Memtestx86 error deciphering

[cap1tal1sm]

http://i.imgur.com/fj3PuIZ.jpg

That ^ is a return I got from a memtestx86 diagnostic I ran on my computer, it was crashing and failing to restart after a crash and I'm not sure if it's a video card issue or RAM issue.

https://pcpartpicker.com/forums/topic/95512-need-diagnostic-help-computer-crashingfailing-to-restart

That is a post I put on another forum asking for general help, but what I'm wondering is what those errors mean and if they're consequential. The weird thing is is that after those errors came up, I ran the memtest again a few times, and those attempts came up clean -- multiple times! So initially it gave me an error then it was coming out fine...?

[keith]

They look like real errors. However, they seem to appear in the 2nd pass and not the 1st pass which indicates that it may be intermittent.

Does the machine still fail to restart after passing the tests?

Also, you may want to try running the latest version of MemTest86 to see if you get the same results.

[wprodrig]

Let's look at the failures:

ADDR: 688C
Assuming a 64 bit bus, starting at offset 0, this means DQ bit 50 is stuck low.
(0x6888 - 0x688F) = (DQ0-DQ63) on a 64 bit bus.

If we think about how the CPU requested data, lets assume it is cache line aligned, and burst length = 8. (Maybe it was?)
So we see a failure on the first falling edge of the DQS (ADDR: 0x6888=0x688F)

So the first failure (if assumptions are correct) is first falling edge, Byte 6, DQ 50.

ADDR: 6914 - Byte 6, bit 2 stuck low. 2nd rising edge, DQ 50.
ADDR: 6934 - Byte 6, bit 2 stuck low. 4th rising edge, DQ 50.
ADDR: 67D4 - Byte 6, bit 2 stuck low. 2nd rising edge, DQ 50.
ADDR: 67DC - Byte 6, bit 2 stuck low. 2nd falling edge, DQ 50.
ADDR: 67EC - Byte 6, bit 2 stuck low. 3rd falling edge, DQ 50.
ADDR: 67F4 - Byte 6, bit 2 stuck low. 4th rising edge, DQ 50.

It appears to me this is a problem with DQ50, not in the memory array, but on the bus itself.
(As the failures you show with mulltiple addresses hit different column addresses, but more than likely within the same row)

There is no dependency on the pattern at face value, but if this is Intel, you could have the memory scrambler enabled and it would mask the data pattern. (IE - it could be bit 50 is always stuck low or always stuck high).

Assuming the data in the burst is cache line aligned, there also was no dependence on which strobe of the burst is failing. a little wierd.

Can You reinstall the DRAM? Clean the slot (especially DQ50Maybe the problem will disappear?

It would be nice if, when the system fails - the memtest86+ team could tell us if it is a read or a write failure by flushing cache and rereading multiple times.
This would lead to a much better diagnosis