memtest86

[precip9]

My thanks to RealBlackStuff for recommending memtest+, which I used to diagnose my X61s, which has two 4G modules. I downloaded v5.01 and burned a CD.

The machine was freezing at decent intervals. Memtest86 has already been a help, but the results are puzzling:

1. Began with no symmetric multiprocessing (SMP), with just one core running.
2. On first run, it immediately blew up, reporting every address as bad.
3. I turned on the BIOS boot memory check. No good reason; I just wanted to see what would happen.
4. Memtest+ ran, but failed at test 7, block moves:
Test Pass Failing Address Good Bad Err-Bits Count CPU
7 0 0015537d0f0 - 5459.8MB ffdeffff fffeffff 00200000 1 0
7 0 0017537d0d0 - 5971.8MB ffdeffff fffeffff 00200000 2 0

5. Both addresses are even, and I noted that Memtest was running in "64 bit" mode, so it seemed to point to one of the modules. The module originally in slot 1 has a few defective gold fingers, with cuts in the gold that may go all the way down to the fiberglass, or perhaps only the nickel. By moving the module to the other socket, the spring forces and physical situation would cause some change in the way it mates, which might be helpful. So I moved it.

6. With the single finger-damaged module installed in slot 0, Memtest86 ran through without errors, even with symmetric multiprocessing, two cores, operating.

7. Inserting the undamaged module in slot 0, it, too, passed all tests with both cores running.

8. Inserting the finger-damaged module back in slot 0, and the undamaged module in slot 1, which is a swap from the original position, Memtest86 freezes on test 7. With SMP disabled, all tests pass.

9. This sounds flaky to me, but read http://forum.canardpc.com/threads/83126-5.01-freezes . Test 7 freezing when executed with SMP is common!

It is an unfortunate fact that all CPUs and chipsets, except for the simple, ultra reliability controller types, have pages and pages of errata, detailing instruction sequences that must not be executed for reasons like the occurrences discussed in the thread. The upshot is, I don't know whether to declare this X61s flaky or not. As I write this, it has been happily churning away for 2:27 without errors -- with SMP disabled.

Comments?

[ajkula66]

You own two other X61s units, correct?

How do they behave under the same tests?

[precip9]

You would have to ask that, wouldn't you? I wanted to play "ostrich".

I'll get back to you.

[precip9]

Here it is, and it's rather interesting. I've tried a second machine, and it freezes at exactly the same point, 53% in test 7,

"Testing: 4096M - 6144M 2048 of 8118M"

So it seems that memtest86 is either:

1. Executing an instruction sequence forbidden by Intel errata.
2. Or it has a genuine bug.

[ajkula66]

So it seems that memtest86 is either:

1. Executing an instruction sequence forbidden by Intel errata.
2. Or it has a genuine bug.

Let's presume for a split second that one of these holds true and that there's nothing wrong with the RAM and/or slots.

What SSD do you have in the problematic machine?

[precip9]

I'm deferring the answer. The memtest results are now identical to the other, "no problem" machine. The assumed cure is swapping the SODIMMs between the two slots. I wouldn't want a careless reader to pick up on the SSD if it's blameless.