Dell XPS 15 9570 benchmarks + temps

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

  1. custom90gt

    custom90gt Dellerator Moderator

    Reputations:
    405
    Messages:
    2,313
    Likes Received:
    906
    Trophy Points:
    131
    Well my 9570 is due for delivery today so I figured I'd start a thread and lay out the groundwork so i can test it out today. I plan on doing this in several phases. I'll get temps and some benches totally stock, undervolt and get temps, repaste with temps, repaste+undervolt and get temps, and then finally I'll do iunlock's mod and get temps and benches.

    I am open to running benchmarks that are requested provided I have the benchmark and time to run it. I'll also try to put up a few pictures of the repaste and iunlock's mod. Also let me know if you want me to put stuff in the spoiler boxes to make this thread look less jumbled.

    Specs of my 9570:
    CPU: i7-8750H
    GPU: 1050 Ti Max-Q
    Ram: 16GB DDR4 2666MHz
    SSD: 512GB 960 Pro
    Screen: FHD
    Battery: 97WHr

    Stock Benchmarks with windows power slider at "better peformance" and dell fan settings set to "optimized":

    Time Spy - 2492
    Time Spy Extreme - 1158
    Fire Strike - 6840
    Fire Strike Extreme - 3454
    Fire Strike Ultra - 1748
    Sky Diver - 21648
    Cloud Gate - 27919
    Ice Storm - 98789
    Ice Storm Unlimited - 96910
    3DMark 11 Performance - Cannot complete at stock
    3DMark 11 Extreme - Cannot complete at stock
    Unigine Heaven Extreme - Cannot complete at stock
    RealBench - Realbench Bench Stock.PNG
    Cinebench R15 -115.2 fps, 1203 cb, 174 cb
    PCMark 10 - 5173
    PCMark 8 Home Accelerated - 4148
    PCMark 8 Work Accelerated - 5497
    PCMark 8 Home Conventional Battery - 6:38


    Thermals:
    Please note I am using an older version of Prime95 (28.7 build 1) and doing Small FFTs.
    Undervolt settings are -170mv on the cpu and -145mv on cache for all undervolt tests.


    Results.png

    I am also highlighting in red areas that I consider troublesome.

    DIMM throttling:
    So the 9570 has issues with DIMM temperatures. Once it measures above 62C it throttles the CPU to 800MHz. Putting thermal pad between the ram and the case helped me to avoid throttling but there isn't much headroom. Other users are cooling the sensor itself to avoid throttling.

    Check out @abujafar's findings:
    http://forum.notebookreview.com/thr...benchmarks-temps.817970/page-29#post-10767160
     
    Last edited: Jul 23, 2018
    Arondel, vs40, Vasudev and 5 others like this.
  2. custom90gt

    custom90gt Dellerator Moderator

    Reputations:
    405
    Messages:
    2,313
    Likes Received:
    906
    Trophy Points:
    131
    Undervolting:

    As you can see from the results above, the biggest bang for the buck temp and performance wise is an undervolt. I won't bore you with how to do it, but if you're a novice you can find plenty of guides on youtube. I like to steer people to @Eason's post on ultrabookreview.com to support a fellow NBR member:
    https://www.ultrabookreview.com/10167-laptop-undervolting-overcloking/

    I will tell you that I had good luck testing the stability of my undervolt with realbench stress test to quickly get me in the reasonable numbers. I solidified my results by running AIDA64's stress test overnight to ensure it would be stable. So far it's way more stable than when Dell shipped it to me. I can actually run 3dmark 2011 and Haven benchmarks.

    Every CPU is different in how it will respond to undervolting (otherwise intel would have the lower voltage already applied), however here is where I ended up:
    CPU core -170mv
    CPU cache -145mv


    Repasting:

    I noticed some really good temp drops/higher clocks with repasting, especially when combined with an undervolt. You can see my results above.

    I will post some pictures on the repaste, but there are a ton of guides out there so I won't go over that. In my application I used Kryonaut thermal paste for the CPU and GPU and I replaced the stock Dell VRAM thermal pads with K5 Pro. While K5 Pro is not anywhere near as good as my Fujipoly thermal pads, it's better suited for the uneven and quite frankly piss poor heatsink. It allows the memory to make contact with the heatpipes in the areas that were previously not even close to touching.

    One thing I did notice (and have in pretty much every XPS I've owned) is that the heatsink was bent slightly. Another quality issue with Dell. I took my time and very very very carefully bent it flat. I won't go over this here, but if you try it on your own be careful. I can't stress enough how easy it is to bend those heatpipes and cause real damage to it.
     
    Last edited: Jun 9, 2018
  3. custom90gt

    custom90gt Dellerator Moderator

    Reputations:
    405
    Messages:
    2,313
    Likes Received:
    906
    Trophy Points:
    131
    Iunlock's mod:
    Credit goes to @iunlock for all of his work on the 9560:
    http://forum.notebookreview.com/thr...ures-benchmarks-xps-15-9560-kaby-lake.802345/

    Pictures:
    Sadly just the one to show where I placed the heatsinks for the Mk5 variant of the iunlock mod. I have pictures of the others but since results were less than desirable I decided not to post them.

    Mk5.jpg

    I don't have a picture of the thermal pad on top of the ram/heatsinks but that's only because they are stuck to the bottom cover. I used 2mm cheap thermal pad for the ram and vrm heatsinks.

    Results:
    So this has been a total struggle. Here are my attempts and there is a chart that shows the results in a 30 minute realbench stress test when compared to the undervolt+repaste+ram padded with cheap thermal pad.

    Mk1 - all VRM components with sinks and thermal pad between the copper sheet and all VRM heatsinks as well as chokes. Ram with thermal pad to bottom cover and PCH with heatsinks and thermalpad between the sinks and bottom cover.

    Mk2 - all VRM components with sinks and thermal pad between the copper sheet and all VRM heatsinks but NOT the chokes. Ram with thermal pad to bottom cover and PCH with heatsinks and thermalpad between the sinks and bottom cover.

    Mk3 - all VRM components with sinks and thermal pad between the copper sheet and all VRM heatsinks but NOT the chokes. Ram with thermal pad to bottom cover and PCH with heatsinks and thermalpad between the sinks and bottom cover. Formed a copper bridge from the bottom of the ram to the copper shield with some very thin copper sheets - this was a just for fun.

    Mk4 - all VRM components with sinks and no thermal pad. Ram with thermal pad to bottom cover and PCH with heatsinks and thermalpad between the sinks and bottom cover.

    Mk5 - only the VRM near the first ambient sensor padded (5 total) and their heatsinks sinked to the bottom of the case. Ram with thermal pad to bottom cover and PCH with heatsinks and thermalpad between the sinks and bottom cover.

    These are the 30 minute Realbench stress test results below which show the ambient temps:

    iunlock 30 min realbench.png


    Benchmarks:
    I've decided to re-benchmark the system with the -170mv undervolt, repaste, and heatsinks on the VRM surrounding Ambient sensor #1 (same settings as the stock runs for comparison):
    Time Spy - 2600
    Time Spy Extreme - 1201
    Fire Strike - 7048
    Fire Strike Extreme - 3536
    Fire Strike Ultra - 1759
    Sky Diver - 22354
    Cloud Gate - 28702
    Ice Storm - 99246
    Ice Storm Extreme - 97088
    3DMark 11 Performance - 9667
    3DMark 11 Extreme - 3460
    Unigine Heaven Extreme - 1136
    RealBench - Realbench Mod.PNG
    Cinebench R15 - 118.9fps, 1254 cb, 171 cb
    PCMark 10 - 5249
    PCMark 8 Home Accelerated - 4189
    PCMark 8 Work Accelerated - 5474

    Encoding - I wanted to do a little bit on encoding (or is it transcoding in this case) for people that do that sort of thing. I am a total novice when it comes to this stuff so I will explain the tests I ran. I downloaded a 96fps "honey bees" 4k clip and a 1 minute Elysium 4k clip from http://4ksamples.com. I used Handbrake to convert them from 4k to 1080p mkv using the "H.265 MKV 1080p30" preset. I then looked at the logs to get the time taken. I also converted the first 10 chapters of Deadpool in 4k to 1080p using the "fast 1080p30" preset. I did this twice to show some potential differences. The first time I did it with the modded system including the undervolt. The second time I did it with all of the mods in place except for the undervolt. Sadly I didn't think to do it when it was stock or differences would have been even greater. After all that blathering here are the results:

    Modded WITHOUT undervolt (mm:ss):
    Honey Bees - 3:32
    Elysium - 1:34
    Deadpool - 43:03**

    Modded WITH undervolt (mm:ss):
    Honey Bees - 3:24
    Elysium - 1:30
    Deadpool - 35:47**

    **surprised as I am? I double checked the encoding logs and hwinfo logs and everything is correct. It's amazing what an extra 572mhz x 6 cores will do with encoding. Just imagine how long it would have taken at stock...

    Tabulated results:
    Bench results.png

    Looking at those results it may not seem worth it to mod the laptop. I will say that the increased stability was worth it, I had more BSOD's than I could count out of the box. Also with demanding applications it was very much worth it. I would recommend doing an undervolt at the very least as that had the biggest impact.

    I ran an hour of realbench and I didn't have any of the massive throttling that I had before. The Dimm temp stayed below 62C the whole time and the laptop bottom was warm but not hot.

    Over all I'm pretty satisfied with how everything worked out. It was a ton of work, but hopefully I've been able to help someone else here.

    Summary:
    If I had to rank order what I think is worthwhile here is how I would do it.

    1. Undervolt - by far the best
    2. Cheap thermal pad between the ram and case bottom - super cheap and easy to do
    3. Repaste - a little more advanced but worthwhile, especially in keeping the max temps down
    4. iunlock mk5 mod - basically just try to cool some of the VRM near that ambient sensor - only really if you want max performance, but I don't know if the juice was worth the squeeze so to speak on that one.
     
    Last edited: Jun 22, 2018
    Maleko48, iunlock, Arondel and 3 others like this.
  4. backslashfr

    backslashfr Notebook Enthusiast

    Reputations:
    0
    Messages:
    36
    Likes Received:
    6
    Trophy Points:
    16
    Awesome project ! I’ll get mine today too.
    After your test, i’ll probably do the same as you !
     
  5. custom90gt

    custom90gt Dellerator Moderator

    Reputations:
    405
    Messages:
    2,313
    Likes Received:
    906
    Trophy Points:
    131
    So just got my 9570 and reinstalled windows. Sadly prime95 throttles in about 20 seconds, lol. We've got some work to do.
     
    pressing likes this.
  6. custom90gt

    custom90gt Dellerator Moderator

    Reputations:
    405
    Messages:
    2,313
    Likes Received:
    906
    Trophy Points:
    131
    Slowly adding results. I still think it's quieter than my 9575 was, and no annoying whistle!!! CPU runs at 2700mhz with prime on all 6 cores. Not terrible but not good by any means.
     
    abujafar likes this.
  7. custom90gt

    custom90gt Dellerator Moderator

    Reputations:
    405
    Messages:
    2,313
    Likes Received:
    906
    Trophy Points:
    131
    Well I'll probably run PCMark 8 a round or two and then call it a night while I prime95 to test potential undervolts
     
  8. abujafar

    abujafar Notebook Consultant

    Reputations:
    27
    Messages:
    274
    Likes Received:
    99
    Trophy Points:
    41
    Agreed. I just got mine as well.
    Prime95 is tough but at least the CPU doesn't collapse to the 15W/800Mhz state.

    Realbeanch stress test seems to be the most brutal and the one that causes the PL throttling also in your results.
     
    custom90gt likes this.
  9. abujafar

    abujafar Notebook Consultant

    Reputations:
    27
    Messages:
    274
    Likes Received:
    99
    Trophy Points:
    41
    @custom90gt the max temp will always be 97c because the Dell Fan Management doesn't start the fans right away. It has like a 5-10sec delay.
    This is done to avoid "making noise" when only a short heavy-load happens.
     
  10. custom90gt

    custom90gt Dellerator Moderator

    Reputations:
    405
    Messages:
    2,313
    Likes Received:
    906
    Trophy Points:
    131
    Well I would be curious if a repaste will help distribute the heat faster and allow the fans to turn on before it gets up that high.
     
Loading...

Share This Page