System Freeze with ECC Injection

[axjcs]

I'm using Memtest86 Pro 6.0.0 to try and verify the ECC functionality of my system (AMD FX-8350, Gigabyte GA-990FXA-UD7, Crucial CT2KIT102472BD160B). MemTest86 shows ECC as available and enabled, but I get strange results when testing.

When I test with ECC error injection disabled, the test doesn't seem to freeze.

When I enable ECC error injection, the test will inevitably freeze at some point (system completely unresponsive, the only way to restart is by using the physical reset button). This happens no matter what CPU mode I choose, but it's not always at the same point.

Most of the time it freezes after the second error injection, but sometimes (and in certain CPU modes) it goes on for a while and then freezes. Sometimes it reports that errors have been corrected by ECC, but other times I just get the error injection message with no message about the error being corrected. Occasionally in one of the CPU modes I get a bunch of error messages saying the CPUs cannot be started.

Also, regardless of which CPU mode I select, it always lists all the CPU cores is that normal?

No matter what happens, or what mode I'm in the Errors count never goes beyond 0.

Firstly, from this can you confirm if my system actually supports ECC functionality (it reports it, but I'm never sure if it would actually correct errors when they occurred).
Is this a UEFI bug (I have the latest BIOS from Gigabyte for this board), or a sign of faulty RAM or CPU?

Any help would be very much appreciated.

[keith]

Regarding the ECC issues, please send us the MemTest86.log file under EFI\BOOT so we can have a better idea of what's going on when detecting ECC functionality.

Also, regardless of which CPU mode I select, it always lists all the CPU cores is that normal?

If you are referring to the cores listed in "CPU:" and "State:", yes it is normal. Cores that are not in use are in the "W" (waiting) state.

No matter what happens, or what mode I'm in the Errors count never goes beyond 0.

ECC errors are not counted in the Errors count but maintained under a separate "ECC Error Count" counter; Only errors as a result of mismatch between the actual and expected values from running the tests are counted.

[axjcs]

I've emailed the logfile as requested. The only option I changed was to enable ECC error injection and it lasted 2 seconds before freezing! It reports that it corrected the first error, then froze after the second error was injected.