Memory Errors and crashes

[ranger]

(RE: My previous post "Games Crashing" on 10-28; and thanks DaveW and
kony - I've tried everything suggested to no avail, but at least I've
discovered memory errors.)

Can any methods can help pinpoint whether memory errors (indicated by
running Memtest86 and Memtest) are actually caused by the memory or by
the motherboard, or possibly even by memory timings?

My Athlon64 socket 939 Chaintech VNF4ultra MOBO sets the default timing
for Corsair TwinX1024-3200C2PT at 2.5,2,4,8 and 2T, but the manufacturer
lists the rated timing at 2.5,3,3,6 1T. Either one produces memory
errors, the latter producing more errors in less time. (What if anything
does that mean?). I can get 5 errors in the first 5 minutes.

What are some other lower timings I can try and is there much benefit to
bother? Should I just go ahead and get some new RAM (that is listed by
my MOBO maker as having tested well).

It seems odd that I have crashes in any and all benchmarks in exactly
the same places in the benchmark everytime, i.e., at the very end of
AquaMark3 where the results score screen should appear, at when near the
end of 3dmark2003 in the exact same spot - where 2 bombers collide in
mid-air, and likewise in 3dmark2005. SiSoft Sandra runs a complete
benchmark and likewise always crashes at the result screen. Is this
typical of memory errors? Always crashing in the exact same spot seems
like it would point to something specific. Nothing has ever been
overclocked in the system. The Sapphire X800GTO is a brand new model and
has been RMA'd and other than running cooler the problem persists in
exactly the same places with the replacement board. Could this be
something odd in this specific video card models BIOS or something? I've
tried the earlier Cat 5.9 drivers, and installed microsofts .net framework.

I have reinstalled XP Home SP1a and have flashed the MOBO with the
latest BIOS and the latest nforce4 ultra chipset drivers, etc., The
system runs a consistent 34c, the CPU 29c, the video card never tops 70c
and generally stays in the 45-60 range.

I have had no problems whatever doing anything else with this rig, to
include defragging, downloading huge files, playing MPEG videos, even
full screen, running all kinds of programs etc., fast and to perfection.
It's only 3d benchmarks and about 5 minutes into any 3d games I try to
run. And the system, CPU, and GPU temps are as listed above. Wouldn't
memory errors, especially when they are so quickly picked up by Memtest,
cause problems in other programs besides benchmarks?

 

[kony]

(RE: My previous post "Games Crashing" on 10-28; and thanks DaveW and
kony - I've tried everything suggested to no avail, but at least I've
discovered memory errors.)

Can any methods can help pinpoint whether memory errors (indicated by
running Memtest86 and Memtest) are actually caused by the memory or by
the motherboard, or possibly even by memory timings?

My Athlon64 socket 939 Chaintech VNF4ultra MOBO sets the default timing
for Corsair TwinX1024-3200C2PT at 2.5,2,4,8 and 2T, but the manufacturer
lists the rated timing at 2.5,3,3,6 1T. Either one produces memory
errors, the latter producing more errors in less time. (What if anything
does that mean?). I can get 5 errors in the first 5 minutes.

What are some other lower timings I can try and is there much benefit to
bother? Should I just go ahead and get some new RAM (that is listed by
my MOBO maker as having tested well).

You could just get different memory, or try more manual
timings with the present modules. If you try more with the
present modules, first update your motherboard bios as it is
common for a board bios to address memory instability. Also
search with Google groups and other misc forums as there are
occasionally some lesser mentioned memory settings that can
make a difference on particular boards, like "precharge
control", but I don't know whether your particular board has
any such issues.

Easiest would be to ignore the memory's spec and start out
trying very conservative timings, incrementally making them
more aggressive until you see the onset of instability. For
example, 2.5,4,4,9. If even that is instable, try
3,4,4,4,10. That should not be necessary, and I don't know
if you have the opportunity to return the memory to the
seller if it can't run at the spec'd timings. Regardless,
if it passes at any given timing you can then try more
aggressive numbers, like 2.5,3,4,8. If your bios also has
settings like "turbo" or some other aggressively worded
option, do not use it.

It seems odd that I have crashes in any and all benchmarks in exactly
the same places in the benchmark everytime, i.e., at the very end of
AquaMark3 where the results score screen should appear, at when near the
end of 3dmark2003 in the exact same spot - where 2 bombers collide in
mid-air, and likewise in 3dmark2005.

That seems more like a driver issue, or maybe DirectX.
Regardless, it is best to go ahead and tackle the known
problem if the memory instability shown in memtest, then go
back and see if these other issues persist.

SiSoft Sandra runs a complete
benchmark and likewise always crashes at the result screen. Is this
typical of memory errors?

No, memory errors would not generally crash it at the same
spot every time, BUT, file corruption might. You may
ultimately need to completely reinstall _everything_ after
you have the memory errors eliminated, including windows
itself. This is certainly a PITA, but ultimately if you
have file corruption, which is easy when there is memory
instability from the very beginning, it is the only way to
get the system working corectly, to be sure you don't have
remnants of the earlier memory error condition now
continuing to cause problems even after the memory has
become stable.

Always crashing in the exact same spot seems
like it would point to something specific. Nothing has ever been
overclocked in the system. The Sapphire X800GTO is a brand new model and
has been RMA'd and other than running cooler the problem persists in
exactly the same places with the replacement board. Could this be
something odd in this specific video card models BIOS or something? I've
tried the earlier Cat 5.9 drivers, and installed microsofts .net framework.

Just do as mentioned above, first get the memory stable in
memtest and go from there.

I have reinstalled XP Home SP1a and have flashed the MOBO with the
latest BIOS and the latest nforce4 ultra chipset drivers, etc., The
system runs a consistent 34c, the CPU 29c, the video card never tops 70c
and generally stays in the 45-60 range.

It seems cool enough, but AFTER you try all other things, if
it still seems to crash in high load situations like games,
then try pointing a fan at the system and take voltage
readings.

I have had no problems whatever doing anything else with this rig, to
include defragging,

Defragging a system that has memory errors can be quite
catastrophic. Any and every file written during that period
is "potentially" corrupt. If at all possible, rely on
nothing on that system, deferring instead to backups that
are offline.

downloading huge files, playing MPEG videos, even
full screen, running all kinds of programs etc., fast and to perfection.

Maybe, but a program, especially a complex one, might seem
to run great while some memory errors are occuring, if they
dont' randomly coincide to the output. Likewise playing a
video, would you notice if every 3rd frame had 1 pixel that
was a little different hue than it should've been? Hundreds
of MB per second of data may be travelling but sooner or
later it may become an issue- or may not, you may just have
specific files corrupted, but tracing those down can be
quite an undertaking, far more time consuming in some cases
than simply redoing all the OS and app installations.

It's only 3d benchmarks and about 5 minutes into any 3d games I try to
run. And the system, CPU, and GPU temps are as listed above. Wouldn't
memory errors, especially when they are so quickly picked up by Memtest,
cause problems in other programs besides benchmarks?

Not necessarily. In light windows' use, your system is
essentially idle, but orders of magnitude more memory I/O in
benchmarks. It's not necessarily the reason though, could
be file corruption. No point wondering about this yet
though, first address the obvious- getting memtest86+ to
test stable and THEN seeing which further issues remain. If
Sandra or other things still crash, then consider
reinstalling those things, but frankly I would copy off data
once you have corrected the memory instability, NOT until
you have if that is the only copy of some personal data (or
whatever), then wipe the hard drive and do a fresh install
of everything.

 

[MrGrumpy]

Since you have Corsair TwinX, which is usually v.reliable, have you
contacted Corsair Tech?
There is a lifetime warranty on the memory, and they may offer usefull
advice on configeration, if that is the problem.

 

[ranger]

Turned up the DIMM voltage to 2.8 and ran MemTest86+ for 24 hours with
no errors, so I assumed my problem was solved.

Flashed the MOBO to the newest version, made sure the DIMM voltage was
again set to 2.8 (actually running at 2.75v), reformatted, reinstalled
WinXP Home SP1a, the newnforce 4 6.70 chipset drivers (without NV
firewall and NV network) but did opt for the other stuff. Made sure I've
got dx 9.0c. Installed ATI Cat 5.10 with Control Panel (not the version
that requres .net framework). After restarting a few times and
installing some other basic stuff like FireFox and AdAware etc., I ran
MemTest86+ for 8 hours - no errors.

Then, installed AquaMark3, 3dMark2003, 3dMark2005 and fully inspected
all would be normal, but, the benchmarks crash in exactly the same
places they did before!

I have 6 days to RMA the Sapphire X800GTO video card, and since I'm out
of options, that one has to be tested - so I'm getting a different brand
card (nvidia 6600GT this time). If the same thing occurs, then I've got
to assusme it's a voltage problem in the MOBO and start shopping for a
new one. Remember, this is already the second Sapphire X800GTO card -
unless they just reapplied some thermal paste and shot the same one back
to me(?)

Any thoughts guys?

 

[kony]

Then, installed AquaMark3, 3dMark2003, 3dMark2005 and fully inspected
all would be normal, but, the benchmarks crash in exactly the same
places they did before!

I have 6 days to RMA the Sapphire X800GTO video card, and since I'm out
of options, that one has to be tested - so I'm getting a different brand
card (nvidia 6600GT this time). If the same thing occurs, then I've got
to assusme it's a voltage problem in the MOBO and start shopping for a
new one. Remember, this is already the second Sapphire X800GTO card -
unless they just reapplied some thermal paste and shot the same one back
to me(?)

Any thoughts guys?

Did you reinstall windows and DirectX?

If so, and since you had two of those cards and it always
crashes in same spot, I would suspect the video driver. You
could try another driver but going to an nVidia card also
has the obvious benefit of a different driver.

 

[ranger]

I reinstalled Windows and DX9.0c, the latter direct from M'soft.

I also tried the Cat 5.9 drivers for the video card - same problem.

When I RMA'd the video card the first time the reason I gave for wanting
a repair/replacement was for apparent over-heating, and the card was
hitting 80c+. They may very well have put some new thermal paste on the
same card and returned it. The card does run 10c cooler.

If the new card (nvidia this time) makes no difference I've got to
assume the MOBO has a voltage gremlin somewhere that manifests itself
when it hits a peak load spot or something of that sort.

The memory was tested 24 hours at the timings the MOBO set but once I
had reinstalled Windows and everything I tested it for 8 more hours at
2.5,3,3,6 1T and no errors.

Checked out the Sapphire web site forums and there are several other new
X800 GTO owners who are having similar problems as crashes and reboots,
failures to boot at all, or crashes early into games. I feel pretty
confident the matter is either going to be resolved by the video card
swap or the MOBO swap.

Thanks for your help!!