Correctable ECC errors in Free and Pro version

[hhware]

I have MemTest86 Pro 9.4 and the latest Free version, 11.2. The Pro version reports a lot of correctable ECC errors on a particular set of hardware (no uncorrectable errors), whereas the newer Free version doesn't report anything on the same hardware, just successful completion of the test. ECC polling is enabled in both cases. Is that intended?

Sorry, could not find a clear answer in the FAQ. Thanks!

--Alex

[David (PassMark)]

No, that isn't intended.
The Free version can report ECC errors.

Is the behavior repeatable? Or might it just been a co-incidence (i.e some other external factor, like temperature, causing the errors during a short period)?

[hhware]

Yes, it is repeatable, and the errors occur throughout the test, not just in the beginning. I ran the tests again and have the same result, attached reports from both.

I have an old Supermicro server board with ECC DDR3 memory that I bought together, used. I use it for a file server with ZFS on top of encrypted block device, so it is really important to avoid memory errors. That old memory showed a bunch of correctable ECC errors and no uncorrectable errors on MemTest86 Pro. I bought a brand new replacement memory on Newegg, and that is what I am testing and inquiring about here: it also has correctable errors. If these errors are real, the new memory is not any better than the old one, and I should return it.

I tried MemTest86 11.2 Free version because the MemTest86 9.4 Pro reports locations of the errors as N/A, as you can see from the attached files. I thought maybe the new version would be better at that. Instead, the new version does not report any errors at all. I am confused.

There isn't even an ECC-related section in the HTML report of the Free version. Why is that, if you are saying that ECC errors are supposed to be reported?

Does it make any sense for me to buy version 11.2 Pro? Should I expect an output that is any different from that of the Free version?

Lastly, a general question. Could it be that ECC memory is simply produced with lower quality than non-ECC? If new non-ECC memory has errors, sooner or later it would result in OS or application crash because these errors are not corrected. In case of ECC, 1-bit errors are corrected, could it be that it encourages manufacturers to lower the quality bar?

Thank you for your help!

[David (PassMark)]

ECC reporting is in the Free Edition.

See,
https://www.memtest86.com/compare.html

We aren't aware of any quality difference between the chips used in ECC and non ECC RAM. I don't think it is in the interest of the big name manufacturers to knowing ship RAM with errors. Their reputation would be trash and support costs very high. Anything is possible with cheap no name RAM.

We'll have a look at the logs.

[keith]

Yes, it is repeatable, and the errors occur throughout the test, not just in the beginning. I ran the tests again and have the same result, attached reports from both.

Thanks for the logs. The ECC errors reported by v9.4 were false positives that were fixed in later MemTest86 versions. So we believe that v11.2 should be the correct behaviour.

Please let us know if this isn't the case (eg. Windows event logs or Linux MCE logs report actual ECC errors)