G.Skill 4GB DDR2 fails in MemTest86+ test -- Do I RMA this memory??

[DRT]

I just bought some brand new 4GB (2x2GB) of DDR2 800MHz memory.

G.Skill F2-6400CL4D-4GBPK

Its a higher end enthusiast memory with CAS 4 latency and is about
double the cost of the regular more common CAS 5 memory out there.

I did my usual MemTest86+ (v1.65 and v1.70) and it reports an error
on the memory right away! :-( It is a one bit difference from the
expected value. It happens on the exact same address of memory on
the several times I've run MemTest86 on it. In other words, the
error is reproducible.

The memory is *not* being overclocked and I've run it in both CAS4
and CAS5 latency settings.

I've also done about 36 hours of Stress Prime 95 (SP 2004 utility)
and got no errors at all! Could MemTest86+ not be reading this
larger 4GB of RAM properly?

On the other hand, I believe it is possible that such a small area
of memory might not be accessible by a user-mode Windows program.
That is, the memory is not being used but is reserved by the
Windows kernel -- a complete guess on my part and I have nothing to
back up this statement.

I'm of the opinion that if any memory fails MemTest86+ that I
should RMA it for a replacement. None of my OCZ memory has failed
in MemTest86+ but they were the smaller 1GB sticks. Can anyone
comment on Newegg and RMA on memory? Do they accept the Memtest86+
as a valid reason for replacement?

Thanks for any info...


---

 

[krw]

I just bought some brand new 4GB (2x2GB) of DDR2 800MHz memory.

G.Skill F2-6400CL4D-4GBPK

Its a higher end enthusiast memory with CAS 4 latency and is about
double the cost of the regular more common CAS 5 memory out there.

I did my usual MemTest86+ (v1.65 and v1.70) and it reports an error
on the memory right away! :-( It is a one bit difference from the
expected value. It happens on the exact same address of memory on
the several times I've run MemTest86 on it. In other words, the
error is reproducible.

The memory is *not* being overclocked and I've run it in both CAS4
and CAS5 latency settings.

I've also done about 36 hours of Stress Prime 95 (SP 2004 utility)
and got no errors at all! Could MemTest86+ not be reading this
larger 4GB of RAM properly?

On the other hand, I believe it is possible that such a small area
of memory might not be accessible by a user-mode Windows program.
That is, the memory is not being used but is reserved by the
Windows kernel -- a complete guess on my part and I have nothing to
back up this statement.

One bit in one location? Wouldn't that be a tad "fine-grained" for
memory management?

I'm of the opinion that if any memory fails MemTest86+ that I
should RMA it for a replacement. None of my OCZ memory has failed
in MemTest86+ but they were the smaller 1GB sticks. Can anyone
comment on Newegg and RMA on memory? Do they accept the Memtest86+
as a valid reason for replacement?

Send it back.

 

[DRT]

One bit in one location? Wouldn't that be a tad "fine-grained" for
memory management?

If I understand you correctly, do you mean the one bit in that one
location isn't directly addresseable? Probably but doesn't matter.
My point was that a user-mode program (such as SP2004) could
never directly see this memory cell *if* the Windows kernel has
claimed it as kernel space memory -- hence I could run SP2004 for
ages and it could never report a failure -- at least on the memory
cell that Memtest86+ has reported a problem with.

Send it back.

Yeah, I'm going to send this G.Skill memory back to Newegg with a
screen shot of the Memtest86+ errors.


---

 

[krw]

If I understand you correctly, do you mean the one bit in that one
location isn't directly addresseable? Probably but doesn't matter.

No, I'm saying that the memory manager can't protect/map one bit so
that location *is* toast.

My point was that a user-mode program (such as SP2004) could
never directly see this memory cell *if* the Windows kernel has
claimed it as kernel space memory -- hence I could run SP2004 for
ages and it could never report a failure -- at least on the memory
cell that Memtest86+ has reported a problem with.

Possible, but not interesting. It's more likely that SP2004 doesn't
trip upon the pattern that Memtest86+ does, or can't address all
memory. In any case, the memory is bad.

Yeah, I'm going to send this G.Skill memory back to Newegg with a
screen shot of the Memtest86+ errors.

They shouldn't balk, as long as you're requesting a replacement.