The Notebook Review forums were hosted by TechTarget, who shut down them down on January 31, 2022. This static read-only archive was pulled by NBR forum users between January 20 and January 31, 2022, in an effort to make sure that the valuable technical information that had been posted on the forums is preserved. For current discussions, many NBR forum users moved over to NotebookTalk.net after the shutdown.
Problems? See this thread at archive.org.

    Power Consumption GTX 860m Kepler vs Maxwell

    Discussion in 'Hardware Components and Aftermarket Upgrades' started by refllect, Jul 20, 2014.

  1. refllect

    refllect Notebook Consultant

    Reputations:
    155
    Messages:
    121
    Likes Received:
    6
    Trophy Points:
    31
    How much more efficient is Maxwell? Will these differences carry over to the GTX 870mx also?
     
  2. octiceps

    octiceps Nimrod

    Reputations:
    3,147
    Messages:
    9,944
    Likes Received:
    4,194
    Trophy Points:
    431
    Almost twice.

    Yes.
     
  3. HTWingNut

    HTWingNut Potato

    Reputations:
    21,580
    Messages:
    35,370
    Likes Received:
    9,877
    Trophy Points:
    931
    860m Maxwell is about 50W TDP, Kepler is about 75W TDP, so about 50% increase.
     
  4. octiceps

    octiceps Nimrod

    Reputations:
    3,147
    Messages:
    9,944
    Likes Received:
    4,194
    Trophy Points:
    431
    But Maxwell 860M is faster.
     
  5. HTWingNut

    HTWingNut Potato

    Reputations:
    21,580
    Messages:
    35,370
    Likes Received:
    9,877
    Trophy Points:
    931
    True, by about 10%. But I still wouldn't say double TDP.
     
  6. Marksman30k

    Marksman30k Notebook Deity

    Reputations:
    2,080
    Messages:
    1,068
    Likes Received:
    180
    Trophy Points:
    81
    IPC wise, a 128 Maxwell CUDA SMM can deliver 90% the performance of a 192 Kepler CUDA SMX due to improvements to the scheduling efficiency. This means that the CUDA Maxwell unit is 40% faster at the same clock speed. Considering the TDP of Maxwell is also 50% less than Kepler. This is where the double performance per watt figure comes from.
    Another way to look at it is a 640 core Maxwell SKU is equivalent to a 896 core Kepler SKU. Also take in to consideration that Maxwell can also be clocked higher due to the lower starting TDP.
     
    octiceps likes this.
  7. refllect

    refllect Notebook Consultant

    Reputations:
    155
    Messages:
    121
    Likes Received:
    6
    Trophy Points:
    31
    But real world tests don't seem to indicate this at least from what I've read the Maxwell 860 is drawing almost as much as the Kepler 860
     
  8. octiceps

    octiceps Nimrod

    Reputations:
    3,147
    Messages:
    9,944
    Likes Received:
    4,194
    Trophy Points:
    431
    They measured power draw directly on MXM slot/mobo? o_O
     
  9. refllect

    refllect Notebook Consultant

    Reputations:
    155
    Messages:
    121
    Likes Received:
    6
    Trophy Points:
    31
  10. octiceps

    octiceps Nimrod

    Reputations:
    3,147
    Messages:
    9,944
    Likes Received:
    4,194
    Trophy Points:
    431
  11. refllect

    refllect Notebook Consultant

    Reputations:
    155
    Messages:
    121
    Likes Received:
    6
    Trophy Points:
    31
    Contrary to our expectations, both GPUs are very similar in regard to the power consumption. 88-110 Watts of the Schenker W504 in the first scene of 3DMark06 is barely more than the Schenker XMG A504 with the identical Core i7-4700MQ CPU (83-107 Watts). Both devices consume up to 150 Watts under maximum load.

    We were even able to determine slightly lower values of the Schenker W504 during idle (13-23 vs. 16-26 Watts). One reason for that should be the additional hard drive of the XMG A504. We did not compare the temperatures and the system noise due to the huge differences of the two barebones.

    They have identical CPUs and I don't think screens and HDDs will create a huge difference in power draw. You can infer that power draw from both GPUs is similar.
     
  12. Marksman30k

    Marksman30k Notebook Deity

    Reputations:
    2,080
    Messages:
    1,068
    Likes Received:
    180
    Trophy Points:
    81
    I'm going by the figures on Anandtech for the desktop parts. I guess for the notebook space, they didn't really down clock Maxwell and instead to use more of the TDP advatange to improve performance, thus similar power consumption.
     
  13. nipsen

    nipsen Notebook Ditty

    Reputations:
    694
    Messages:
    1,686
    Likes Received:
    131
    Trophy Points:
    81
    ...Anandtech got something right? Weirdest thing I've heard of in a while.

    But it's not as linear as people make it out to be. The major..if you can call it that.. differences between maxwell and kepler is that maxwell can make better use of the dynamic downclocking, and maintain better output on lower clock-speeds. I.e., the kepler functions are slightly better (giving it lower idle/low-burn power drap), and it ends up at similar performance at peak with slightly lower power-draw.

    The idea that it's half the power-draw is bonk, though. That's the normal anandtech reading of whitepapers: "Hoi, I see a number from two different contexts that tell me exactly what the advertisement says! It must be right, let's write that down in 6000 words!".