Delidding my 7700K

Discussion in 'Desktop Hardware' started by DukeCLR, Apr 26, 2017.

  1. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    2,880
    Messages:
    4,877
    Likes Received:
    3,029
    Trophy Points:
    331
    Thanks, will check that out in a minute.

    I've just run 50mins of the latest Prime95 (v28.1), and ran small FFT's for maximum heat. Temperature differences are now 'only' 10 degC between cores, whereas immediately after delid they were at 13degC differences. And wow, these are not some bad temperatures right for air cooling 6700K operating at 130W?! (only 62 degC CPU Package Temperature when operating at 130W).
    Prime95 small FFT (21 degC room).jpg
     
    Last edited: May 13, 2017
  2. Papusan

    Papusan BGABOOKS = That sucks!! STAHP! Dont buy FILTH...

    Reputations:
    6,474
    Messages:
    14,870
    Likes Received:
    19,827
    Trophy Points:
    931
    It is possible higher load will make them more similar. Hard to say. See if this is cosistent during tests later. Or re-try previous tests.
     
    Last edited: May 13, 2017
  3. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    2,880
    Messages:
    4,877
    Likes Received:
    3,029
    Trophy Points:
    331
    http://www.tomshardware.co.uk/forum/id-1800828/intel-temperature-guide.html

    That's an interesting source of information kind of related to delidding & testing temperatures - stuff in there I knew, stuff I didn't.

    Will run Prime95 v26.6 small FFT's as per guide above at link above. Previously I was running Blend test. Still thinking giving my CPU a bit of abuse within 48 hrs of assembly could help core temperatures even out while liquid metal still has a more runny consistency. Will let this test run for an hour or two....

    @Papusan , at higher temperatures the Intel temperature sensors are more accurate, so maybe that's why they're closer in my instance when testing that 130W CPU load. +/- 5degC is the specification for accuracy of any given Intel temperature sensor, which is why they expect all cores to be within 10 degC of each other (I'm just on that limit at the moment). This info was gleaned from the link at the top of this post (but I have heard others mention it too).
     
    Last edited: May 13, 2017
    jaug1337 likes this.
  4. Papusan

    Papusan BGABOOKS = That sucks!! STAHP! Dont buy FILTH...

    Reputations:
    6,474
    Messages:
    14,870
    Likes Received:
    19,827
    Trophy Points:
    931
    Yeah, I have it bookmarked from before. A nice guide.
     
  5. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    2,880
    Messages:
    4,877
    Likes Received:
    3,029
    Trophy Points:
    331
    Well I've upped my OC to 4.7Ghz (+100Mhz) (on core & cache), and testing it with Prime95 v26.6 small FFT's, thought I might as well kill 2 birds with 1 stone seeing as I'm running stress tests to try & even out core temperatures anyway. At the moment it's sitting at 104W CPU consumption and 59 degC @1.376V. My previous overclock was 4.6Ghz@1.328V. I read in that link above in my last post that most Intel CPUs scale with +50mv for every +100Mhz when the CPU is not too far overclocked. I was gonna then try 4.8Ghz at 1.426V, but my motherboard changed the voltage offset numbers in the BIOS from green to red, I guess they too think that over 1.4V is not particularly safe. At the link above 1.4V max safe voltage is mentioned too. To be honest I'll probably back down to 4.6Ghz for everyday usage, and then if I'm bored one time I might try for 4.8Ghz benches in the "Red Zone".

    (Please note for those only just tuning into the discussion, I have a 6700K, not a 7700K, but I figured delidding is delidding & overclocking is overclocking, so I did think it kind of belonged in this thread anyway, just be aware I've got a 6700K which is why I'm not hitting the 5Ghz that you guys are able to achieve!).
     
    Last edited: May 13, 2017
  6. Papusan

    Papusan BGABOOKS = That sucks!! STAHP! Dont buy FILTH...

    Reputations:
    6,474
    Messages:
    14,870
    Likes Received:
    19,827
    Trophy Points:
    931
    Maybe as well try 4.8GHz@1.4V or 1.408v
     
    Robbo99999 likes this.
  7. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    2,880
    Messages:
    4,877
    Likes Received:
    3,029
    Trophy Points:
    331
    I'll try it now. I did initially increase the voltage on my previous 4.6Ghz overclock to pass hours of OCCT, whereas it was Prime95 stable prior to that, and I've based the +50mv voltage increases today off that baseline - so there's a 'good' chance that 4.8Ghz is Prime stable at 1.4V, I'll try it now. Just did 25mins of Prime95 at 4.7Ghz, 59 degC, stable. Now onto 4.8Ghz....

    EDIT: @Papusan , 4.8Ghz (on core & cache) running on Prime95v26.6 small FFT @1.4V, 109W, 62 degC hottest core, 5mins stable so far...will let run for a while....

    EDIT 2: 15+mins stable at 4.8Ghz@1.4V and still 62degC hottest core. Will now try 4.9Ghz at the same 1.4V (expecting this to boot but to fail quickly on Prime95)........

    EDIT 3: yep, I was right, it booted but failed immediately when running Prime95 to blue screen at 4.9Ghz @1.4V. Some people on Overclockers say that 1.45V is the safe 24/7 max voltage for Skylake (& Kaby too I think), but I'm not so sure. 1.4V feels dangerous to me even! I reckon I could get 4.9Ghz with 1.45V, praps I'll try that another time, but for now 4.8Ghz@1.4V is my own personal limit, and I've now dialed it down to my previous 4.6Ghz for everyday. Ha, I bet in a few weeks or months I'll be running 4.8Ghz@1.4V! Always easy to keep setting new limits, just bit by bit, feeling safe...then BAM, your chip's dead! (@Papusan , my last conclusions for the day!)

    (One thing's for sure, without delidding I wouldn't feel safe even trying to run 1.4V, it'd be in the high 70's on Prime95 without delidding I can extrapolate from my before & after delidding temperature results)
     
    Last edited: May 13, 2017
    bloodhawk likes this.
  8. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    2,880
    Messages:
    4,877
    Likes Received:
    3,029
    Trophy Points:
    331
    DOH! 4.8Ghz wasn't truly stable as it failed in OCCT (even up to 1.45V, so I'm voltage limited here, even if temperatures never went over 63degC on hottest core), was Prime stable though. Gone down to 4.7Ghz @1.376V (I know from previous testing that next step down of voltage is unstable at this level), and currently testing an OCCT run, and so far stable for 30+ minutes with temperatures of 62 degC max & 54 degC average with average pull of 109W, 22 degC room temperature. At least it looks like delidding gave me the confidence to run an extra 100Mhz on top of my previous overclock while running cooler at the same time. I'm still pleased with the result - was it strictly necessary - no - did I enjoy doing it - yes, and I'm pleased I've extracted maximum performance out of my components! (Will update this post later just with confirmation that 4.7Ghz was hours stable with OCCT - will post screenshot!)

    EDIT: strange black screen failure @4.7Ghz@1.376V during OCCT after about an hour of testing, now increased voltage by 10mv to 1.384V & retesting....
     
    Last edited: May 14, 2017
  9. Papusan

    Papusan BGABOOKS = That sucks!! STAHP! Dont buy FILTH...

    Reputations:
    6,474
    Messages:
    14,870
    Likes Received:
    19,827
    Trophy Points:
    931
    Different types load - tests don't have same behavior. And although your processor is stable in those tests with even 24 hours testing doesn't mean you are guaranteed 110% stability.
    What's matter is that you have stability in your daily work flow. Not that you can go through 1-2, 4 or even 24 hours Stress testing with OCCT, P95 or Aida64 tests without errors. Only stock clocks can be terminated as stable enough if your work depends on fully stability!! You don't run oc'd processors if your workflow is controlling important infrastructure like nuclear weapons or similar types work.
     
    Last edited: May 14, 2017
  10. Robbo99999

    Robbo99999 Notebook Prophet

    Reputations:
    2,880
    Messages:
    4,877
    Likes Received:
    3,029
    Trophy Points:
    331
    Well, in all my overclocking/undervolting experience of CPUs & GPUs, stress testing overclocks/undervolts with demanding stress tests beyond the requirements of my general workload results in rock stable overclocks that I can rely on during all of my workflows (& game flows!). But, I think I know what you're getting at, you're saying I'm testing too thoroughly for how I normally use my PC - you might have a point. My 4.7Ghz at my increased voltage @1.384V just crashed after 1hr 45min of OCCT. I've now increased voltage by ANOTHER +10mv, but I don't have enough time to test OCCT before I go to bed (don't want to leave a super demanding stress test running completely unattended that can result in blue screen hangs), so just running a less demanding test overnight based on x264, which is more real world test apparently (from this thread: http://www.overclock.net/t/1570313/skylake-overclocking-guide-with-statistics). If it passes the overnight x264 test, then I'll probably test with OCCT tomorrow, so I can be there to check up on it once in a while. Good chance that 1.394V will be stable on OCCT as it would follow with the +50mv 'rule' associated with +100Mhz increments of overclock. And at this point I wouldn't be willing to push voltage any further, but I think I'm confident that anything under 1.4V would be fine for the health of my CPU especially as idle temperatures are in the high teens, and gaming temperatures are in the 30's & 40's, so really those are the real world load temperatures I'll be seeing @1.4V.
     
    Last edited: May 14, 2017
    Papusan likes this.
Loading...

Share This Page