The consequences of repasting an XPS 15 9560 and why most people shouldn't do it

Discussion in 'Dell XPS and Studio XPS' started by Althernai, Mar 18, 2017.

Thread Status:
Not open for further replies.
  1. Althernai

    Althernai Notebook Virtuoso

    Reputations:
    919
    Messages:
    2,233
    Likes Received:
    98
    Trophy Points:
    66
    There are quite a few threads on this forum for people who modify the cooling design of the XPS 15 9650, mainly via "repasting" (i.e. changing the thermal paste of the CPU and GPU to a better one than Dell uses). In this thread, I would like to make an argument for not doing this... or at least that it is a lot harder to do right than it first appears and stopping halfway might cause problems. I'm not an expert in this so more experienced people should feel free to correct me if I'm wrong.

    Two phenomena combine to make this harder than it looks. The first is the nature of modern CPUs and GPUs: rather than operating at a fixed clock frequency when under load, they operate between a base frequency and a boosted one. The boost is intended to speed up tasks which do not take enough time to overheat the chip. If the maximum boost is sustained for a sufficiently long time, the temperature of the chip will increase which causes the chip to lower the clock speed. It will continue to lower it as higher temperatures are reached until it stabilizes at some point between the boosted frequency and the base one. Now, the crucial point here is that (as far as I can tell; again, people can correct me if I'm wrong) the input to the clock-lowering algorithm is the temperature of the chip itself, not that of everything else in the system.

    The second phenomenon has to do with what repasting actually does. It is often lumped together with undervolting, but the two are fundamentally different in that undervolting literally lowers the amount of heat produced by the chip being undervolted at a given level of performance. The power used by the chip is roughly proportional to the square of the voltage so decreasing the voltage causes it to use less power and thus produce less heat. Thus, as long as you stay above the true minimum voltage, even the simple version of undervolting is safe and efficacious. Repasting does not lower the power usage or the heat output -- it merely moves the heat away from the chip better than the stock cooling solution.

    So, what is the result of all this? The temperature of the chip itself is lower in a repasted system and thus it will operate at higher frequencies longer than it otherwise would... but the heat thereby produced has to go somewhere. In large machines with powerful cooling systems (such as the Clevo P650), the heat goes outside of the laptop and you wind up with a machine that produces more heat, but has a higher performance. Unfortunately, the cooling system of the XPS 15 isn't that powerful and the inside of the laptop also heats up.

    This is bad news because in addition to the CPU and GPU, there are also a bunch of little chips that serve various purposes such as regulating the power delivered to different parts of the system. They do not benefit from your repasting, but they also produce heat and must be cooled by the overwhelmed cooling system. The more efficient your repaste, the greater the likelihood that one of these chips will overheat. Furthermore, they don't have algorithms that gradually decreases the clock speeds of the CPU and GPU; they simply wait until they reach temperatures close to boiling point (i.e. 95-99C) at which point an emergency override kicks in to prevent them from frying and they stop doing whatever it is that they're supposed to do. Since the ones that overheat first are typically those involved in delivering power, this causes the CPU and/or GPU to instantly downclock -- most likely not just falling from the boosted frequency, but below even the base one.

    Now, if you were sufficiently dedicated to this, could you also provide cooling solutions to these tiny chips and thus end up with a system that runs hotter, but performs better? Yes, of course. You can read in the other threads about attaching thermal pads to them and which chips need more help than others... but how many people really want to do this? Unless the factory paste job of your specific machine is really bad, it's much easier to simply undervolt the CPU, accept Dell's intended performance level and enjoy your new laptop.
     
    Anthony Accioly likes this.
  2. pressing

    pressing Notebook Deity

    Reputations:
    378
    Messages:
    1,969
    Likes Received:
    1,012
    Trophy Points:
    181
    Agreed - those with very limited skills should think twice before repasting.

    One key issue is that the Dell factory thermal paste job is frequently terrible. There is pleny of photographic evidence in this forum. I can't believe even Texan engineers accounted for that so plenty of users are getting crippled performance out of the box.

    Good 9560 data does not exist but a lot of 9550 data indicates that:

    -undervolt can provide ~10*C temp decrease
    -repaste can provide ~5*C decrease

    These steps can unlock a lot of performance for nearly free. Obviously other bottlenecks will appear. And when those bottlenecks are addressed, new ones will pop up.

    * Note Kaby Lake has less potential upside to undervolting as it is essentially a factory overclocked SkyLake chip

    * And YMMV as each chip varies. Heatsink warpage varies. Factory paste and user repaste vary. etc...
     
    Pete Light likes this.
  3. GoNz0

    GoNz0 Notebook Virtuoso

    Reputations:
    259
    Messages:
    3,947
    Likes Received:
    1,378
    Trophy Points:
    231
    I thought you were going to say you broke your new XPS doing it.

    Your post does have some weight but i'm afraid with this laptop it will still overheat and kick in the hardware throttle, the VRM is the main cause. The only difference is how long it takes to do it.
     
    Pete Light and pressing like this.
  4. _sem_

    _sem_ Notebook Deity

    Reputations:
    100
    Messages:
    776
    Likes Received:
    271
    Trophy Points:
    76
    The first line of throttling is indeed like this, and kicks in very quickly.
    However, there are further lines of defence against the laptop frying itself to death, like the power limit throttling, which acts based on other temperature sensors around the laptop.
    You are right that with an improved paste job, temperatures elsewhere tend to rise higher. But this should still not be dangerous.
    The problem is that in order to produce a laptop as thin, the thermal design here is so borderline that power limit throttling tends to kick in after a few minutes of heavy load, and that power limit throttling isn't smooth. I don't think an intentional bad paste job is a solution, hehe. It is possible to undervolt, pad the VRMs, limit the CPU and iGPU in Advanced power settings etc. Question is then, does it make sense to have the i7 in such a laptop, if it overheats instantly. But some folks don't play games so much, and are happy if the i7 runs unthrottled when the GPU is idle...

    Hehe, me too ;)
     
  5. pressing

    pressing Notebook Deity

    Reputations:
    378
    Messages:
    1,969
    Likes Received:
    1,012
    Trophy Points:
    181
    I thought the same thing.

    Agreed. Particularly for demanding games.

    But for medium-duty tasks and games with lower graphics settings, the repaste and undervolt can open up better performance without throttling.

    Of course, that better performance window is limited until someone figures how to keep the mosfets (well the whole vrm) below the factory set throttling threshold of ~78*C
     
  6. Philaphlous

    Philaphlous Notebook Evangelist

    Reputations:
    32
    Messages:
    363
    Likes Received:
    134
    Trophy Points:
    56
    oh this logic is bad... sorry but it is...

    Thermal limits on alot of the components in the motherboard are rated for 120C in most cases. VRM, chokes, etc...they can handle more heat than your normal CPU or GPU because the internal components are made at a much much larger structure than the 14nm components of a chip... Your chip will fail in most cases before your VRM or motherboard components do if you're not running the VRM at a constant 110-120C which no one here at stock has seen temps even that close... I've seen 95C on mine but that's about the highest I've seen....

    You might as well just massively underclock your GPU and use toothpaste as your thermal paste if you apply this logic...and be happy with a 1GHz CPU clock...
     
  7. custom90gt

    custom90gt Doc Mod Super Moderator

    Reputations:
    7,845
    Messages:
    3,833
    Likes Received:
    4,505
    Trophy Points:
    331
    I figured I'd throw my two cents into this even though they are exactly that...

     
    Eason likes this.
  8. Pete Light

    Pete Light Notebook Deity

    Reputations:
    172
    Messages:
    1,052
    Likes Received:
    643
    Trophy Points:
    131
    Exactly right in my experience. The repaste with undervolt has actually enabled me to run unthrottled but it requires a -400/-500 under clock on the GPU to be completely stable and unhindered. Next step is to tackle the VRMs

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

    Pete Light Notebook Deity

    Reputations:
    172
    Messages:
    1,052
    Likes Received:
    643
    Trophy Points:
    131
    You sure it's 78? I have some pretty conclusive evidence to suggest it's 98C when it gimps the GPU clock

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

    Pete Light Notebook Deity

    Reputations:
    172
    Messages:
    1,052
    Likes Received:
    643
    Trophy Points:
    131
    Mine goes up to 98C before it gimps the GPU (assuming the "Ambient" sensor in hwinfo is a VRM)

    Sent from my SM-G935F using Tapatalk
     
Loading...
Similar Threads - consequences repasting 9560
  1. tiner
    Replies:
    9
    Views:
    483
Thread Status:
Not open for further replies.

Share This Page