Originally Posted by Derp
Memtest86 has been consistently updated btw, the latest release is from the end of February and it loads all CPU cores now.
Yeah, Memtest86 has been multi-threaded for a while and is the first test I run.
Originally Posted by lombardsoup
So memtest86 is still the better option when testing for defective memory?
It tests more
memory than most anything that needs an OS to run, but doesn't test as hard as some.
A very strenuous test that only tests most memory will find more intermittent errors faster than a less stressful test that tests more memory, but no test can test memory addresses that aren't made available for testing.
I had a defective stick of Crucial Ballistix DDR4 2400 (which Crucial replaced promptly and without fuss, to their credit) recently that would pass atleast five complete passes of Memtest86 7.30, but would fail in less than three minutes of stressapptest. Testing with just test #7 in Memtest86 would also prompt a failure after just a few loops.
Since this failure was intermittent, highly temperature dependent, and on the same address each time, I strongly suspect that there were a few weak cells that made it through basic QA, but which would fail to hold their charge at standard tREF and tREFI settings under more demanding circumstances. The full run of Memtest86 tests wasn't getting the memory warm enough to reliably find this error (test 7 is one of the higher load tests, but ).
If the weak cells had been in an area that stressapptest (run from a bootable Lubuntu USB stick) couldn't test frequently enough (because the OS was using them), then stressapptest may not have found it. If it was a more subtle error, memtest86 may not have found it.
All of the above is why I use multiple tests (Memtest86 first, stressapptest second, and various tests in Windows last) and try to test in worst case scenarios (if the hottest the inside of my case will get in normal use is 30C, I turn fans off until I can get the inside to 40C during memory testing).