Win 7's Ideal Core/Core Parking

Discussion in 'PC Hardware, Software and Displays' started by ECH, Jan 26, 2010.

  1. ECH

    ECH
    Regular

    Joined:
    May 24, 2007
    Messages:
    692
    Likes Received:
    30
    My understanding is rudimentary at best. From what I know Core Parking moves background processes to single core allowing other cores to sleep. This prevents all cores from always running at a low utilization which consumes power. This prevents rabid thread migration. Also, it seems that Core Parking in Win7 is more efficient then in Vista at idle state. And, Windows 7 P-state promotion policies are more aggressive by ramping up Turbo Boost faster (when needed) resulting in better performance at the cost of more power consumption then Vista. Overall, Core Park seems more ideal to i7s, i6s, i5s more so then core 2 dual/quad when power consumption is a concern. Perhaps for laptop, etc that use C1E?

    So, I wonder if disabling it offers any benefit or not? I've found no testing on this that would provide information on this yet. Also, what is Core Parking really for? Any CPU or what's mentioned above?

    If you are interested in changing core parking you can do so by finding the following key in the registry:
    -0cc5b647-c1df-4637-891a-dec35c318583

    -Find ValueMax. This value represents the % number of cores the system will park. Default is 100% IE: all Cores are potentially park-able

    -Change the value from 64 to 0. Both ValueMin and ValueMax are both zero

    -Continue to search for the 0cc5b647-c1df-4637-891a-dec35c318583 key as it will be found more then once in the registry.

    -Do a full shutdown then a cold-re-start
     
  2. Silent_Buddha

    Legend

    Joined:
    Mar 13, 2007
    Messages:
    18,406
    Likes Received:
    8,843
    From my understanding at least with regards to Core i7/i5/i3 disabling core parking would be detrimental to overall performance as well as power savings.

    One of the benefits of Core iX is that when other cores are inactive/not consuming much power and thus not contributing to the thermal envelope of the chip that it will then give a large "turbo" boost to the remaining active core(s).

    Core parking would theoretically enhance this providing greater single threaded performance by attempting to reduce the number of cores in use and thus lowering power consumption while increasing single threaded performance on Core iX.

    Regards,
    SB
     
  3. Albuquerque

    Albuquerque Red-headed step child
    Veteran

    Joined:
    Jun 17, 2004
    Messages:
    4,052
    Likes Received:
    639
    Location:
    35.1415,-90.056
    I see what you're trying to say, but turbo boost isn't defined by the hardware only -- it is also defined by the operating system (in an indirect way.)

    First, you are generally right in that lesser numbers of active cores can equate to a higher clock increase in turbo mode. However, turbo can still be enabled even when all cores of the processor are in an active state, so long as thermal and power limits aren't exceeded. Also, the number of cores that must be inactive before the "extra turbo" comes available isn't static across the Core i(x) line. I'm guessing you probably already knew this stuff, I'm more reiterating it for someone else wandering through the thread :)

    Ok, here's the OS part of turbo mode: In order for turbo mode to even be considered, the OS must have the physical processor package in the highest performance state (ACPI P0.) If Windows 7 is trying to save power, there's a very strong chance that the performance state isn't at P0. And if you're parking cores (power state C3/clockstopped or C6/powergated+clockstopped) then there's a very solid chance that you're still not in a high performance state.

    However, you can still be correct - if Windows leaves multiple cores in a C3 or C6 deep sleep state and makes a performance state change to P0, the other remaining cores can potentially use turbo mode, thus soaking up a bit more power. The true question is, what piece of the processor actually uses more power? Running one or two cores at 105-110% of their rated clockspeed, or running all four cores at 100% of their rated clockspeed? Don't forget the extra PLL's and stuff too.

    I'm thinking that a dual core at 110% is going to soak up a bit less power than a quad core at 100% clock. Just my assumption though. Of course, I'm also assuming that Windows 7 parked those extra cores because whatever you're running isn't "threaded enough" to use them anyway.
     
    #3 Albuquerque, Jan 27, 2010
    Last edited by a moderator: Jan 27, 2010
  4. Davros

    Legend

    Joined:
    Jun 7, 2004
    Messages:
    17,346
    Likes Received:
    4,770
    If 2 cores at 110% was fast enough then wouldnt 4 cores running at a reduced multi also be fast enough (aka speedstep) wouldnt that happen
     
  5. ECH

    ECH
    Regular

    Joined:
    May 24, 2007
    Messages:
    692
    Likes Received:
    30
    Does this mean that in order to properly use tubo mode via core parking you have to keep the Power Options in Win7 at Balanced (Processor Power Management sub section)? Is that what Balanced is recommended?

    Can anyone check to see if Maximum Processor State allows for anything higher then 100% for an i5, i6 or i7 CPU?
     
  6. Albuquerque

    Albuquerque Red-headed step child
    Veteran

    Joined:
    Jun 17, 2004
    Messages:
    4,052
    Likes Received:
    639
    Location:
    35.1415,-90.056
    Only if your app is threaded in a way that allowed four threads at reduced speed to be equal to 2 threads at increased speed. The entire concept behind core parking is that the OS recognizes that whatever your current applications are, they aren't multi-threaded.

    ACPI performance states are not percentages, they're states. However, this isn't entirely "intuitive", as Windows 7 exposes this setting indeed as a percentage. Rather, think of Windows 7 settings as amounts of forced processor idle time rather than a true processor performance state. It's a bit odd that they went that way, honestly.

    Anyway, P0 is the highest performance state, which you could consider as equivalent to "100%". There's nothing above that; the "turbo" mode for the Core i(x) line is not directly exposed to the OS via a power or performance state request. Rather, the OS must not be requesting any less than 100% before the CPU decides if turbo mode should be used.

    Edit: didnt' answer your first question ECH
    I can't immediately find the performance state information for Windows 7's default power profiles. However, so long as the profile allows for the "maximum processor state" at 100%, then that should mean turbo will be available for that power profile on the given battery / AC mode.
     
    #6 Albuquerque, Feb 4, 2010
    Last edited by a moderator: Feb 4, 2010
  7. ECH

    ECH
    Regular

    Joined:
    May 24, 2007
    Messages:
    692
    Likes Received:
    30
    Thanks for the info:!:
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...