Bit Fade test fails, other tests fine.

[SteveInAustin]

I'm getting a Bit Fade error in Memtest86 version 4.2.0 on USB stick. Yet, I have no problems in all other tests. I'm not sure what's bad, but something clearly isn't right.

My setup: ASUS P9X79 Pro motherboard, Intel i7 3930k CPU, G.Skill Ripjaws Z 64GB DDR3 2400 RAM (model F3-19200CL10Q2-64GBZHD), Corsair HX850 power supply, MSI GTX 660Ti Power Edition graphics card.

I tried just about everything I can think of in the BIOS to mitigate the problem, but nothing I do changes the results one bit. I of course flashed the BIOS with the latest version. Then I reset all to defaults in the BIOS. Gave that a try. No luck. Then turned on XMP in the BIOS and backed off from the DDR3-2400 setting to DDR3-2133 instead (DDR3-2400 doesn't boot for me). No luck. Then tried boosting DRAM voltages, VCCSA voltage, VTTCPU, etc., but no luck. Also changed DRAM REF Clock Time from Auto (around 200) to 500 to give it more time around refreshes. No difference. Then tried changing the DRAM timings from 10-12-12-31-2 to 11-13-13-32-3, but no difference.

I've not tried removing the RAMs and installing individual sticks and testing them one at a time. I'll do that next. But so far, is there anything anyone thinks is going on here? Does it look like the problem is with the memory, the motherboard, or the CPU? I wouldn't want to return all three, but I'm running out of ideas.

Memtest86 is reporting this problem from the bottom of the address space to the top, all in between. I'm amazed anything works, even Memtest86 or the BIOS. How is it that it's able to keep running when my RAMs can't remember anything for more than 1 or 2 minutes? Weird!

Thanks, - Steve

[David (PassMark)]

Yes, try one stick at a time.
Can you borrow some RAM from another machine or a friend for testing in your machine?

[SteveInAustin]

Oh this is interesting.

I unplugged all my memory sticks and only left one stick plugged in. I used Memtest86 and passed every single test, including test #10 (Bit Fade test). Great. I think this memory stick is fine, and so I unplug that memory stick and plug the next one in, and I try running Memtest86 again. It passes every test, just like the first memory stick. So I move on to the 3rd memory stick.

Now, here's where it gets interesting. I ran Memtest86 on the third memory stick. It passed all tests and began to loop back around to the first test again. So about 20 seconds into that second loop, I press "c" for the configuration screen. I select "1", then "3", and then type in "10" followed by the enter key. This is the command sequence for selecting just Test 10 (Bit Fade test). And so the bit fade test starts up. I get about a couple minutes into the test, and it shows errors.

Very interesting. It didn't show any errors the first pass through, when everything was on automated tests. It only showed errors when I manually selected that particular test.

So I rebooted the machine and had Memtest86 start through the automated tests once again. And sure enough, it got through the entire test suite, including test 10 (bit fade test). Once again, I manually selected test 10 to run, and once again it reported errors.

Hmmm.

I did notice a small difference between the manual and automated runs of test 10. In the automated run, it shows "Iterations: 80". In the manual runs of test 10, I see "Iterations:240". Everything else kind of looks the same.

The errors it reports begins at address "00000100000" (1.0MB) exactly. And it looks like it fails on every address after that. But it does not report on any address prior to that. And this address is consistent. I get it with any of the sticks I test. And I get this exact start address when I use all 8 sticks at the same time.

So to me this looks suspiciously like a software problem with Memtest86 or some kind of interaction problem with my motherboard / BIOS. But I can't be sure.

Any thoughts?

[SteveInAustin]

Memtest86+ (note the plus sign) version 4.20 (latest version) does not report any problems with the Bit Fade test on my memories. I selected the test manually and waited a very long time for the results. It passed all other tests just fine also.

So I'm tentatively concluding that this is a bug in Memtest86's Bit Fade check, resulting in a false error. And I think my memories are working just fine.

I am using a USB wireless keyboard and mouse setup (Logitech MK520). From the revision history of Memtest86, it appears that in the past the USB keyboard devices were memory mapped, and so keyboard input sometimes wrote to memory, and that resulted in the corruption of the test. But it looked like that problem was fixed. Could that still be going on in my case? Remember, Memtest86 only fails the Bit Fade test when I select it manually (by using the keyboard). When it runs in its automated start-up loop, it passes the Bit Fade test without errors.

Developers: Let me know if you want me to run some diagnostic tools on this. I'd be happy to help debug it.

- Steve

[David (PassMark)]

We have reproduced similar behaviour. So I agree it looks like a bug when running just a single test. We'll get it fixed.

[SteveInAustin]

Cool, thanks! - Steve

[David (PassMark)]

Bug has been fixed.
When test 10 was run by itself it was skipping an initialisation step, which subsequently causes false errors later on.

Fix should be made public in the next release, probably V4.3.

[SteveInAustin]

Wow, that was fast! Good job! I'll watch for the update.

Thanks,
- Steve

[ctk]

It appears to have been present in later versions too, e.g. bit fade on 5.01 selected standalone fails on all my machines, including qemu which is easy to screenshot:


6.20 is fine however.

[David (PassMark)]

This was a bug fixed more than 10 years ago.

Also we don't support MemTest86+, which isn't the same software as the original MemTest86.