How reliable is Memtest86+?

I'm trying to troubleshoot a sick Windows PC (reboots randomly or
bluescreens with DRIVER_IRQL_NOT_LESS_OR_EQUAL errors.) I may have more
questions to ask about that later, but for now, I've run Memtest86+
twice now for about 4 passes of it's default tests each time, about 2.5
hours each time. No errors are reported. Can I confidently conclude
that the RAM is OK?

On 4/3/2011 3:06 PM, Nil wrote:
I'm trying to troubleshoot a sick Windows PC (reboots randomly or
bluescreens with DRIVER_IRQL_NOT_LESS_OR_EQUAL errors.) I may have more
questions to ask about that later, but for now, I've run Memtest86+
twice now for about 4 passes of it's default tests each time, about 2.5
hours each time. No errors are reported. Can I confidently conclude
that the RAM is OK?

No. The only conclusion you can really draw from Memtest86+ is that the
memory is bad, or possibly that it has a mean time between failures
roughly equivalent to the length of your most recent test.

I had some bad RIMM that was sporadically preventing the machine from
POSTing. If it did POST, and I was able to boot up Memtest, it would
report no errors with the RAM. Memtest checks for certain types of
failure, but it is not exhaustive.

On 4/3/2011 3:06 PM, Nil wrote:
I'm trying to troubleshoot a sick Windows PC (reboots randomly or
bluescreens with DRIVER_IRQL_NOT_LESS_OR_EQUAL errors.) I may have more
questions to ask about that later, but for now, I've run Memtest86+
twice now for about 4 passes of it's default tests each time, about 2.5
hours each time. No errors are reported. Can I confidently conclude
that the RAM is OK?

I should have added in my last posting, that in spite of Memtest's
limitations, it is still a very useful tool, and I routinely use it to
check for gross defects in new RAM.

Nil wrote:
I'm trying to troubleshoot a sick Windows PC (reboots randomly or
bluescreens with DRIVER_IRQL_NOT_LESS_OR_EQUAL errors.) I may have more
questions to ask about that later, but for now, I've run Memtest86+
twice now for about 4 passes of it's default tests each time, about 2.5
hours each time. No errors are reported. Can I confidently conclude
that the RAM is OK?

Paul said here one time that prime95 was more stressful but less
diagnostic than memtest86.

http://bit.ly/eqq3d1

--
Mike Easter

Nil wrote:

I'm trying to troubleshoot a sick Windows PC (reboots randomly or
bluescreens with DRIVER_IRQL_NOT_LESS_OR_EQUAL errors.) I may have more
questions to ask about that later, but for now, I've run Memtest86+
twice now for about 4 passes of it's default tests each time, about 2.5
hours each time. No errors are reported. Can I confidently conclude
that the RAM is OK?

One person found no errors with MemTest86+, but when he tried Gold
Memory 6.92 he found one bad location, and it took another 9 hours to
detect that error again. IOW your 10 hours of testing may not be
enough. Gold Memory rated well in RealWorldTech.com 's evaluation
from several years ago, beating MemTest86 and bettered only by the
expensive PhD RST products. BTW, despite MemTest86+ being based on
MemTest86, I've found the two give different results. The same goes
for Gold Memory v. 5.07 and v. 6.92.

Unless you're using modules whose RAM chips are clearly marked with
the actual chip manufacturer's (Samsung/SEC, Micron/Inotera/Nanya,
Hynix, ProMOS, PowerChip) logo or part number, be very suspicous of
the quality, especially if the voltage rating is higher than normal
(over 1.5V for DDR3, overl 1.8V for DDR2).

On 03 Apr 2011, Mike Easter wrote in
alt.comp.hardware.pc-homebuilt:

Paul said here one time that prime95 was more stressful but less
diagnostic than memtest86.

http://bit.ly/eqq3d1

I'd like to try it, but it's a Windows application, and I can't get
Windows to stay up for more than a few minutes at a time.

Nil wrote:
Mike Easter

Paul said here one time that prime95 was more stressful but less
diagnostic than memtest86.

http://bit.ly/eqq3d1

I'd like to try it, but it's a Windows application, and I can't get
Windows to stay up for more than a few minutes at a time.

The latest Hiren's has prime95.

http://blog.sisq.info/en/2011/02/10/...s-bootcd-13-1/ Latest
Hiren’s BootCD 13.1 - New apps: Prime95

http://www.hiren.info/pages/bootcd Testing Tools - Prime95 25.11 This
will detect for errors in CPU or RAM within a matter of minutes if an
overclock is not stable, you can run Torture Test (burn-in) overnight to
ensure long-term stability of the hardware.

Be careful where you get it and be nice about copyright issues :-)


--
Mike Easter

Nil wrote:
On 03 Apr 2011, Mike Easter wrote in
alt.comp.hardware.pc-homebuilt:

Paul said here one time that prime95 was more stressful but less
diagnostic than memtest86.

http://bit.ly/eqq3d1

I'd like to try it, but it's a Windows application, and I can't get
Windows to stay up for more than a few minutes at a time.

Prime95 is available for Linux as well as Windows. Boot
a Linux LiveCD, go to mersenne.org/freesoft and get your
Linux test software there.

http://www.mersenne.org/freesoft/

Before Prime95 had a multi-threaded version, you could
run multiple single copies of Prime95, by keeping each
of them in a separate folder in the Linux home directory.
That's how I used to thrash a machine with the single
threaded version. Now that multi-threaded versions
exist, that job should be easier to do, with a single
copy running. (The multi-threaded version, means you
get to run at least a single thread per core.)

Using that technique, I ran four copies of Prime95 in
Linux, and could consistently get failures in a particular
"quadrant" of the memory space. But that doesn't necessarily
mean I can easily map what I'm seeing, to a particular
DIMM. So while it was fun to do, it didn't make locating
the culprit any easier.

One benefit of using multiple separate copies, would be
the ability to test machines with very large memory. I
don't know what upper limit Prime95 has for its memory
footprint. It may have been around 1600MB on the Windows
version.

An even more stressful test, is Prime95 running at the
same time as a 3D application. I used to use 3DMark2001
for that, but there are other options. That creates more
stress, than Prime95 alone. You start Prime running, then
start 3DMark. I've had a computer crash in 3DMark
by doing that, with the audio stuck in a loop, so it
does add an extra bit of stress.

Memtest86+ is mainly useful for stuck-at faults. It
detects dynamic faults too, but if the dynamic faults
only occur under extreme stress, you'll never find them.

There are better things than Prime95, but I've stopped
tracking "burn in" options. There is some Intel tool
for example, that the enthusiasts like. But I haven't
investigated replacements for Prime95, in quite a while.

Paul

In article , Nil
wrote:
I'm trying to troubleshoot a sick Windows PC (reboots randomly or
bluescreens with DRIVER_IRQL_NOT_LESS_OR_EQUAL errors.) I may have more
questions to ask about that later, but for now, I've run Memtest86+
twice now for about 4 passes of it's default tests each time, about 2.5
hours each time. No errors are reported. Can I confidently conclude
that the RAM is OK?

Try updating your video cards drivers. I dont IMHO think its ram related.

Paul nospam needed.com wrote:

Using that technique, I ran four copies of Prime95 in Linux, and
could consistently get failures in a particular "quadrant" of
the memory space. But that doesn't necessarily mean I can easily
map what I'm seeing, to a particular DIMM. So while it was fun
to do, it didn't make locating the culprit any easier.

That is some serious techieness.