memtest86+ errors

bdzyub DeleteThis @gmail.com wrote:
> I was just stress testing a core 2 quad computer with 4 GB of ram at
> the default frequencies. I am going to use this computer for fairly
> high precision scientific calculations. I just ran memtest86+
> overnight, and it produced 2 errors. How stable is this computer?
> Can someone point me to somewhere that explains what the errors can
> mean.
>
> Just recieving the 2 errors surprised me, because I had a bad memory
> dimm once in one of my laptops, and when memtest86+ produced errors,
> it produced them by the hundreds.
>
> How bad are 2 errors in a 12 hour period?
>

Solution is simple. Bump up the Vdimm voltage to the memory.

Standard voltage for DDR2 is 1.8V. I was looking at a Micron datasheet
last night, and absolute max is 2.3V. So you could easily dial up the
Vdimm voltage to 2.1V and retest.

For memtest86+, I recommend a couple passes or a couple hours. There
is another test, which is better, and can be used after you are
finished with memtest86+. And that is the Prime95 Torture Test
(mersenne.org). That carries out some math calculations with a known
answer, and also allocates a large chunk of RAM, for test purposes.
Prime95 seems to be more sensitive to computational correctness, than
memtest86+.

The Orthos version, allowed more than one thread to execute, with that
test code. What I don't know, is if it starts four threads for a quad
or not. You can download that version of Prime95 tester here:

http://sp2004.fre3.com/beta/beta2.htm

If you are still having problems, then post back with the brand
and model of RAM sticks used, the brand and model of motherboard,
for further comments.

If you are relying on the accuracy of these calculations, you
should be using RAM equipped with ECC. There are some motherboards
that have just come out, using the X38 chipset, and those are
supposed to support ECC for an Intel processor with FSB up to
FSB1333.

Paul
 >> Stay informed about: memtest86+ errors 

bdzyub DeleteThis @gmail.com wrote:
> ok, so what torture test should I perform: the blend or the in place
> large FFT?
>

Blend is fine. It does a little of everything.

Small FFTs tend to stay in cache. This stresses the CPU, heats it up,
tests your cooling and your CPU.

Large FFTs are larger than the cache. So tend to stay more in system
memory, results in some transfers from main memory. More of memory
gets tested that way.

If you want finer control of Prime95, you can run the mersenne.org version
under Linux. As long as each copy of Prime95 under Linux, is given its own
directory, you can run multiple copies. For example, I've run four copies
on my P4 system (with Hyperthreading, my processor looks like two cores). I
assign a different amount of RAM to each program, and each program executes
the blend. But with the different amounts of RAM, the programs cannot use
the same test pattern. I find that is a pretty good test.

Note that Prime95 style testing, doesn't say "location X in memory is bad".
It reports a rounding error, but without details. By using the Linux testing
method, and launching multiple copies, you can give each copy its own section
of RAM to test, and I have found consistent failures that way. (For example,
with four test programs, the third program always gave errors, so the section
of memory I gave it, must have a bad location in it. Obviously, the order of
execution, and the size of memory allocation, must be consistent from run to
run, to note something like that. A reboot between runs, will put the memory
allocation system in a consistent state.)

Paul
 >> Stay informed about: memtest86+ errors 

bdzyub DeleteThis @gmail.com wrote:

> I was just stress testing a core 2 quad computer with 4 GB of ram
> at the default frequencies. I am going to use this computer for
> fairly high precision scientific calculations.

If your application is critical, take Paul's advice and use ECC
mainboard/memory.

> I just ran memtest86+ overnight, and it produced 2 errors. How
> stable is this computer?

How critical is your application?

> Can someone point me to somewhere that explains what the errors
> can mean.
>
> Just recieving the 2 errors surprised me, because I had a bad
> memory dimm once in one of my laptops, and when memtest86+
> produced errors, it produced them by the hundreds.

Because you got lots of errors before doesn't mean that two errors
is okay. Typically you should get zero errors during a test.














>
> How bad are 2 errors in a 12 hour period?
>
 >> Stay informed about: memtest86+ errors 

Paul wrote:
>
> The Orthos version, allowed more than one thread to execute, with that
> test code. What I don't know, is if it starts four threads for a quad
> or not. You can download that version of Prime95 tester here:
>
> http://sp2004.fre3.com/beta/beta2.htm
>

It does not as yet. Simply create a second Orthos directory, then run two
instances concurrently.
 >> Stay informed about: memtest86+ errors 

<bdzyub RemoveThis @gmail.com> wrote in message news:1192476384.471522.49520@v23g2000prn.googlegroups.com...
>I was just stress testing a core 2 quad computer with 4 GB of ram at
> the default frequencies. I am going to use this computer for fairly
> high precision scientific calculations.

If you're doing high precision scientific calculations, you
should definitely be using a computer with ECC memory.
Better yet, a computer with Chipkill ECC memory.

The only acceptable result for memtest86+ is zero errors.

-- Bob Day
http://bobday.vze.com


> I just ran memtest86+
> overnight, and it produced 2 errors. How stable is this computer?
> Can someone point me to somewhere that explains what the errors can
> mean.
>
> Just recieving the 2 errors surprised me, because I had a bad memory
> dimm once in one of my laptops, and when memtest86+ produced errors,
> it produced them by the hundreds.
>
> How bad are 2 errors in a 12 hour period?
>
 >> Stay informed about: memtest86+ errors 

bdz....DeleteThis@gmail.com wrote:

> I was just stress testing a core 2 quad computer with 4 GB of ram at
> the default frequencies. I am going to use this computer for fairly
> high precision scientific calculations. I just ran memtest86+
> overnight, and it produced 2 errors. How stable is this computer?
> Can someone point me to somewhere that explains what the errors can
> mean.
>
> Just recieving the 2 errors surprised me, because I had a bad memory
> dimm once in one of my laptops, and when memtest86+ produced errors,
> it produced them by the hundreds.
>
> How bad are 2 errors in a 12 hour period?

Totally unacceptable. If the rest of your computer is OK, then the
memory should always work perfectly at its SPD defaults, even without
any tweaking.

What brand of memory is this? I've had bad luck with Kingston,
Mushkin, and K-byte, while Micron, Corsair, and PNY always worked
fine.

You may also want to try the www.goldmemory.cz diagnostic because it
sometimes finds errors better than MemTest86 does, but it's important
to test with both because I've also had memory modules pass Gold
Memory but fail MemTest86.

Here are some evaluations of several memory diagnostics:

www.realworldtech.com/page.cfm?ArticleID=RWT052001232443

www.realworldtech.com/page.cfm?ArticleID=RWT120901222920
 >> Stay informed about: memtest86+ errors 

bdzyub.TakeThisOut@gmail.com wrote:

> How bad are 2 errors in a 12 hour period?

If you don't care whether your computer gives correct answers, then the
errors don't matter at all.
 >> Stay informed about: memtest86+ errors