problem with memory modules

Hello,

my system has a problem with the memory:

Running memtest on my four modules results in errors.

Running memtest on 2 of them also results in errors.

But running memtest on each of the 2 above does not result in errors.

Can you help?

http://www.bavaria64.de/tmp/memtest/

Magnus

On Sat, 11 Dec 2010 17:19:08 +0100, Magnus Warker
wrote:

Hello,

my system has a problem with the memory:

Running memtest on my four modules results in errors.

Running memtest on 2 of them also results in errors.

But running memtest on each of the 2 above does not result in errors.

Can you help?

http://www.bavaria64.de/tmp/memtest/

Magnus

Now I am no expert at all, only a average computer user...
But I have understood that Memtest (86+?) scans using random patterns,
so it *could* be (I think) that a first time you find an error and a
second time -Memtest having chosen another searchpattern- it will not
find this error and that you will have to scan each module or
combination of modules several (many?) times before the outcomes will
be the same.

Just my 2 cents.

By the way, have you disabled legacy usb in the BIOS while testing? If
I remember well that could be the cause of fake errors in the past.

Massimo

Magnus Warker wrote:
Hello,

my system has a problem with the memory:

Running memtest on my four modules results in errors.

Running memtest on 2 of them also results in errors.

But running memtest on each of the 2 above does not result in errors.

Can you help?

http://www.bavaria64.de/tmp/memtest/

Magnus

First, AMD memory controllers are load sensitive. I can't find
datasheets to verify the numbers, but it's possible your
processor supports 2 sticks in dual channel mode at DDR3-1333 and
4 sticks in dual channel mode at DDR3-1066. You might try
dropping the memory bus clock and retesting the 16GB case.

I'm willing to bet, if you switch to DDR3-1066 setting with
the four sticks, it will work.

*******

In the BIOS, set the 1T/2T command timing to 2T.
The motherboard is probably already doing that.

http://www.xbitlabs.com/images/mainb...bios_dram2.png

Check the operating voltage (Vdimm) setting that
is allowed by the RAM. I can't find a good datasheet
for a Corsair 4GB module as an example for you.
The DIMMs intended for Intel DDR3 motherboards,
are voltage limited to 1.65V by the processor itself.
So 1.65V is not necessarily the practical limit.
It's a recommended limit for Intel systems. AMD is different.

Perhaps the proper is recorded on the memory package,
as to what voltage was used for the Corsair testing.
The AMD processor and its memory controller, can take
more voltage than the Intel one on the Core i7.

*******

Given your memory test results so far, you could
send back the memory that isn't passing. You may
get a working memory in return.

Time S1 S2 S3 S4 Result
2010-12-09 1 1 2 3 4 Error
2010-12-10 1 1 - 3 - Error
2010-12-10 2 2 - 4 - OK
2010-12-11 1 1 - - - OK
2010-12-11-2 3 - - - OK

I would add the following test cases. These test
single channel operation, but with two sticks.

New 1 3 - - ???
New 3 1 - - ???

Both tests should fail. Note the address value reported
in each case. Perhaps with those test cases, you'll be
able to tell which stick is bad ? That is, if the sticks
can be returned separately. Some dual channel kits (2 x 4GB),
when you return (RMA) them, they both have to be returned
at the same time.

Paul

Hello Paul!

Thanks for your recommendations!

Paul wrote:

First, AMD memory controllers are load sensitive. I can't find
datasheets to verify the numbers, but it's possible your
processor supports 2 sticks in dual channel mode at DDR3-1333 and
4 sticks in dual channel mode at DDR3-1066. You might try
dropping the memory bus clock and retesting the 16GB case.

I'm willing to bet, if you switch to DDR3-1066 setting with
the four sticks, it will work.

Ok, I would like to try this! How can I make this setting (if it's not
the "1T/2T command timing" setting below)?

In the BIOS, set the 1T/2T command timing to 2T.
The motherboard is probably already doing that.

Hm, when I set "DDR Timing Items" to "Manual", the option "1T/2T command
timing" is already set to "2T". Can I assume that it always was set
to "2T", even in the "Auto" case for "DDR Timing Items"?

Check the operating voltage (Vdimm) setting that
is allowed by the RAM. I can't find a good datasheet
for a Corsair 4GB module as an example for you.
The DIMMs intended for Intel DDR3 motherboards,
are voltage limited to 1.65V by the processor itself.
So 1.65V is not necessarily the practical limit.
It's a recommended limit for Intel systems. AMD is different.

Perhaps the proper is recorded on the memory package,
as to what voltage was used for the Corsair testing.
The AMD processor and its memory controller, can take
more voltage than the Intel one on the Core i7.

Unfortunately, I do not have a package anymore.

Given your memory test results so far, you could
send back the memory that isn't passing. You may
get a working memory in return.

But didn't you say above that there is probably an incompatibility between
the board settings and the modules?

I still cannot tell if some modules are defective, and which modules...

I would add the following test cases. These test
single channel operation, but with two sticks.

New 1 3 - - ???
New 3 1 - - ???

Ok, the scan is running...

How long would you recommend scanning? I usually wait until pass 3.

Both tests should fail. Note the address value reported
in each case.

Ok!

Perhaps with those test cases, you'll be
able to tell which stick is bad ? That is, if the sticks
can be returned separately. Some dual channel kits (2 x 4GB),
when you return (RMA) them, they both have to be returned
at the same time.

Well, the sticks were bought at the same time, but separately. I could
return them separately, but maybe it would be better to return them all? I
still don't know if some are defective or not...

Magnus

Magnus Warker wrote:
Hello Paul!

Thanks for your recommendations!

Paul wrote:

First, AMD memory controllers are load sensitive. I can't find
datasheets to verify the numbers, but it's possible your
processor supports 2 sticks in dual channel mode at DDR3-1333 and
4 sticks in dual channel mode at DDR3-1066. You might try
dropping the memory bus clock and retesting the 16GB case.

I'm willing to bet, if you switch to DDR3-1066 setting with
the four sticks, it will work.

Ok, I would like to try this! How can I make this setting (if it's not
the "1T/2T command timing" setting below)?

In the BIOS, set the 1T/2T command timing to 2T.
The motherboard is probably already doing that.

Hm, when I set "DDR Timing Items" to "Manual", the option "1T/2T command
timing" is already set to "2T". Can I assume that it always was set
to "2T", even in the "Auto" case for "DDR Timing Items"?

Check the operating voltage (Vdimm) setting that
is allowed by the RAM. I can't find a good datasheet
for a Corsair 4GB module as an example for you.
The DIMMs intended for Intel DDR3 motherboards,
are voltage limited to 1.65V by the processor itself.
So 1.65V is not necessarily the practical limit.
It's a recommended limit for Intel systems. AMD is different.

Perhaps the proper is recorded on the memory package,
as to what voltage was used for the Corsair testing.
The AMD processor and its memory controller, can take
more voltage than the Intel one on the Core i7.

Unfortunately, I do not have a package anymore.

Given your memory test results so far, you could
send back the memory that isn't passing. You may
get a working memory in return.

But didn't you say above that there is probably an incompatibility between
the board settings and the modules?

I still cannot tell if some modules are defective, and which modules...

I would add the following test cases. These test
single channel operation, but with two sticks.

New 1 3 - - ???
New 3 1 - - ???

Ok, the scan is running...

How long would you recommend scanning? I usually wait until pass 3.

Both tests should fail. Note the address value reported
in each case.

Ok!

Perhaps with those test cases, you'll be
able to tell which stick is bad ? That is, if the sticks
can be returned separately. Some dual channel kits (2 x 4GB),
when you return (RMA) them, they both have to be returned
at the same time.

Well, the sticks were bought at the same time, but separately. I could
return them separately, but maybe it would be better to return them all? I
still don't know if some are defective or not...

Magnus


http://techreport.com/articles.x/16382

"One caveat: the Phenom II only supports 1333MHz DDR3 - at least,
officially - with a single DIMM in each memory channel. With
four DDR3 DIMMs, 1066MHz is the standard."

And that is the reason I suggested testing four DIMMs,
with the clock rate on the memory turned down.

I'm not keeping careful records, of the current rules
for AMD processors. The AMD site now has poor documentation,
so you can't really determine the rules, by consulting with
AMD. I have to dig up articles like the one above,
to get the information.

To control memory clock, set the "Set Memory Clock" to [Manual]
and "Memory Clock" [5.33x], 5.33 times 200MHz = DDR3-1066.

http://images.hardwarecanucks.com/im...e/20-Bios5.jpg

*******

The memory can have a "stuck-at" fault. That would be a case,
where the same memory bit is bad, each time it is tested.
None of the information below, will change a stuck-at fault
type.

Memory can also have dynamic faults. Those can be caused by
slow bus timing (timing failure on either the memory
controller or the memory).

For dynamic faults, there are several considerations.

1) Does the processor manufacturer guarantee operation
under the conditions you're using ? I have trouble
finding recommendations directly from AMD. The Techreport
article above, says Phenom II supports four DIMMs at
DDR3-1066 and two DIMMs at DDR3-1333. And those figures,
are not a function of your memory. Your memory is rated
for DDR3-1333. If you test with two DIMMs, in dual channel
configuration, then if the memory passes there at
DDR3-1333, the memory is considered good.

By those rules, these modules are OK (testing at DDR3-1333)
has proved the memory's specification is met).

2010-12-10 2 2 - 4 - OK

and this pair of modules is bad, as you tested these at
1333 and they failed. Granted, the processor might be
defective, but the previous test has conveniently proved
that it is not.

2010-12-10 1 1 - 3 - Error

If we change the memory clock to DDR3-1066, since it is
a suggested speed for four DIMMs, then we need to retest
this test case. If the DIMMs were actually good, this
might be a means of proving the processor is OK at this
level of bus loading. This speed should not stress the RAM
that much.

2010-12-09 1 1 2 3 4 ??? at DDR3-1066

2) The Command Rate 2T setting, increases the setup time
when passing address or command information to the DIMMs.
Normally, the BIOS would use the relaxed timing 2T, whenever
a stressful memory configuration is present (four DIMMs
present, or the user has selected a high clock speed).
Since you have examined the Default value, and it reads
2T, then this setting is already "relaxed".

3) Memory bus performance, is affected by the voltage supplied.
You can get slight improvements in timing, with higher
voltages. Higher voltages also cause the module to get
warmer. You either download a datasheet for the memory
module, to get voltage information, or you consult the
label affixed to the DIMM packaging. (You shouldn't throw
out the tray the memory comes in, as it is an antistatic
plastic material, suited for transport of the modules.)

*******

At some point, you have to make a decision as to whether
a DIMM module is bad or not. You can play with the adjustments,
and try and make them work. Or, you can be insistent with the
supplier, that the modules are defective.

The two new test cases, are intended to indicate which of the
two sticks is bad. By using (1,3,x,x) and (3,1,x,x), that is
a single channel configuration, with two DIMMs on the same
channel. My purpose in doing that, is to make the same loading
as your failing (four DIMM) test case. The four DIMM case,
puts two DIMMs on a channel, and so does (1,3,x,x).

By placing them on the same channel, it makes interpreting
the failure addresses of the memory, easier to understand.
Since the modules are on the channel, the addresses are
sequential (they are not interleaved). One module will
own the low_addresses, the other module will own the
high_addresses. Doing (1,3,x,x) and (3,1,x,x) should
cause the failure, to follow one of the modules when
it is moved.

*******

I don't rely on memtest86+, as a complete test result.
And because of that, a couple passes as a quick test
is sufficient. Memtest86+ works best as a stuck-at
fault tester. For dynamic faults, Prime95 is better.

If you wish to accelerate memtest86+ testing a little
bit, you can use the controls to advance to Test 5
and do your testing there. That will help you get
to the failing cases faster.

I use Prime95 stress tester, as an enhanced test. You
can boot a Linux LiveCD (selecting an appropriate version
for your 16GB memory space), and run multiple copies of
Prime95 (download from mersenne.org/freesoft).
If Prime95 will run error free for a period of four
hours, I'm pretty happy with that. You may need
multiple copies of Prime95, in order to cover
the majority of the 16GB address space.

There are other stress testers, but Prime95 is the
one I'm familiar with.

If you need more help, please post the exact
part number of the memory. The information below
is not sufficient, for me to look for the
voltage information. The number will be
something like CMX8GX3M2A1333C9 (8GB, DDR3, 2 modules).
Their single module products, use an entirely
different number.

4 * Corsair DDR3 1333 PC3-10667U 4096MB XMS3 CL9

HTH,
Paul

"Magnus Warker" wrote in message
...
Hello,

my system has a problem with the memory:

Running memtest on my four modules results in errors.

Running memtest on 2 of them also results in errors.

But running memtest on each of the 2 above does not result in errors.

Can you help?

http://www.bavaria64.de/tmp/memtest/

Magnus

I had a system with identical problems - in my case the problems only arose
after a problem with the graphics card meant that the graphics card got
fried - I ended up replacing the motherboard and after that it worked fine.
Although it was difficult to find a compatible motherboard because it was
quote old.

So, has it ever worked? Are all the components new or old? Has anything else
gone wrong recently?
--
Brian Cryer
http://www.cryer.co.uk/brian