning into lower C-states than ever before: [​IMG]

    (Just imagine that each gaps allow the CPU to enter idle state. The longer the gap, the longer the time the CPU is in idle, and the lower power state it is in)

    Windows 8 moves from a periodic, timer based interrupt in previous OSes to a new, activity based interrupt. That significantly increases the amount of idle time for the CPU.
     
    #3 IntelUser2000, Apr 24, 2012
    Last edited: Apr 24, 2012
  5. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    4. Turbo Mode and Configurable TDP Explained

    Now, I will deviate from the topic a bit and explain about Turbo Mode and Configurable TDP. The former was introduced in Nehalem, the latter with Ivy Bridge(the systems based on it aren't available yet).

    How do we differentiate between Turbo Mode, Configurable TDP, EIST, and Throttling? Aren't they one and the same?

    Throttling: It's there simply to protect the CPU. If the CPU has a 25W TDP but the system can only take 17W, it will get hot and throttle. Throttle brings two disadvantages: Lower performance, and hotter system. The reason the latter happens is because the CPU constantly tries to reach maximum performance within the thermal limit. Throttling is also bad because its not controlled.

    EIST: Enhanced Intel SpeedStep, or EIST is there to save power usage when in light load. Not all applications need full speed from the CPU. For example, the FFDShow for QuickSync only runs the CPU at LFM: http://www.anandtech.com/show/4895/intel-engineer-ports-quicksync-video-decoding-to-ffdshow

    Turbo Mode: Turbo Mode only activates when OS asks for P0 P-State. Speeds above Base at P0 is controlled by the CPU, not the Operating System. Turbo Mode is an opportunistic, temporary speed boost system. Speeds are not guaranteed, and works off Base frequency. Turbo Mode 1.0 in Nehalem works solely within power limits, while Turbo Mode 2.0 takes advantage of both power and thermal headroom to boost. The duration is limited by two factors:

    -Power/Thermal
    -Arbitrary timer

    Because of that, in mobile systems the max Turbo Mode expires before it reaches the full Timer limit. The duration also depends on application. Think of Turbo Mode as a light load usage boost. In browsing, you are unlikely to encounter a scenario where Turbo Mode can't go to full speed. But in rendering like Cinema4D(and the benchmark Cinebench) it won't be able to maintain speeds.

    Configurable TDP: Also known as cTDP. There's a cTDPup and cTDPdown. When a certain outside factors triggers it, the CPU with the feature will “fix” itself to different TDP. Configurable TDP chips are also said to have the same max Turbo Boost ceiling, so there's little/no impact for light load usage. But a cTDPdown system won't be able to sustain the Turbo clock as long as Base, and Base won't be able to sustain it as long as a cTDPup system.

    Triggers include:

    -Fixed: Using a dock for cTDPup, and cTDPdown for going from a Clamshell to a Tablet mode.
    -User switch: Perhaps a Turbo Button?
    -Vendor specific switch: May be application dependent

    cTDP is shown in this slide:[​IMG]

    As you can see, the cTDP down chip set to 14W TDP loses Turbo frequency quicker than the Base 17W set one in Cinebench, but there's no difference in a light load Sysmark benchmark.
     
    #4 IntelUser2000, Apr 24, 2012
    Last edited: Apr 24, 2012
  6. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    5. Steps taken to achieve 2013 power management

    As I said earlier, it would have took Intel and its ecosystem partners 5 years for this level of power management to be achieved. The target platform is Shark Bay with the Haswell microarchitecture in 2013, so since 2008 when first Atom chips were introduced the efforts have been going on.

    There were signs:
    -Haswell was said to have “Revolutionary” power management. That probably refers to modifying how interrupts work as it hasn't been touched in years, along with FIVR
    -IDF Fall 2011 claimed Haswell would achieve “All-day Battery life” and 10-day Connected Standby
    -More subtle ones like this excerpt from Intel Labs: http://techresearch.intel.com/ProjectDetails.aspx?Id=136

    The way I see it there were other steps taken. For example.

    -The first generation Smartphone Atom platform Moorestown, and its largely based off Oak Trail was unable to reach battery life anywhere near claimed. I put the blame on the ecosystem. Android advances fast, and being vendors unwilling to work with Intel to optimize for their platform, they weren't able to get the work needed to lower power(specifically platform idle power).

    With Medfield Intel created their own phone! Little before that Google announced that future Android versions will treat IA chips as same class citizen. Intel must have realized during Smartphone hardship that great efforts would have been required to get the PC ecosystem work like a proprietary system.

    -I don't see Turbo Boost as a coincidence. It's all part of a plan. It took years to do because proper implementation takes time. Turbo to maximize performance in general, Turbo 2.0 to optimize performance in light usage load scenarios with thermally limited systems, Configurable TDP to allow smaller than ever x86 Tablets/Convertibles/Laptops without sacrificing responsiveness.
     
    #5 IntelUser2000, Apr 24, 2012
    Last edited: Apr 24, 2012
  7. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    6. And Finally, Haswell

    The work that has been going on can not happen with changes happening with Haswell. Haswell integrates the Voltage Regulator, which has traditionally been on the board. Less distance travelled and smaller components mean power usage will drop. Not only that, transitions between frequencies and power states can be faster than ever before: http://www.anandtech.com/show/1770

    That allows further power reductions both in moderate load and light load. New architecture optimized on the 22nm Tri-Gate process will allow lower in-use power with better performance. On package Lynx Point PCH with the Ultrabook parts allow smaller form factors and further lower power consumption.

    C10 power state(C7 on non-Ultrabooks) is coming for even lower power, and the whole new ultra low power state for the platform is called S0ix.

    List of changes coming for various 2013 Ultrabook parts:

    Power Optimizer framework – Software and hardware management to ensure devices use power properly. One of the important things devices report to Power Optimizer is latency, so the framework can figure out how everything should work together.

    PCI Express(both PHY Link and device power), USB 2.0 and 3.0(3.0 allows lower power), SATA, WLAN, Audio, various sensors, cameras and other devices ALL need to work together to achieve the goal. Like I said earlier, one bad device can mess up everything.

    A big one...

    Display - Panel Self Refresh: Stores active display state on low power DRAM to save power when monitor is idle.

    The whole PC ecosystem is set to benefit big from this. You'll finally see low-power(aka Atom) based Tablets getting 8-12 hour battery life, without needing a ridiculously sized battery. High performance devices like Ultrabooks will achieve 10+ hours in battery life in light load, while allowing high performance when needed(of course with a lower battery life than at light load).
     
    #6 IntelUser2000, Apr 24, 2012
    Last edited: Apr 24, 2012
  8. cbn

    cbn Lifer

    Joined:
    Mar 27, 2009
    Messages:
    12,503
    Likes Received:
    183
    Thanks for the thread IntelUser2000.

    So with the optimizations in Windows 8 and the Intel hardware, how much closer will the laptop/tablet device makers be too each other with respect to power and performance optimization?
     
  9. cbn

    cbn Lifer

    Joined:
    Mar 27, 2009
    Messages:
    12,503
    Likes Received:
    183
    I'm also wondering how much impact these changes will have on white box manufacturer strategy?
     
  10. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    Right now there's a significant difference between Tablet and PC hardware. Soon that will blur. I wouldn't be surprised if development of some of the hardware becomes largely shared.
     
  11. cbn

    cbn Lifer

    Joined:
    Mar 27, 2009
    Messages:
    12,503
    Likes Received:
    183
    Any good threads or articles you could direct us to for that information? (I haven't been following tech as much I used to......not that I ever knew much to begin with :D, but I am guessing *some* of what you mentioning is related to Thunderbolt and docking Tablets/phones to PC peripherals)

    Also I think it will be interesting to see how Intel handles cTDP at the level of "Form Factor Reference Design". I particularly like what you mentioned here:

    So a Tablet that benefits from greater cooling when it is docked to a Keyboard? That would be very interesting to learn about! I like it!
     
  12. twjr

    twjr Senior member

    Joined:
    Jul 5, 2006
    Messages:
    399
    Likes Received:
    34
    Interesting and informative thread. Developments in the mobile area are happening very rapidly.

    I think that convertible x86 Windows 8 tablets similar to the ASUS Transformer are going to be the future. Especially now given the potential Thunderbolt has to offload all the power hungry components to a dock when you are on the move and don't need the power/performance.
     
  13. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    You should check out the link for Connected Standby I gave in post #2. The Microsoft guy there gives a presentation and is a good overview of what you are wondering about.

    Yes, they were talking about concepts like cooling through the keyboard. Maybe you can use a cooling pad, or a dock with better cooling/fan.

    Note that Configurable TDP isn't there to give you better battery life. It's there for flexible system designs(and hybrids). Configurable TDP is already here with Ivy Bridge with the high end 3920XM. But Ultrabook SKUs that will take better advantage isn't here yet.
     
  14. podspi

    podspi Golden Member

    Joined:
    Jan 11, 2011
    Messages:
    1,908
    Likes Received:
    2
    :thumbsup: great thread. It is hard to tell what is more exciting, the increases we're seeing in performance, or the increases in battery life.

    I still say the holy grail is a smartphone that will be able to dock to a keyboard, mouse, LCD, etc. For most people that would be enough, once they get to ~ C2D-class speeds.
     
  15. jaydee

    jaydee Diamond Member

    Joined:
    May 6, 2000
    Messages:
    4,485
    Likes Received:
    2
    So is "Shark Bay" going to be the exclusive platform for mobile Haswell for all laptops/ultrabooks? Or is "Shark Bay" going to be the $1000+ Ultrabook platform, while the sub-$1000 laptop market will give us a higher power consuming "other" platform?
     
  16. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    Shark Bay applies to ALL mobile Haswell based platforms. There are differences of course.

    Existing differences like...
    -ULV chips being better binned at all frequency/power levels than standard ones
    -Smaller laptops usually get more power efficient components

    To newer ones like:
    -C10 C-state rather than C7 for regular laptops
    -Deeper S0ix states? Maybe regular laptops get S0i1 while Ultrabooks get S0i3. In practice difference in S0ix states may not have too much of an impact in power usage in load and applies most for when the computer is in deeper sleep modes.
    -Regular laptops have less restrictions. That means maybe opting for components with less aggressive power states. The new power management benefits from restrictions like Ultrabooks get. Right now, they use power efficient components by choice. But standards will lift the average bar.

    Remember I said in #2 that one non-compliant device can negate any benefits? I can see this exposing manufacturers that use substandard components to save costs.
     
  17. sm625

    sm625 Diamond Member

    Joined:
    May 6, 2011
    Messages:
    8,175
    Likes Received:
    134
    I dont see what the big deal is with interrupt timers. Nothing really needed to be changed, except the frequency of the interrupt timer. Simply lower the clock so that interrupts only occur a few times a second when sleeping vs a few hundred times a second. A computer doesnt need latencies into the low milliseconds when sleeping. Who cares if a phone call or a wake timer has a 300 mS lag? No one is going to notice that. They could have fixed this a long time ago.
     
  18. IntelUser2000

    IntelUser2000 Elite Member

    Joined:
    Oct 14, 2003
    Messages:
    5,505
    Likes Received:
    551
    It wasn't fixed because it wasn't seen as necessary and wasn't simple to fix either. It's also a hardware issue, and that takes long time when you are looking for everyone to cooperate.

    Good example is with Smartbooks. What is a Smartbook? The simplest definition is "ARM Netbooks".

    Despite using same CPU/GPU/OS in Smartphones there no idle battery life gain over traditional x86 based Netbooks. That's because rest of the components were built using off-the shelf PC components.
     
    #17 IntelUser2000, May 18, 2012
    Last edited: May 18, 2012
  19. cbn

    cbn Lifer

    Joined:
    Mar 27, 2009
    Messages:
    12,503
    Likes Received:
    183
    cTDP does seem like a really good idea for the versatility/flexibility in design.

    I just wonder what direction Intel will go with their cooling patents?

    Convertible laptop

    vs.

    Tablet + lap dock?

    In any event it will be interesting to see how Intel tackles the problem of Tablet/laptop battery durability when external power is connected and the cTDP is turned up?

    (For new tech readers like myself) According to this wikipedia link Lithium ion batteries do not like high temps or constantly being topped off to 100% charge