Dell XPS 15 9570 benchmarks + temps

Discussion in 'Dell XPS and Studio XPS' started by custom90gt, Jun 7, 2018.

  1. micmax

    micmax Notebook Enthusiast

    Reputations:
    0
    Messages:
    40
    Likes Received:
    5
    Trophy Points:
    16
    Stock mine got 115 °C. Undervolting and padding is on your todo-list. I would repast too.
     
  2. zeverus

    zeverus Notebook Enthusiast

    Reputations:
    5
    Messages:
    48
    Likes Received:
    13
    Trophy Points:
    16
    I'v got paste, pads and heat sinks lying around, but not sure if i should return it and get one that runs better on stock, before tearing the whole machine apart....... 121 C seems extreme, so maybe defect vrm.

    How are your temps after mod?
     
  3. custom90gt

    custom90gt Doc Mod Super Moderator

    Reputations:
    7,368
    Messages:
    3,346
    Likes Received:
    3,620
    Trophy Points:
    331
    My max before messing with the VRM was 86C...

    You can see my trial results in the iunlock section.

    *on edit*
    those results were undrvolted which I'm sure took a big load off of the VRM. Looks like stock it got up to a max of 90C while running realbench...
     
    Last edited: Jul 20, 2018
  4. abujafar

    abujafar Notebook Evangelist

    Reputations:
    56
    Messages:
    357
    Likes Received:
    195
    Trophy Points:
    56
    I write here my personal insights after working on the laptop for a couple of weeks. This can be considered an addendum to the excellent @custom90gt original post

    post.jpg

    1) Undervolting is the first thing to do. I was able to push to -175 core / 140 cache -125 sysagent/igpu.
    I CAN undervolt more the cache and it appears to be stable on prime95 BUT it is not stable on throttlestop's stress test. I believe this is because with prime95 the clock goes down soon, while TS bench runs at 3.9 for most of the test.
    Even though Core and Cache seem connected at first, under some stress tests it appears that undervolting the vcore more then vcache does result in a measurable difference in VID (as measured by HWINFO64.

    2) The "catastrophic" throttling to 15W/800mhz is triggered only by the DIMM sensor reaching 62c. I don't believe this is due to the RAM getting hot. I believe it is due to the proximity of the sensor to the heatpipe. Padding this sensor to the backplate helps to get rid (or delaying a lot) this throttling during stress tests.
    After doing this, I got similar results to the ones reported by @custom90gt and also consistently so (about 118000 points in Realbench/1250 cinebench).

    3) Repasting helps HOWEVER it wasn't a straightforward job for me. I didn't have much improvement with kryonaut but I did much better with an old tube of GC Extreme that I had laying around. This can be explained by the fact that GCE is more viscous and sticky paste with respect to kryonaut. Viscosity helps in notebooks' heatsinks which sometimes are not perfectly even/balanced.

    4) Related to point 3. The contact of the heatsink with the VRAM thermal pads is uneven. In the picture, the orange pad is very squished while the green pad barely makes contact. Maybe one underrated thing done by @custom90gt is that he swapped the pads for a thick paste (k5pro). Maybe this change made its heatsink seat more properly and that is why his repaste with kryonaout worked so well.

    5) It is possible to undervolt the GPU as well. However, it's not really an undervolt, it's more a remapping of the voltage/frequency curve.
    If the temperature of the card is below 78c, the card will try to work at the 1V point (1680mhz). This is probably the "maxq" Voltage of the card.
    Once the temperature hits 78c, the card will use less voltage (and work at a lower frequency) trying to maintain 78c.
    I used MSI afterburner to change the voltage/frequency curve. To simplify, I pushed the whole curve up by 125mhz, ESPECIALLY the beginning of the curve.
    This essentially tells the card that it can work at a higher freq by using the same voltage as before, e.g. 1V/1800mhz
    Or, equivalently, it can work at a lower voltage for reaching the same frequency as before (e.g. 0.8V/1680mhz).
    I broke 1200 points in Unigine Heaven Extreme (from about 1120).

    6) The laptop does not throttle based on VRMs temps (As opposed to the 9550/9560).
    My hypothesis is that:
    - There are 3 chokes and 6 ICs supporting the GPU (In Blue). 2 ICs per choke, IC codes: 6314 GV8N1E and 6354 GV7315
    - There are 3 chokes and 3 ICs supporting the CPU (In Violet). IC code: FDMF 3035
    While the laptop doesn't throttle because of the VRM temps, they might get hot anyway.
    For example, the first ambient sensor is very close to the GPU mosfets and it can reach 100c under GPU stress tests (like Heaven).

    I am considering padding all the MOSFETs ICs with (a stack of) thermal pads up to the backplate.
    Any feedback is appreciated.

    Edit: some modifications to CPU undervolting section
     
    Last edited: Oct 10, 2018
  5. pressing

    pressing Notebook Deity

    Reputations:
    273
    Messages:
    1,804
    Likes Received:
    895
    Trophy Points:
    131
    Good post @abujafar.

    I don't have the 9570 but have spent some time analyzing the vrm of the the 9550 and 9560. The labels on the chips and board are tough to read but here is a stab:

    ==> As you noted, it looks like Dell doubled up the number of mosfets on the GPU; that reduces work, reduces heat, provides smoother power. Nice change although it was probably min required with the 1050Ti so there's that.

    Indeed the blue chips are on the GPU side and look similar to these AO mosfets
    http://www.aosmd.com/res/data_sheets/AON6314.pdf
    http://www.aosmd.com/res/data_sheets/AON6354.pdf

    ==> On the CPU side, the mosfets appear similar to these
    http://www.onsemi.com/PowerSolutions/product.do?id=FDMF3035

    ==> The PU1001 is probably the Power IC from Richtek which manages the vrm. But I can't read any of the numbers. There appears to be a similar chip (PU1501?) that might be dedicated to the CPU.

    ==> As you can see the labels on the board and chips easily, you can match the parts (9550 scheme will be similar):

    So for example, CPU power mosfets (PU1200, PU1201, PU1202) might be associated with chokes (say PQ1200, PQ1201, PQ1202). Or GPU power mosfets (PQ1002 , PQ1003 ) might be associated with choke (PL1003). I can't see the labels so well. . .

    __________________

    Using the case bottom to heatsink the mosfets has not been a very successful strategy for most people. That bottom gets saturated fast then can't dissipate much heat. Also superheats intake air so that cripples your fans/radiators.

    I always liked the idea of diverting a bit of cool air from the fans into the vrm (and also providing a small escape vent at the back for that hot air).

    Also I used some 3m 105* tape to better seal the fan-radiator connection. That definitely helps improve radiator performance but might boost ambient internal stagnant air temp a bit.
     
    _sem_ likes this.
  6. abujafar

    abujafar Notebook Evangelist

    Reputations:
    56
    Messages:
    357
    Likes Received:
    195
    Trophy Points:
    56
    post2.jpg

    I think you are 100% correct

    PU1001 and PU1501 are the same and the IC has this code stamped is "7J 5C 70C".

    To be fair there are two other chokes (yellow and pink in this image). The pink choke seems to be associated with another IC/mosfet that has code "6962 GA8H2L".

    It seems that DELL reworked the board a bit.

    Not sure how to do that, if it really works, and how much of a hassle/danger it is. I'll have another look at iunlock original post.

    Regarding my padding plans,
    - If I don't pad, the VRMs will get hot (of course here we are talking about heavy sustained load).
    - If I pad, maybe VRMs will be a bit better?
    I do agree with you that the bottom will get hot, but I assumed it was going to be a bit better.

    That is doable and makes somewhat sense. I have the tape. Not sure how much it will help.
     
  7. pressing

    pressing Notebook Deity

    Reputations:
    273
    Messages:
    1,804
    Likes Received:
    895
    Trophy Points:
    131
    Maybe the Richtek 8205 or 8206 variant. Would have to see it better to guess.

    Probably unrelated to the hot CPU & GPU power section. The code typed on the motherboard will indicate what those are. and there are only a handful of possible functions. You can measure temps manually to confirm.

    As little as possible.

    There are a few posts on that mod for the 9550 and 9560. Need to make sure airflow is not exposed to sticky side of tape to attract dust and undo any benefit. I think I wrapped both radiators 360 degrees with 3m tape but maybe not. Can't remember.

    I noticed a significant increase in hot airflow outside the back of the computer - very obvious. Temps might have dropped a few degrees but not much more.
     
    _sem_ and abujafar like this.
  8. micmax

    micmax Notebook Enthusiast

    Reputations:
    0
    Messages:
    40
    Likes Received:
    5
    Trophy Points:
    16
    115-120°C of course was at running prime and furmark simultaniously. By doing normal desktop/Office/web work Ambient1 was around 70-80°C max.
    If you do the whole modding thing you gonna get similar temps as in post 1 in this thread.
    And yes, imo to get these way better temps it's necessary to pad the vrm's and may be the chokes too to the alloy bottom plate. Otherwise much thermal energy will stay in the case.
    I surely would have kept the dell but the fan control in idle modus is so desperate poor, i couldn't stand it.
     
  9. pressing

    pressing Notebook Deity

    Reputations:
    273
    Messages:
    1,804
    Likes Received:
    895
    Trophy Points:
    131
    These temps are too high; 120C risks killing the laptop & likely reduces its lifespan.
    The temp sensors are not so accurate so individual components likely are significantly hotter.

    Also, I don't think you should be running Furmark on a uber-thin laptop; running Prime & Furmark together is crazy. For example, Prime alone or ROG RealBench alone are rough enough benchmarks for this laptop.
     
    _sem_ likes this.
  10. abujafar

    abujafar Notebook Evangelist

    Reputations:
    56
    Messages:
    357
    Likes Received:
    195
    Trophy Points:
    56
    It does look like a Richtek
    post3.jpg

    Pink is **700 and Yellow is **900. I uploaded 3 high res picture here: https://imgur.com/a/9Rxvy4F

    360 wrapping seems hard but I can definitely put some on the top.

    You mean from the screen hinge, that's where the exhaust is.
    BTW, does keeping the laptop closed while working using an external screen/keyboard harm the exhaust?
     
    pressing likes this.
Loading...

Share This Page