Repeated bit 15 errors (0x8000) at addresses ending 7FA4 – AMD Threadripper X399

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • PetrZ
    Junior Member
    • Nov 2026
    • 4

    #1

    Repeated bit 15 errors (0x8000) at addresses ending 7FA4 – AMD Threadripper X399

    MemTest86 11.7
    Hello PassMark Support,

    I am investigating an intermittent memory error on an AMD Threadripper workstation and would appreciate your technical opinion.

    Hardware configuration:
    • CPU: AMD Ryzen Threadripper 1920X
    • Motherboard: ASRock X399 Professional Gaming
    • BIOS: P3.70 (August 27, 2019)
    • Memory: 8 × 16 GB Kingston HyperX KHX2400C15/16G DDR4 (128 GB total)
    • Tested memory speed: DDR4-2133 (2134 MT/s reported)
    • Timings: 15-15-15-35
    • Voltage: 1.20 V
    • MemTest86: v11.7 Free, build 1000

    Four memory errors recorded:
    Nov 8, 2026 128 GB 7 0x40E6F7FA4
    Nov 8, 2026 128 GB 9 0x463747FA4
    Nov 12, 2026 64 GB 7 0x28B2B7FA4
    Nov 16, 2026 128 GB 9 0x431657FA4
    Every error has exactly the same characteristics:
    • XOR between expected and actual data: 0x00008000
    • Only bit 15 is affected (1 → 0).
    • Every failing physical address ends in 0x7FA4.
    • Errors have occurred in both Single CPU 0 and Parallel modes.

    Example from the latest test:
    Test: 9
    CPU: 0
    Address: 0x431657FA4
    Expected: E141B64B
    Actual: E141364B
    XOR: 00008000

    Troubleshooting already performed:

    We have tested multiple 64 GB configurations, pairs of DIMMs, and individual modules. Many configurations passed four iterations of tests 7 and 9 without errors.

    The original 64 GB configuration that previously produced an error later passed the same tests without errors.

    The latest 128 GB test ran four complete passes in Single CPU 0 mode and produced one error during pass 2.

    MemTest86 also repeatedly logs a warning about a possible BIOS multiprocessing bug, generally involving CPU #15. However, memory errors have also occurred in Single CPU 0 mode.

    Questions:
    1. Does the repeated XOR mask 0x8000 at different physical addresses suggest a specific data line, memory channel, or memory-controller issue?
    2. Is the identical address suffix 0x7FA4 significant, or could it result from the access patterns used by tests 7 and 9?
    3. Is there a known MemTest86 issue with the AMD X399/Threadripper platform that could explain this pattern?
    4. Can we identify the affected DIMM or memory channel from these physical addresses?
    5. Would you recommend further testing of the CPU memory controller, motherboard, or DIMMs?

    Our next planned test is to repeat the full 128 GB configuration at DDR4-1866.

    I can provide the complete MemTest86 logs if required.

    Thank you for your assistance.
    Attached Files
  • David (PassMark)
    Administrator
    • Jan 2003
    • 11122

    #2
    A small number of single bit errors are generally bad RAM.
    But if the error comes and goes, it is likely marginal (just on the edge of working / not working).

    The Pro version of the software can do some decoding of the addresses. See,
    https://www.memtest86.com/tech_DIMM_Decoding.html
    But we haven't done any testing of feature with your CPU model with 8 sticks. So this might not help.

    The multiprocessing warning message probably means the UEFI BIOS is old and doesn't support threading very well. This was somewhat common in older machines. This doesn't mean the errors aren't real errors however.

    Comment

    Working...