PassMark Software

Announcement

Collapse
No announcement yet.

Incompatibilities with MemTest86 and newer Apple BOOTROMs

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

  • Incompatibilities with MemTest86 and newer Apple BOOTROMs

    Hello everyone!

    I've been doing some research and experiments with MemTest86 and some 2015 13" MacBook Airs and 2015 15" MacBook Pros. Among these MacBook Airs that I had, I observed MemTest86 failing test 10 pretty spectacularly, similar to how it did for this gentleman. The MacBook Pros were failing test 13 pretty spectacularly as well, where we observed one computer reporting 130,000 errors after four passes. This computer also had some amount of errors in test 10, but we will get to that later.

    The skinny of it: We have determined that the BOOTROM update that comes with OS X 10.13 (High Sierra) and later BOOTROM updates cause false positives when running MemTest86 on the system. The specifics of the incompatibility elude us.

    The setup: We had five 2015 13" MacBook Airs and two 2015 15" MacBook Pros. We had a handful of USB drives loaded with MemTest86 v7.3, with one more stick loaded with MemTest86 v7.5, though this made little difference other than stopping testing if a laptop failed more than 10,000 tests.

    Three of the MacBook airs were failing test 10 and generating an amount of errors that were multiples of 1026. Out of four passes, the laptops that failed usually only failed one pass for a total of 1026 errors; but we had one laptop that generated 3078 errors out of four passes. I updated this last laptop through Apple's App Store hoping that a UEFI update would allow for test 10 to function properly. Running MemTest86 again on this updated machine yielded 1026 errors in the second pass of test 10. No luck.

    Two of the MacBook airs completed four passes of MemTest86 without a hitch. After the initial four passes, we left these untouched until the later part of our research.

    We had one MacBook Pro that was generating tens of thousands of errors on test 13. After four passes, the total error count was over 130,000 with most being in test 13, but the remainder being in test 10. (The exact error count eludes me at the moment, but I believe test 10 generated at least 1000 errors).

    We brought in a second MacBook Pro (same model and year as the first) to double-check if the former system was bad or if the test was not behaving properly. We had conducted MemTest86 on this laptop sometime within the past year, and it was noted to have passed MemTest86. However, when this latter MacBook Pro had arrived at test 13, it started failing spectacularly. We observed tens of thousands of errors in the test before we prematurely ended the test.

    The MacBook Airs were running OS X 10.12. The former MacBook Pro was running OS X 10.13 and the latter was running OS X 10.11

    Further research: When looking into the issue, there were several threads on this site that pointed to some errors being related to the UEFI or BIOS related. Besides the thread linked above, there were several threads where Apple systems undergoing MemTest86 were struggling with tests 10 and 13. My general experience has been that MemTest has been generally reliable for Apple notebooks until we had a handful that were simultaneously failing MemTest86 test 10 in the same manner. This prompted me to double-check the BOOTROM versions of all the computers involved thus far in this research.

    The three MacBook Airs that were failing had a newer BOOTROM version: Two were using the version that comes with OS X 10.13 while the other was the one that I had updated through App Store. The two MacBook Pros had the BOOTROM version that came with OS X 10.13. However, the two remaining MacBook Airs had a BOOTROM version that precedes the 10.13 OS X update.

    Next experiment: We ran the two non-updated MacBook Airs through twenty passes of MemTest86, with no errors generated. While this method isn't 100% proof positive of these two notebooks having good RAM, we feel that it is a pretty good indicator. Afterwards, we installed OS X 10.13 on one of these MacBook Airs, which updated the BOOTROM to a newer version.

    We ran the two through MemTest86 again. Surely enough, the updated MacBook Air started failing at test 13 while the other notebook passed all its tests. The former was running on MemTest86 v7.5 and eventually autoquit after reaching over 10,000 errors, while the latter was running on MemTest86 v7.3 and completed four passes with no errors generated.

    At this point, we feel pretty comfortable in saying that BOOTROM versions released with OS X 10.13 and afterwards are incompatible with MemTest86 versions 7.3 and 7.5.

    I hope this has been informative to anyone that has read this far.
    Last edited by fwangberg; 05-07-2018, 10:14 PM. Reason: added Tags

  • #2
    Thanks for the post. We'd already guessed there was an issue. But didn't know the exact versions of the UEFI BIOS that caused the problem.

    We don't have any good contacts in Apple, but we have reported the problem to bugreport.apple.com.

    Comment


    • #3
      Thanks David.

      I've got a question.

      I kept running into the false positives in the address range of 0x0 - 0xFFF. I've run MemTest86 for one of these previously failing computers twice; first on just the Address range of 0x1000 - {Address Max}, and the second time between the address range of 0x0 - 0x1000. The test did not report any errors or any of the false positives. To me, it seems that this method should provide sufficient coverage for a test of an Apple computer's RAM

      My question is: Might there exist any exceptional circumstances where the described method doesn't provide a thorough test of all the computer's RAM?

      Comment


      • #4
        I think there is a typo in the address ranges you gave in your post. But yes, as previously noted, if you avoid testing this 0 to 0x1000 address range you won't trigger the bug.

        And yes, it still provides reasonable coverage for the rest of the RAM.

        Comment


        • #5
          Cool. Thanks again, David.

          Comment

          Working...
          X