1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Memory RAM Instability

Discussion in 'Hardware' started by dec, 15 Apr 2019.

  1. dec

    dec [blank space]

    Joined:
    10 Jan 2009
    Posts:
    323
    Likes Received:
    12
    The work I do has forced me to mix and match RAM in my PC to temporarily increase capacity. The following is what was being used before any changes.
    i5 4690K 4200MHz VCore = 1.1V VCache = 1.1V VInput = 1.9V. All other CPU voltages left to Auto in BIOS. Core, cache, and input voltages were all set to fixed voltage.
    ASRock Z97 Extreme 4
    GSkill DDR3-2133 8GB (2x4GB) overclocked to 2400. VDRAM (overclocked) = 1.65V. Timings (overclocked) = 11-13-13-31
    This setup has been the daily usage setup since 2014. No stability issues in any form.

    I added the following kit. GSkill DDR3-1600 16GB (2x8GB). Rated for VDRAM = 1.5V. Timings = 10-12-12-30. Then I configured the combined RAM (24GB) to run at DDR3-1600 VDRAM = 1.5V

    My workload is RAM intensive as it requires reading multiple gigabytes into RAM, doing calculations (single thread) on the data in RAM, then writing the result back to disk. This was set to happen 20 times. This workload was sufficient to send me back to BIOS (by the 5th or 6th of 20 iterations) and only occurred when all 4 memory slots were populated. I expect to find instability in mis-matched RAM kits when the RAM slots fully populated (maybe I didn't get the best silicon lottery draw of IMC). For reference the workload (MATLAB2019) was run on Debian Linux 4.19 kernel.

    After doing some reading on the specifics of Devils Canyon (a little different than Haswell) in how Intel set up all the voltage names and CPU parts, I made the following adjustments in BIOS. All of these were done with the RAM at stock DDR3-1600 VDRAM = 1.5V Timings = 10-12-12-30.
    CPU core clock = 4000MHz (decreased. stock is 3500 with 3900 turbo)
    VCore = 1.1V (CPU core voltage. no change)
    VCache = 1.1V ("Last Level Cache"/Ring voltage. no change)
    VInput = 1.9V (Voltage Regulator Input voltage. no change)
    VSA = 1.05V (System Sgent voltage increased from stock)
    VIOA = 1.1V (Analog Input/Output voltage. increased from stock)
    VIOD = 1.1V (Digital Input/Output voltage. increased from stock)
    From what I read the system agent and IOD/A voltages all influence the IMC so raising those a little (100 to 200mV more than stock) should help stability under load. Despite all of these changes however, my workload would still send me back to BIOS. I couldn't reproduce the behavior in Prime95 (custom high RAM usage test) and OCCT (Linpack test) on Windows 7.

    Presently I'm using these voltages with only the 16GB kit at its stock settings and everything is fine. Has anyone had a similar experience with Haswell era parts? I was considering buying more RAM before DDR3 becomes impossible to find. I wouldn't be considering a DDR3-1600 kit however and would likely aim for a higher speed 2x8GB 2400 or 2133 kit. Could the cause of my RAM issues be the mismatching sticks? Would using more similar sticks (my 2x4GB 2133 kit along with a 2x4GB 2400 kit) lead to better results? Last but not least, is this a software issue?
     
    Last edited: 16 Apr 2019
  2. KayinBlack

    KayinBlack Unrepentant Savage

    Joined:
    2 Jul 2004
    Posts:
    5,913
    Likes Received:
    533
    I think it's both RAM mismatch and your overclock. Can you set it back to stock and still error? If so, RAM needs MEMTEST run. If not, IMC hates your RAM mismatch. While DDR3 is available I would go for another 16gb set to match the one you got. 32 GB should be that board's limit.
     
  3. dec

    dec [blank space]

    Joined:
    10 Jan 2009
    Posts:
    323
    Likes Received:
    12
    The testing was done at stock speed and still getting the problem. The 1600 16GB kit belongs to my work PC so I cannot match to that kit for this PC. The 2133 8GB kit is what i would be working with. How can Memtest tell me anything when each kit (2133 8GB and 1600 16GB) works fine alone but only when together is the problem present?
     
  4. KayinBlack

    KayinBlack Unrepentant Savage

    Joined:
    2 Jul 2004
    Posts:
    5,913
    Likes Received:
    533
    Thanks for the additional info. Sounds like an IMC problem then. Buying 2x16gb kitsch would benefit you most, but matching the 8gb kit should fix the issue.

    Memtest doesn't show errors on either kit run separately, have you tried them together? I'm sorry if you have tried it. I'm just trying to cover the bases.
     
  5. TheMadDutchDude

    TheMadDutchDude The Flying Dutchman

    Joined:
    23 Aug 2013
    Posts:
    4,739
    Likes Received:
    523
    The issue is more than likely too much voltage for the speed on the faster stuff.

    It’s never a good idea to mix RAM, and this just shows you why.
     
  6. The_Crapman

    The_Crapman World's worst stuntman. Lover of bit-tech

    Joined:
    5 Dec 2011
    Posts:
    7,653
    Likes Received:
    3,909
    Sacrifice a newt and spray it's blood on you keyboard, press the keys the blood touches, pray to the almighty timing titans for mercy and boot. It's about as good as you can do sometimes.
     
  7. KayinBlack

    KayinBlack Unrepentant Savage

    Joined:
    2 Jul 2004
    Posts:
    5,913
    Likes Received:
    533
    All I have is a lizard and an xbox controller.
     
    The_Crapman likes this.
  8. dec

    dec [blank space]

    Joined:
    10 Jan 2009
    Posts:
    323
    Likes Received:
    12
    After your suggestion here I tried "splitting the difference". One kit is rated for DDR3-1600 VDRAM = 1.5V and the other is rated for DDR3-2133 VDRAM = 1.6V so I set both to run at DDR3-1866 VDRAM = 1.55V with 11-13-13-31 timings. So far they are both happy.
     

Share This Page