Overclocking CPU/GPU/Memory Stability Testing Guidelines

Discussion in 'CPUs and Overclocking' started by Idontcare, Sep 29, 2011.

  1. Idontcare

    Idontcare Elite Member

    Joined:
    Oct 10, 1999
    Messages:
    21,130
    Likes Received:
    10
    Welcome to the forums.

    Prime95 will automatically detect the proper number of threads needed for your processor. Just load it and run the torture test.

    Prime95 really makes it easy.
     
  2. Costas Athan

    Costas Athan Senior member

    Joined:
    Sep 21, 2011
    Messages:
    314
    Likes Received:
    0
    As many as the threads of the CPU probably.
     
  3. Costas Athan

    Costas Athan Senior member

    Joined:
    Sep 21, 2011
    Messages:
    314
    Likes Received:
    0
    I would suggest running a bootable memory test instead of one that runs under Windows for checking all the available memory.
     
  4. nsafreak

    nsafreak Diamond Member

    Joined:
    Oct 16, 2001
    Messages:
    7,093
    Likes Received:
    2
    There is a deluxe edition of HCI memtest that is available for purchase for $14. Well worth it I would say. I've been having issues for the past month with random BSODs popping up every now and then (it could be days or a week or two in between BSODs) with no rhyme or reason that I could figure out. The dumpfiles indicated that there was a memory corruption issue but when I ran a bootable CD with memtest86+ for over 8 hours it couldn't find a single issue. It could've been that I tested with all RAM inserted although I would think it would still point out that I had a RAM issue, I'd just have to further narrow down the stick. Since it didn't I continued looking elsewhere for possible causes. I found out about HCI memtest and the bootable CD option and figured I'd give it a shot since my other troubleshooting didn't turn up anything.

    I tested out my RAM one stick at a time and indeed I do have one bad stick of RAM. It's been removed while I go through the RMA process with Kingston but my system has been stable since. I would recommend getting this memory testing tool if memtest86+ finds nothing but your system is still having issues related to memory.
     
  5. Idontcare

    Idontcare Elite Member

    Joined:
    Oct 10, 1999
    Messages:
    21,130
    Likes Received:
    10
    Yep, this is my experience as well, which is why the sticky recommends people use HCI Memtest over memtest86+.

    I have found multiple cases where memtest86+ failed to detect bad memory, but in every case HCI correctly detected the problem (even when running in windows).

    But I have yet to encounter the case, or read of one being reported, in which memtest86+ detects an error while HCI memtest fails to detect it.
     
  6. Durp

    Durp Member

    Joined:
    Jan 29, 2013
    Messages:
    132
    Likes Received:
    0
    Now that memtest86 is being updated again (passmark) do you think it would be the better choice over HCI?

    HCI has been much better than the older versions of memtest and current memtest86+ in my experience but a better bootable test would be nice. (I don't consider the deluxe version of HCI a fair trade at $14)
     
  7. sub-80

    sub-80 Senior member

    Joined:
    Jan 11, 2014
    Messages:
    227
    Likes Received:
    0
    I am a noob to overclocking. I read in this blog that to do a proper stability test to see if both your cpu and gou are stable under load is to run IBT and Furmark toghether.

    "
    Run FurMark and IntelBurnTest at the same time for a few hours – at least as long as you expect to operate with combined CPU+GPU load. For example, if your typical gaming session is 3 hours, you should run IBT+FurMark for at least 3 hours. Of course more is better.
    One tricky part to combined CPU+GPU stress testing is that FurMark needs some free CPU cycles in order to stress the GPU properly. If you fully load the CPU with IBT, Furmark will be bottlenecked and the GPU load will be much less than 100% – on my system it was around 75%.
    The solution is to find the right settings for IntelBurnTest that will result in stressing the CPU while leaving just enough free CPU cycles to allow FurMark to get GPU load above 95%. In my case, the highest combined load was produced with IBT set to “High” and 3 threads." Link: http://blog.szynalski.com/2012/11/25/the-right-way-to-stress-test-an-overclocked-pc/


    I only believe this is true if your psu capacity is at limit of what you need and does not apply to people who have 300watts more than they need. What do you think?
     
  8. mark jorden

    mark jorden Junior Member

    Joined:
    Jan 31, 2014
    Messages:
    1
    Likes Received:
    0
    #183 mark jorden, Jan 31, 2014
    Last edited: Jan 31, 2014
  9. Matt1970

    Matt1970 Lifer

    Joined:
    Mar 19, 2007
    Messages:
    12,323
    Likes Received:
    0
    You said your name was rudra.

     
  10. wonadd

    wonadd Junior Member

    Joined:
    Mar 6, 2014
    Messages:
    5
    Likes Received:
    0
    From my limited experience with it, it would seem the program applies its settings each time windows boots, settings in the bios remained as I set them originally, not implementing the changes I made in ai suite.
     
  11. Chrome_ops

    Chrome_ops Junior Member

    Joined:
    Mar 22, 2014
    Messages:
    3
    Likes Received:
    0
    Ok what you need to do is insert the cpu motherboard into the comp and then smash it with a hammer





    This is the tech forum and threadcrapping/neffing isn't allowed.


    esquared
    Anandtech Forum Director
     
    #186 Chrome_ops, Mar 22, 2014
    Last edited by a moderator: Mar 22, 2014
  12. Chrome_ops

    Chrome_ops Junior Member

    Joined:
    Mar 22, 2014
    Messages:
    3
    Likes Received:
    0
    Das right :cool:
     
  13. Galatian

    Galatian Senior member

    Joined:
    Dec 7, 2012
    Messages:
    372
    Likes Received:
    0
    I have a little question about the OCCT GPU test:

    1. Is it only the memory overclock or the GPU Core overclock or both that is checked?

    2. I had 2 errors pop up after 2 hours running OCCT. Should I consider it an unstable overclock?
     
  14. Galatian

    Galatian Senior member

    Joined:
    Dec 7, 2012
    Messages:
    372
    Likes Received:
    0
    More questions (but I guess no one reading them :-( ):

    I'm currently overclocking my i7-4790K. I have no problem getting it stable to run Prime95 at higher clocks with more voltage, but the heat is just killing me. I have a custom water loop but at 4,7 GHz I hit 95 degrees C easily using small FFT. Are large FFT still the way to go?

    Also is Linx still to be used? At overclockers.net I read I should use different stability tests, that don't heat up the CPU as much. What would be your advise?
     
  15. Idontcare

    Idontcare Elite Member

    Joined:
    Oct 10, 1999
    Messages:
    21,130
    Likes Received:
    10
    Hi!

    Use small FFT for checking the stability of the CPU cores themselves. Their ability to correctly add 2+2 and so on.

    Use large FFT for checking the cache and memory subsystem. Their ability to store and retrieve the bits correctly without introducing errors at the voltage and clockspeed for which you have set with your OC'ing efforts.

    In short, do both but understand why you are doing them and what it means when either one fails.

    The heat may be too high for your liking but this is just a stability test, unless you run Prime95 as an application then it is highly unlikely that your CPU is going to approach those temperatures while running your everyday apps.
     
  16. Galatian

    Galatian Senior member

    Joined:
    Dec 7, 2012
    Messages:
    372
    Likes Received:
    0
    But it does throttle at 100 degrees C ;-)
    Or perhaps I can live with those few throttles in between as long as it passes the test?

    My 4790K is running at 4,7 GHz right now 2h primestable so far, hottest core was 97 degrees, but the average is around 80. This is for large FFT.

    You wrote that LinX is better for Core logic, so I use this rather then Small FTT, as the heat is not so much. I changed something in their new version of Prime95 which really heats up those Haswell processors :rolleyes:
     
  17. Idontcare

    Idontcare Elite Member

    Joined:
    Oct 10, 1999
    Messages:
    21,130
    Likes Received:
    10
    If you are throttling then the stability test is no longer testing what it needs to test (which is stability at high temperature at the targeted OC clockspeed).

    You need to dial back your OC or increase your cooling or decide to not worry about stability. (three options to choose from)

    LinX is better for checking high temperature but it doesn't check all the same math operations as Prime95, ideally you should use both programs.

    If you are pressed for time or just don't want to get into endless rounds of stability testing then you can drop the small FFT test and just do the LinX test.
     
  18. Galatian

    Galatian Senior member

    Joined:
    Dec 7, 2012
    Messages:
    372
    Likes Received:
    0

    Ugh this is with a custom water loop, so I fear unless I use some exotic form of cooling, I won't get any better temperatures. Delidding is always an option, but after I destroyed 1 1/2 Ivys like that, I wanted to keep Haswell lidded.

    I ordered some Liquid Ultra and maybe I gain a few degrees. After all 1 or 2 degrees might be all I need for a stability check.

    Thanks for your answers!
     
  19. NoobyDoo

    NoobyDoo Senior member

    Joined:
    Nov 13, 2006
    Messages:
    463
    Likes Received:
    0
  20. Osjur

    Osjur Member

    Joined:
    Sep 21, 2013
    Messages:
    92
    Likes Received:
    19
    My testing has shown me that old Prime95 26.6 (not AVX enabled) is still the best way to test final stability of Sandy Bridge based CPU's if using offset voltage. I don't know if this applies to Ivy or Haswell because frankly, I still haven't even tried one yet :oops:
    Couple of SB-E's have been more than enough to not warrant upgrade

    I have been running IBT and Prime95 (AVX) for hours on many SB and SB-E based CPU's without a problem only to get a 0x101 bsod the moment I start old Prime95.

    For some reason, old Prime95 pulls about .02v less compared to other programs and that makes it unstable. It also doesn't generate as much heat compared AVX programs so in my opinion it is better program to test core logic stability.

    My way of doing things:
    IBT : 15 passes for initial stability testing until it passes and also for heat testing.

    Old Prime95 26.6:
    Small FFT: 1h to check if I need pump up vCore offset voltage by 0.015-0.02v. This test usually fails within minutes if vCore is too low.

    Large FFT: 2-4h. This might take little longer to see if it fails, and usually its VCCSA / VTT voltage that needs upping and not vCore.

    Blend: 8h or more for final check. This can give any of vCore / VTT / VCCSA bsod codes so up the voltages according to the code you get if this one fails (0x101, 0x124, 0x0A, etc.)

    Just my 2 cents ;)
     
  21. Idontcare

    Idontcare Elite Member

    Joined:
    Oct 10, 1999
    Messages:
    21,130
    Likes Received:
    10
    Yeah, it entirely depends on which circuits (which instructions from your specific chip's ISA) are the weakest and causing an instability at the specific operating voltage, temperature, and clockspeed when you test for stability.

    If your chip's weakest link is not the AVX circuits, but in some other circuit on the chip, then naturally testing the AVX instructions is not going to highlight an instability.

    Your chip has a thousand or more instructions, each could be the weakest link at any given point in overclocking. You'll never know if you've truly found the weakest circuit and instruction unless you exhaustively test them all.

    [​IMG]

    And since no two chips are identical, the weakest link in your chip won't necessarily be the weakest link in your neighbor's chip. That is why it behooves you to really test with as many different stability test programs as possible and don't be surprised if one program works a lot better than another program for your specific chip.

    At the time when AVX was coming out, and this sticky was written, the added temperature effect (and its ability to destabilize a CPU) trumped the question of which circuit was the weakest link.

    I.e. running the chip at 80C while testing the AVX circuits would be more indicative of stability than say running the chip at 70C while checking other instructions.

    That guideline really doesn't apply anymore because Intel has changed the voltage characteristics when processing AVX instructions.

    At the end of the day all that an OC guide can really say is you must test every single instruction in your CPU's ISA for stability at your chosen clockspeed and voltage, with an emphasis added on elevating the operating temperature during the test so as to highlight the frequency of failures that you can expect for that given instruction (accelerated testing in engineering vernacular).

    Short of that, try out the handful of stability apps that do exist and hope for the best :D :p
     
  22. Osjur

    Osjur Member

    Joined:
    Sep 21, 2013
    Messages:
    92
    Likes Received:
    19
    So that is reason why non AVX based programs tend to pull little less vCore voltage and if the load is high enough, CPU might crash. I guess Intel gives a little bump to VID when CPU detects that AVX instruction set going to be used and offset voltage follows that command by giving little more than normally.

    No wonder I have always thought that AVX is only good for testing temperature stability and the hype around it being the "best" way to test is just fud.

    I guess many will just check stability with IBT, LinX, etc. and if its not stable, then increase vCore like 0.025-0.05v and recheck and be done with it if its stable.

    I personally always want to use smallest voltages that are possible without hampering my stability, so it's tedious work to check if things really are stable, because the next 0.01v decrease might make things unstable.
     
  23. Tr4nd

    Tr4nd Member

    Joined:
    Oct 27, 2014
    Messages:
    40
    Likes Received:
    0
    This question...o_O
     
    #198 Tr4nd, Nov 6, 2014
    Last edited: Nov 15, 2014
  24. exar333

    exar333 Diamond Member

    Joined:
    Feb 7, 2004
    Messages:
    8,514
    Likes Received:
    5
    Yet another awesome sticky. I always come back to this thread with every new CPU purchase I make (and then OC, naturally). :)

    Thanks IDC!
     
  25. DilipVasan

    DilipVasan Member

    Joined:
    Dec 10, 2014
    Messages:
    25
    Likes Received:
    0
    Will overclocking cause system failure or over heating ?