irql not less or equal + kmode exceptions with P95 27.9? But not 28.9 + applications with 8700k oc?

Discussion in 'Overclocking & Cooling' started by markm75, Dec 20, 2018.

  1. markm75

    markm75 Gawd

    Messages:
    683
    Joined:
    Sep 12, 2008
    I have the z370 + (from SL) 8700k rated at 5.1ghz.. along with ram that is F4-3200C15D-32GTZ @1.40 and 3400mhz currently.. i had a pretty stable overclock or so i thought.. for almost a year.

    I have a negative avx offset of minus 1 and have done things pretty much by the book..

    I thought perhaps win 10 was unstable (some sleep times not coming fully awake, maybe one bsod here or there but generally stable).. so i decided to test with p95 again 27.9 version..

    Generally speaking by 5-7 hours it will BSOD with the irql or kmode errors or system service exception.. pretty sure ram related.. (fma is off in local.txt but no setting for avx)..
    I tried everything, raising volts on ram from 1.4 to 1.435.. raising vccio to 1.33 and sa to 1.18 or so.. same difference..
    If i run the same test with 28.9 with avx turned off in local.txt (and fma off as well).. i can go 14 hours no issue..

    Also.. 10 standard intel burn tests and 10 maximum, no bsod..

    If i turn the ram down to 2100mhz from 3200mhz p95 27.9 will make it 12 plus hours no issue.

    So i thought it was just a p95 thing, went into X-plane 11 .. ran the sim, at least one time.. it irql'd on me after about 5 or 6 minutes.

    I have yet to find a test that can cause a crash within 2 hours or less to make this easier to troubleshoot.

    Anyone have any thoughts as to what to do here.. short of chopping my ram speed..
    I thought maybe i should lower negative avx offset to -2.. start there, but if its ram then it has to be ram related and not vcore or avx related?

    Thanks in advance for any suggestions
     
  2. SvenBent

    SvenBent 2[H]4U

    Messages:
    3,115
    Joined:
    Sep 13, 2008
    Even thouglowering the ram speed it might nok be specifically ram thats the issue ( even thoug it most likely).
    maybe mobo having ahardtime delveir stable voltage under the load. lessening the load make it easer to keep up.

    Also what setting ar you running in prime.
    rime95 small or tiny mode find erros faster than large if its cpu related.
    or look into linpack
     
  3. markm75

    markm75 Gawd

    Messages:
    683
    Joined:
    Sep 12, 2008
    I have run with blend mode. I cane try small fft as well I might try changing avx offset to 2 from 1.first before increasing vcore

    I'm already around 83c on blend I guess I could try llc 6 rather than 5 too
     
  4. markm75

    markm75 Gawd

    Messages:
    683
    Joined:
    Sep 12, 2008
    I decided to try avx offset of negative 2.. didnt help any.. blend still failed after 3-7 hours..
    Right now i'm trying the same settings with small ffts of 8 and 8 just to see if i can fail it sooner, but at llc5..
    i'm noticing idle 1.392.. then under load its 1.344..
    I'm using prime95 27.9 actually with both the fma and avx = 0 entries (so failing before probably had nothing to do with avx)..
    I think i need to get that under load value close to the bios value which is near the max i see.. 1.395 i think in bios. Might be the issue and nothing to do with ram, though at 5ghz it worked for 12 hours and at 5.1 and 2100mhz it did too.

    I do have speedstep on during these tests.. i've been able to pass with it on before and given it was on when windows itself failed before i dont think its related.
     
  5. markm75

    markm75 Gawd

    Messages:
    683
    Joined:
    Sep 12, 2008
    Small ffts, 8, 8, 0mb memory failed in about 20 minutes with irql error.. (p27.9 with avx=0 fma=0)

    Trying to adjust LLC for now.

    Now.. with vcore 1.395 bios with LLC set to 6 it goes to 1.408 under load unlike with LLC5 it dropped to 1.34 but my temps are getting too hot, 84C on one core max volts 1.424 too

    Putting LLC6 and volts down to around 1.37 in the bios results in 1.376 idle.. roughly 1.392 vcore under load.. and the max so far on the same test.. i also lowered cpu cache to 44 from 46 on this round and.. i set vssa and vccio to 1.05 each (i saw one report where someone said they had to lower this to get stable).. probably too many changes at once but will see. *so far 83C on one core, rest at most 80C* (with case cover on)
     
    Last edited: Dec 20, 2018
  6. markm75

    markm75 Gawd

    Messages:
    683
    Joined:
    Sep 12, 2008
    Well that last round worked a treat.. 18 hours P95 27.9 with fma=0 avx=0.. bios setting 1.375 vcore.. vssa and vccio both at 1.05 and 3200mhz.. switching llc to 6 from 5 might have been key, or lowering the ram voltages (vrms may have been getting too hot).
    Max vcore 1.392 (usually the value during load now).. idle 1.376. I of course also lowered cpu cache to 44 from 46.. i will try for 46 later and maybe set cpu cache min to Auto (saw this mentioned elsewhere).
    I might also try for the adaptive voltage of 1.392 (the max) or maybe try one shot at 5.2ghz.