Announcement

Collapse
No announcement yet.

MemTest86 vs. Windows-based utilities for RAM testing

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • MemTest86 vs. Windows-based utilities for RAM testing

    Hello,

    I would first like to express my gratitude for your software and hard work.

    I am running a Ryzen 9 5950X-based system with an X570 chipset 128GB of RAM at 3200Mhz (two RAM kits of 2x32GB, making it 4x32GB DIMMs).
    Very early into owning the system I started experiencing stability (BSODs, access violations) and data corruption issues.

    I used MemTest86 v9.0 to test the RAM, and the only test that would consistently fail was the Hammering Test. All other MemTest86 tests would pass without error.

    With the Hammering Test, I would first see a note stating that "RAM might be vulnerable to high frequency row hammer bit flips". Some cycles later, concrete errors would start mounting up. Oddly enough, one could see an odd address distribution there: about 50% of the errors have occurred on a single address. That address would continue to "star" in subsequent test passes, even after rebooting (which made sense, as an OS with address abstraction/randomization/virtual memory is not involved here). Additionally, these errors would all happen in the same 36MB-big memory region (around the 23.3GB mark), and all seemed to be single-bit flips, too; I could never see corruption in more than one bit at a time.

    By trial-and-error, I was able to narrow this down to one of two kits of DIMMs (since I purchased two dual-module RAM kits, I would have to RMA both the offending module and its same-kit counterpart. It therefore made no difference to me which of the two modules was the culprit). Having received the replacement kit, the Hammering Test seems to pass without error. No apparent data corruption is visible, too. In my experience the possibility of Hammering Test errors manifesting themselves as real problems is very real.

    I would now wish to test the reliability of my RAM, "once and for all". I am aware that using four dual-rank modules in a Ryzen-based system at 3200MHz is "pushing it".

    Browsing online, I see that some people suggest to rather use Windows-based memory testing software. This is a bit counterintuitive to my software engineering mind, since operating systems usually abstract memory behind virtual memory managers, ASLR and etc. Moreover, because some memory would be occupied by an OS, an exhaustive test would not be possible.

    What I would like to ask is whether there is any advantage of using Windows-based tools over MemTest86 in my scenario?

    Thank you!







  • #2
    Moreover, because some memory would be occupied by an OS, an exhaustive test would not be possible.
    Yes, this is the big problem.
    Disk cache can often consume 10 to 20GB by itself.
    Then when you try and force all this out of RAM, you end up swapping to disk and testing the disk's swap file and not the RAM. Then you have the idea to turn off the swap file to stop this happening and instead blue screen the system as some device driver or the O/S makes a request for RAM for some background task, the request fails and the O/S crashes.

    Then you are left wondering, was that BSOD due to bad RAM, or low free RAM in Windows?? It can be hard to know.

    Having said that we have our own Windows memory tester built into BurnInTest.

    The only advantage to memory testing in Windows in my opinion is that you can also load the rest of the machine at the same time (again using something like BurnInTest).

    If you load the CPU, GPU & Disks to 100%, as well as the RAM, then the PSU is under a lot more load, everything is running hot in the box and you might pickup some errors related to temperature or EMI, that you might otherwise miss. i.e. other hardware and software might interfere with the RAM, when under high load.

    Comment


    • #3
      Hi,

      Thanks for your reply!

      I have a 1200w PSU. For the components I have, that's a lot of room to spare (I don't anticipate a scenario where load could exceed 800w). I was reluctant to use Windows-based tools exactly because of the reasons you stated (the virtual memory model will result in disk swapping, etc). For that reason, I prefer to use MemTest86.

      Under that configuration, would you resort to testing in Windows or would you deem the result obtained by testing in MemTest86 conclusive?

      Thanks!

      Comment


      • #4
        If you now have no errors in Memtest86 you can have a reasonable level of confidence it is OK.
        But there are small sections of RAM that MemTest86 can't test either (memory mapped hardware, BIOS firmware code & data, etc...). Plus there was the possibilities mentioned above (temperature, EMI & power under load). So nothing is perfect.

        1200W PSU is likely massive overkill, unless you have multiple high end video cards. Which probably means the PSU is not efficient at low system load. Big PSUs can go faulty just as easily as small PSUs. Plus some of the power supply regulation functions are on the motherboard & not in the PSU itself.

        Comment


        • #5
          The idea behind purchasing that PSU was to have some room to spare. I have one GeForce RTX 3090 and I might want to put in another one in the future (for some machine learning workload).
          The motherboard itself is an ASUS Crosshair VIII Dark Hero, which should have decent VRMs.

          Thanks
          Last edited by PavelPr; Mar-17-2021, 09:43 AM.

          Comment

          Working...
          X