ATTENTION! Potential Flaw with ALL XPS 15 9560 - GPU Throttling!

Discussion in 'Dell XPS and Studio XPS' started by Pete Light, Mar 17, 2017.

  1. Pete Light

    Pete Light Notebook Deity

    Reputations:
    155
    Messages:
    967
    Likes Received:
    581
    Trophy Points:
    106
    Hi All,

    I've been doing a lot more analysis on my XPS 15 the last few days and have been running many tests both before and after a LM Conductonaut repaste.

    Temps for the CPU (as mentioned by @iunlock ) are pretty good but I've noticed an issue in game also potentially noticed by Doug from ultrabookreview here:

    http://www.ultrabookreview.com/14875-fix-throttling-xps-15/

    The issue is when gaming on a particularly demanding game. Theres no doubt the GTX 1050 can handle BF1 at high settings on 1080 no problem at 50+ FPS consistently and for the first 5 - 10 mins of playing you will notice just that.

    HOWEVER, after a period of time (so long as you have CPU temps cool enough) I noticed that my FPS in game would suddenly drop from 50's to low 30's! After much annoyance and investigation I think I've found some evidence to suggest whats going on.

    [​IMG]

    If you look closely you can see many things for the XPS:
    • The CPU clock is rock stable on all four cores at 3.4GHz suggesting that the CPU is working well and is being sufficiently cooled and powered to maintain those clock speeds.
    • The GPU clock seems to be cycling around values from 1900 down to 1000/1200 then back to 1900 again
    I can confirm that in game when the clock speed is 1900 I get a nice stable 50+ FPS in game but when the clock speed is 1000 to 1200 (I.e. being throttled), the FPS drops to low 30's FPS.

    Conclusion? It's related to the ambient temps and NOT the GPU temperature because as you can see from the graphs, Whenever the Ambient sensor hits 98C the GPU throttles (to 1000 to 1200) and whenever it drops to 78C the GPU clock is enabled to go full speed again (1900). So therefore I conclude that this must be one of the VRMs / MOSFETs from iunlocks diagram.... But which one?

    [​IMG]

    Now begins the fun task to figure it out and to stack 6w/mK pads on it until the correct one is cooled. I already started this to begin with:

    [​IMG]

    But now I will stack each one to the metal chassis until it can be cooled (as per the advice from Douglas from ultrabookreview)

    I'll keep you all posted with the results!
     
    lewdvig and iunlock like this.
  2. Mobius 1

    Mobius 1 nͫiͤcͫeͤ

    Reputations:
    3,210
    Messages:
    8,621
    Likes Received:
    6,019
    Trophy Points:
    681
    Don't bother with those grey squares, it's not worth cooling them. Focus on the black rectangular chips (VRM/mosfet package).
     
    triturbo and pressing like this.
  3. iunlock

    iunlock 7820HK @ 5.1GHz

    Reputations:
    1,087
    Messages:
    3,477
    Likes Received:
    3,793
    Trophy Points:
    331
    Great stuff... yea those darn VRM's....

    Oh and careful not to block the air flow. If you pay attention to the PCH temps, it will run a few degrees hotter if the rear area is blocked / restricted....total catch 22 right? We're kinda screwed both ways, but.... I have an idea, a very wild one that may just work.....

    TBA...
     
    don_svetlio and pressing like this.
  4. pressing

    pressing Notebook Deity

    Reputations:
    183
    Messages:
    1,430
    Likes Received:
    612
    Trophy Points:
    131
    Most of the vrm heat comes from the mosfets. The silver chokes don't create much heat here.

    However...iUnlock's photo shows the chokes are quite warm. What I think is happening is the mosfets use the motherboard for some passive cooling, via the solider joints, with the chokes becoming perverse heat sinks (with no place to displace heat).

    That is supported by two 9550 gamers who resolved gaming throttling via low W/mk thermal pads to the chokes.

    A secondary issue is that the case bottom has little capacity as a heatsink. So a few
    weak pads seemed not to saturate the case bottom.

    For the 9560, I would ignore the CPU chokes and just try
    4 W/mk thermal pads on the 3 GPU vrm chokes.
     
    Pete Light likes this.
  5. Althernai

    Althernai Notebook Virtuoso

    Reputations:
    919
    Messages:
    2,230
    Likes Received:
    97
    Trophy Points:
    66
    I tried to repeat your experiment with the game I'm playing now (Torment: Tides of Numenera), but the behavior of my card is completely different with multiple stages. Per the GPU-Z log:
    1) It very briefly hits 1733.5 MHz (3 seconds or so). Transition temperature: 50C
    2) Also briefly, 1721.0 (10 seconds or so). Transition temperature: 54C
    3) 1708.5 for roughly 30 seconds. Transition temperature: 60C
    4) 1695.5 for a much longer period of time (22.5 minutes). Transition temperature: 70C
    5) 1683.0 for another 11 minutes. Transition temperature: 74C
    6) 1670.5 for 18 minutes. Transition temperature: 70C
    7) Then back to 1683.0 for a minute or so after which point I turned the game off.

    I'm a bit puzzled by these numbers because Nvidia says that the boost clock of the 1050 is only supposed to be 1493 MHz and I did not tweak the GPU at all beyond installing Nvidia's latest drivers, but, except for that 260 MHz offset, mine is kind of behaving the way one would expect: it starts at some maximum boost and gradually decreases to a clock for which the temperature is tolerable. Yours appears to be stuck at some maximum boost value (I don't believe it is 1900 anymore than I believe mine is around 1700; I suspect there is a different offset) which causes it keep generating heat until something else on the motherboard complains.

    Wait... could it be that by repasting your GPU, you have confused the natural downclocking process? After all, the transitions seem to be based on the temperature of the GPU itself, but your repaste will have driven that lower. Unlike undervolting, repasting doesn't actually decrease the amount of heat being generated -- it simply makes the diversion of the heat from the two main sources more efficient. It could be that this is heating up whatever is near the heat sink and fans faster than it was intended to be heated and, at the same time, the GPU does not know to when to slow down anymore so instead of the gradual decrease intended by the manufacturer, you get a sharp emergency drop initiated by something on the motherboard that is about to fry.
     
  6. Eason

    Eason Notebook Virtuoso

    Reputations:
    173
    Messages:
    2,068
    Likes Received:
    674
    Trophy Points:
    131
    @Pete Light are you getting "power" throttling as reported by GPU-Z? I believe that occurs after a long period of thermal throttling. Take a look at the GPU-Z sensors and let me know.
     
    Pete Light likes this.
  7. iunlock

    iunlock 7820HK @ 5.1GHz

    Reputations:
    1,087
    Messages:
    3,477
    Likes Received:
    3,793
    Trophy Points:
    331
    Hey guys just noticed something crazy...

    So on Overwatch I usually play on Ultra Maxed out settings (Not Epic, but second to highest) and the GPU temps before maxed out at 76C.

    Well, I decided to turn it down to Medium for giggles and it seems like with the VRM's carrying less of a load, the GPU temps were in the low 60C's?

    Can those of you with OW test this and see if it decreases your overall GPU temps on Medium settings? I'm really curious now...

    (For a game like OW it looks great on the 4K screen even at Medium)


    [​IMG]

    At medium settings there are a lot less load on the VRM's so in this particular scenario, there is obviously a link there. The temps won't just magically go down LOL so this has me pretty curious and the only thing I did different was turn down the graphics settings. Hmmm....

    Yes, I'm scratching my head too.....
     
    Pete Light likes this.
  8. Pete Light

    Pete Light Notebook Deity

    Reputations:
    155
    Messages:
    967
    Likes Received:
    581
    Trophy Points:
    106
    Great @pressing thanks I'll try all that later and take some pictures for ya too

    Sent from my SM-G935F using Tapatalk
     
  9. Pete Light

    Pete Light Notebook Deity

    Reputations:
    155
    Messages:
    967
    Likes Received:
    581
    Trophy Points:
    106
    Ding this now, results soon

    Sent from my SM-G935F using Tapatalk
     
  10. Pete Light

    Pete Light Notebook Deity

    Reputations:
    155
    Messages:
    967
    Likes Received:
    581
    Trophy Points:
    106
    Here you go, looks like GPU-Z also picks up the throttling. The two colours it uses are "Pwr" and "Util":

    [​IMG]

    Obviously you'll notice I enabled my GPU OC during that test and that seemed to have a big impact on getting that VRM/Ambient sensor to 98C triggering the capping again! Therefore I'm going to run some tests underclocking the GPU to see if I can get to stay below the cap!
     
Loading...

Share This Page