Intel Kaby Lake i7-7700K CPU De-Lid & Re-Lid Results @ [H]

Discussion in 'Intel Processors' started by FrgMstr, Jan 19, 2017.

  1. Singularity_Survivor

    Singularity_Survivor Limp Gawd

    Messages:
    212
    Joined:
    Apr 7, 2016
    Woah, settle down Francis.
     
  2. Singularity_Survivor

    Singularity_Survivor Limp Gawd

    Messages:
    212
    Joined:
    Apr 7, 2016
    Along with your TIM thread, these are genuine Gold. Well done, Sir!
     
  3. sirmonkey1985

    sirmonkey1985 [H]ard|DCer of the Month - July 2010

    Messages:
    21,450
    Joined:
    Sep 13, 2008
    might be memory related then since kyles trying to do 5Ghz with 3.6ghz on the ram. while you're running 3466 and 3000 and being able to hit 5ghz that might be the difference.

    yup i have an old asus board, first year did 1.58v no problem every year that voltage dropped by .2v until the last 3 years where it's gotten worse.. now after 8 years since it was purchased 1.58v set in the bios is actually 1.31v. i was surprised just how fast it degraded though, i figured it would of survived at 1.58v for at least a few years given the board was designed to support up to 225w and my phenom II 940 was no where near that.. i give it maybe another year before it'll no longer be able to supply the required voltage to run that cpu.
     
    Last edited: Jan 22, 2017
  4. Jim Kim

    Jim Kim 2[H]4U

    Messages:
    3,467
    Joined:
    May 24, 2012
    A single person does not a statistic make. Electronics age, ssd's wear out, capacitors leak, shit happens and VOLTAGES DRIFT.
    See sirmonkey1985s post.
    And even this example is not enough data to say one way or the other, but it's starting to look like I might be onto something.
    or on something;)
     
  5. FrgMstr

    FrgMstr Just Plain Mean Staff Member

    Messages:
    48,389
    Joined:
    May 18, 1997
    Well 5.2GHz is a no-go with any load.
     
  6. Brackle

    Brackle Old Timer

    Messages:
    7,224
    Joined:
    Jun 19, 2003
    Holy poop those results....Damn I mean you can almost make a case to say is going cheap or purposely doing that to keep people overclocking too much..

    Impressive Kyle. Love that 5.3ghz SS!
     
  7. Qrash

    Qrash Gawd

    Messages:
    968
    Joined:
    Oct 9, 2014
    Am I wrong here or has it only been possible to run at 5.0 and 5.1 after reducing the memory speed? Does this mean the issue was with the CPU's memory controller? I believe the RAM was able to run at full speed when the CPU speed was less than 5 GHz.

    Perhaps a future BIOS update or an adjustment of memory related settings can allow the memory to run faster at higher CPU speeds. Or maybe different RAM would run better.
     
    Last edited: Jan 22, 2017
  8. Nenu

    Nenu [H]ardened

    Messages:
    18,843
    Joined:
    Apr 28, 2007
    The memory controller does affect overclocks for sure.
    ie I cannot get 4.7GHz on my 6700K if my ram is at 3600MHz but it works fine with ram at 3333MHz.
     
    Singularity_Survivor likes this.
  9. EdKiefer

    EdKiefer Limp Gawd

    Messages:
    226
    Joined:
    Sep 4, 2012
    I have general question, you guys find Asus realbench a good general stability OC test app as its using real app, compared to linx, Prime95 ?
     
  10. EuphoricRage470

    EuphoricRage470 [H]ard|Gawd

    Messages:
    1,108
    Joined:
    Apr 15, 2011
    I use RealBench for stability testing myself.
     
  11. TAP

    TAP Limp Gawd

    Messages:
    236
    Joined:
    Mar 29, 2016
    How much pressure can one put on a delidded cpu?
     
  12. krotch

    krotch [H]ardness Supreme

    Messages:
    4,509
    Joined:
    Aug 12, 2004
    Yes, I know that, but I was having an issue getting the heat off the CPU. The stock Intel cooler wasn't enough, the Coolermaster Hyper something rather (I don't remember what it was), couldn't do it. I ended up going watercooling, which was what did it for me. It's possible delidding and keeping the stock cooler or the Coolermaster would have alleviated the issue for me, without resorting to going watercooling. If I was back in my stateside apt, wouldn't have been much issue, as I had A/C. So the cooling setup didn't have to be all the efficient.
     
  13. krotch

    krotch [H]ardness Supreme

    Messages:
    4,509
    Joined:
    Aug 12, 2004
    Ya, I'd like to see how the temp differences with regular coolers and stock clock. I got a Kaby Lake on the way, but it'll get watercooled, so I'm not worried about temps with it. I'm staying stock on it, but my old Ivy Bridge will be going back to air, so any temp drops would be good. It was already throttling when it was last on air. I had a room fan blowing on it, while I waited to get my watercooler.

    All the random delidding reviews I see are overclocked machines.
     
  14. CSI_PC

    CSI_PC 2[H]4U

    Messages:
    2,200
    Joined:
    Apr 3, 2016
    Out of curiosity you tried to combine it with BLK adjustments increase that work better with Skylake and Kaby Lake?
    You can do some insane combinations, but curious if it will help in this situation.
    Cheers
     
  15. Nenu

    Nenu [H]ardened

    Messages:
    18,843
    Joined:
    Apr 28, 2007
    I have barely touched Bclk as there doesnt seem to be a need.
    Memory and CPU clock settings are independent of each other on Skylake.
     
    FrgMstr likes this.
  16. FrgMstr

    FrgMstr Just Plain Mean Staff Member

    Messages:
    48,389
    Joined:
    May 18, 1997
    Woohoo. New camera, new Core i5-7600K.

    20170122_131306.jpg 20170123_153234.jpg
     
  17. EdKiefer

    EdKiefer Limp Gawd

    Messages:
    226
    Joined:
    Sep 4, 2012
    I feel a 5.0+ghz is a coming :)

    The odds should be in your favor now.
     
    Last edited: Jan 23, 2017
  18. CSI_PC

    CSI_PC 2[H]4U

    Messages:
    2,200
    Joined:
    Apr 3, 2016
    Sort of as the clock frequency is still BLCK x multiplier.
    They may be independent but the BLCK influences the clock speed, set it to 115 and see what I mean; the BLCK is now unlinked from the DMI and PCIe.
    The difference is that Skylake and and Kaby Lake actually mean you can set this to decent levels where before (unless go quite far back) it was not realistically possible.
    There was an article with actual results somewhere that did exactly this for Skylake to show there is a balance between the two and the benefits of a higher BLCK, some have used 150ish in their tests that would be impossible with more recent CPUs from Intel.

    Anyway no harm trying is there to set it to 115 and then other settings?
    Thanks.
     
    Last edited: Jan 24, 2017
  19. CSI_PC

    CSI_PC 2[H]4U

    Messages:
    2,200
    Joined:
    Apr 3, 2016
    Here is an article showing it does affect Overclocking.
    http://www.anandtech.com/show/9848/bclk-overclocking-intels-non-k-skylake-processors-coming-soon

    Although since then Intel has blocked it for anything but K models.
    Separate to this as I mentioned is that some tech journalist showed the benefits of using BLCK overclocking along with the multiplier.
    One of the top scores in a competition and mentioned in the linked article was Elmor using a BLCK of 152MHz with a I3-6300 that gave him 5.8GHz (on LN2 of course).
    Elmor's thread at HWBot: http://forum.hwbot.org/showthread.php?t=149907

    Anyway in this instance I am talking about using more marginally say trying 110 to 125 and also with adjusted multiplier that will be needed to assist breaking 5GHz and what is holding it back for some.
    Cheers
     
    Last edited: Jan 24, 2017
  20. Nenu

    Nenu [H]ardened

    Messages:
    18,843
    Joined:
    Apr 28, 2007
    Interesting but I still cant see how it will help.
    There wasnt any information that it allows for higher overclocks unless running out of multipliers.
    Bclock clocking also allows for finer granularity if you want something inbetween 100MHz steps.

    If you still think its worth a go, can you be more explicit why and I'll try. But not just yet as I am temp limited on one core atm.
    It appears my CPU die either has a bend or it bends when warm because it loses paste from one end or the other after it has got hot a few times, resulting in 1 or 2 of the 4 cores getting 15 to 20C higher temps after about a day! Pointless to push it harder like this.
    I've just bought some Gelid Extreme in a vain hope that the paste will hang around long enough (its quite thick).
     
  21. CSI_PC

    CSI_PC 2[H]4U

    Messages:
    2,200
    Joined:
    Apr 3, 2016
    The issue being experienced goes beyond just a multiplier max GHz issue, and I doubt any of us here know or can exactly quantify why the BLCK works well but not necessarily same as the multiplier when done at the extreme comparisons.
    It is application dependant if you tried to do an extreme of massive BLCK with low multiplier and then massive multiplier and low BLCK at same total frequency, some respond better than others and it does come down to relationship with memory for some of those apps.
    But again I am suggesting a balanced approach to break 5GHz and this would be optimal :)
    Look I have given you an article on the subject and also a very good OCer showing it working, if you still are not interested fine as it may or may not help anyway.

    Cheers
     
    Last edited: Jan 24, 2017
    Nenu likes this.
  22. Nenu

    Nenu [H]ardened

    Messages:
    18,843
    Joined:
    Apr 28, 2007
    I had a good go after using Gelid Extreme paste on the core/lid gap.
    Basic system: 6700K, Corsair LPX 3000 C15
    Tried Bclk at 115 then 151.5MHz (to get almost exactly 4700MHz CPU clock and 3333MHz C16 ram).
    115MHz made no difference, the following was 151.5MHz.

    Each tried separately: lower CPU voltage, lower RAM voltage, higher cache speed, higher core speed, higher ram speed.
    The only change that worked was lower vcore, changed from 1.41V to 1.4V applied in the bios with the same stability.
    (reported load Vcore at the CPU reduced from 1.456V to 1.44V)
    I cant overclock any higher, even with higher vcore. Maybe 40Mhz but I could do that before.

    Cool anyway, slightly less vcore.

    ps
    Gelid Extreme paste is much better if your core temps start varying compared to each other after delidding.
    I applied it a little thick to start with so it could settle to its own thickness all over.
    Temps are very uniform, the core that was +15C now matches its neighbor.

    Its only a few hours after applying so temps may drop a little as it squeezes out and some variance may still occur.
    But it is a stack better than MX-3 or stock white paste for temps between cores not varying which would have been well over 5C by now.
    A moment ago I checked all core temps under heavy load and they were within 1C of each other.
    I havent seen that before. Even after a fresh paste, the previous best has been 4C difference.
    Pleased.

    edit
    Now next day and temps are still within 1C under load. Would normally have around 10C difference by now on at least 1 core.
    Not tested if temps have dropped due to paste movement yet, gotta have similar room temp.
    Gelid, great stuff!
     
    Last edited: Jan 25, 2017
    CSI_PC likes this.
  23. FrgMstr

    FrgMstr Just Plain Mean Staff Member

    Messages:
    48,389
    Joined:
    May 18, 1997
    OK.
     
  24. noko

    noko [H]ardness Supreme

    Messages:
    4,320
    Joined:
    Apr 14, 2010
    Except you might be stepping in another load of BS. Recommend stepping in the one that smells better ;)
     
  25. noko

    noko [H]ardness Supreme

    Messages:
    4,320
    Joined:
    Apr 14, 2010
    Did Intel use crap TIMs to keep CPU cooler manufacturers in business? Or what?

    Knowing this I may just crack open my I7 6700K so I can use a much smaller cooler in a SFF system.
     
  26. Jim Kim

    Jim Kim 2[H]4U

    Messages:
    3,467
    Joined:
    May 24, 2012
    No, they did it to save 47ยข per cpu.
     
  27. sawk

    sawk [H]Lite

    Messages:
    91
    Joined:
    Dec 3, 2010
    Firstly I have no affiliation with the creator of this product. Nor have I used it.

    http://rockitcool.myshopify.com/

    This is very similar to the product that was reviewed.

    I believe it is based in the US and seems to be as easy to use. This was also on Kickstarter a while back.

    The only reason I am mentioning it as the product availability for the Delid-die-Mate 2 is not known and in pre-order only at the moment.
     
    noko likes this.
  28. Qrash

    Qrash Gawd

    Messages:
    968
    Joined:
    Oct 9, 2014
    I saw this product a couple of weeks ago and actually thought it was the one that was used here. I see they also sell relidding parts as well. Thanks for posting.
     
  29. sawk

    sawk [H]Lite

    Messages:
    91
    Joined:
    Dec 3, 2010
    You are welcome. I have ordered the Rockit product and the re-lid tool today.

    I have a i7-7700k, asus z270 Tuf Mk2, Samsung 960 M.2 drive, Corsair 400Q, Corsair H100i v2, as well as a Corsair HX1000 Platnimum Power supply sitting at home ready to assemble :)

    I guess its time to update my Forum Signature.
     
    noko likes this.
  30. Skyblue

    Skyblue Limp Gawd

    Messages:
    261
    Joined:
    Jan 2, 2008
    My understanding of Kyle's article is that while delidding will lead to lower temperatures, it wont affect the overclocking limits of the cpu in any meaningful way. In other words, I suppose Intel made a reasonable decision to not spend money that would be completely wasted. While temps are higher at full load, how often is a cpu for home use at full load?

    Kyle, next test: Run one cpu delidded at full speed, another nondelidded at full speed, and see how long they last. 5 or 7 years? :)
     
  31. FrgMstr

    FrgMstr Just Plain Mean Staff Member

    Messages:
    48,389
    Joined:
    May 18, 1997
    Promise I won't hose the video....too badly, this time.

    20170126_215953.jpg
     
    noko likes this.
  32. FrgMstr

    FrgMstr Just Plain Mean Staff Member

    Messages:
    48,389
    Joined:
    May 18, 1997
    Making headway...

     
    Peter2k and noko like this.
  33. bal3wolf

    bal3wolf Limp Gawd

    Messages:
    264
    Joined:
    May 14, 2008
    Temps don't normaly kill its the vcore so a delided one could run a higher clock while the nondelidded would have to run slower likely to stay under the thottle range or have better cooling.
     
  34. Peter2k

    Peter2k Limp Gawd

    Messages:
    309
    Joined:
    Apr 25, 2016
    a drop in 20 degrees is usually the thing people invest into watercooling for

    even if it means just staying silent

    my 7600k dropped from 88 degrees to 68
    under stress testing, on air

    that would translate into less noisy fans
    especially in more normal cpu load conditions

    it's already a K CPU
    I'd happily throw another 5 bucks at Intel for not using bad TIM (maybe solder again)
     
  35. Nenu

    Nenu [H]ardened

    Messages:
    18,843
    Joined:
    Apr 28, 2007
    Why did you quote me?
     
  36. Nenu

    Nenu [H]ardened

    Messages:
    18,843
    Joined:
    Apr 28, 2007
    Update:

    I noticed my idle temps were a bit higher than normal, ran Prime95 to see if load was also higher and realised its approx 10C higher than I expected, must have missed this.
    After a bit of testing I have confirmed that setting exactly the same voltages and speeds, Bclk at 151.75MHz raises my idle temps by 2C to 3C and load by 9C on each core.
    ie 1.4V vcore, 4.7GHz core, 4.5GHz cache, 3333MHz ram are the settings used.

    I verified Vcore with Asus AIsuite 3 which is one of the only apps that correctly identifies the voltage sent to the cpu.
    Vcore at idle is identical for both at 1.424V
    Vcore under load is 1.44V for 100MHz Bclk and 1.456V or 1.44V (varies) for 151.75MHz Bclk.
    That difference is equivalent to changing from 1.4V applied to 1.41V (at 100MHz Bclk) but it doesnt explain a 9C jump in load temps so something else is happening with higher bclk to make CPU temps jump.

    I decided to go back to 100MHz Bclk.
     
  37. exlink

    exlink [H]ardness Supreme

    Messages:
    4,319
    Joined:
    Dec 16, 2006
    It's really a shame that Intel decided to use such a shitty TIM. I would understand if they used this for their non-K processors, but they really should've gone a better route for the K line processors which are more geared towards the enthusiast crowd.

    Looks like I'll be replacing the TIM with liquid metal once my 7700K comes in and I get my hands on a Delid Die Mate. Thanks for all your research Kyle.
     
    FrgMstr likes this.
  38. Jayham

    Jayham Gawd

    Messages:
    931
    Joined:
    Nov 25, 2008
    Yep, this has further encouraged me to do a delidding/liquid metal on my newly acquired 7700k! Good ole razor method, won't be able reseal but 20-30c temp drop if all goes well is insane!
     
  39. Reggie Jackson

    Reggie Jackson n00b

    Messages:
    2
    Joined:
    Jan 22, 2017
    Hello everyone,

    Long time lurker, not many post.

    I figured I'd share my results as I was pretty happy with the process and end results thus far.

    Relevant System Specs-

    Asus ROG Maximus Hero IX
    16G GSkill Trident(2x8) @ 3600mhz
    7700K
    Kracken x62

    Relevant changes in bios

    Multiplier 50
    avx offset 2
    Vcore 1.35
    Clk - stock @ 100
    XMP mem timings

    Everything else set to Auto or stock (I think)

    The process I went by was to use a bench vice. Set the CPU with pin side facing up. 1 edge of the CPU with the substrate on the face of 1 side of the vice and on the other side at slight angle, the vice catches the bottom edge of the lid. As you can imagine, as you tighten the vice, it "pushes" the CPU away from the lid. Happens fast no muss no fuss. I thought it was really easy and felt no drama or trepidation in doing it. I think the real key is to make sure that both sides of the vice facing into the CPU are clean and flat with no groves to catch on the substrate that would potentially crack or damage the cpu. Cleaning the black goop off was a little harder than I would have liked. I WAS scared of damaging the substrate while scratching that stuff off. I ended up leaving the lid float as I didn't want to re-glue it down this time. I used Noctua thermal compound and suspect I'll probably try liquid metal though based on my results I may not touch it for years....

    Onto the results -
    I'm seeing max temp at load sitting around 85c on Prime95 v26.6
    In Aida64 Stress test, I'm seeing around 78c on avg though I notice when I run that stress test the clockspeed displayed in CPUZ is 4800 instead of 5ghz - I think that has to do with the AVX offset - not sure?

    In gaming (BF1) things hover around 70c. Idle is around 36c

    The CPU was NOT stable @ same settings before delid, however @ same specs temps were still under 100 at around 92 or so. It was stable @ 4.9. I am not 100% I had the same exact options though so it is possible I changed something else that helped enable 5ghz to be stable after I delided.

    Anywho, that's my story for now. Glad that HardOCP has provided so much coverage of this as project to tackle. certainly happy with my results!
     
  40. damstr

    damstr [H]ard|Gawd

    Messages:
    1,362
    Joined:
    Jun 26, 2005
    I've been going back and forth between 5GHz and 4.9GHz but settled on 4.9GHz because of 10C peak load temps. See below for more information. Both are perfectly stable

    • 5GHz
    • 1.305v in UEFI
    • 1.312-1.328v in Windows (mostly stays around 1.312v)
    • OCCT S stress test run for an hour and 5 mins. This test is similar to Prime95 in peak temps vs other stress tests
    • Peak temps across all cores from the above test 76/75/75/76
    • Ambient room temp 25C

    This is the 4.9Ghz profile I settled on. 10C lower temps and lower voltage. 100MHz isn't worth it IMO

    • 4.9GHz
    • 1.26v in UEFI
    • 1.264 in Windows
    • OCCT S stress test run for an hour and 5 mins.
    • Peak temps across all cores running the above test 65/66/66/69
    • Abmient room temp 22C
     
    DPI and FrgMstr like this.