BIOS setting "Memory over voltage" shows value in red

Somewhere on teh intarwebs Paul wrote:
Maurice Batey wrote:
On Thu, 04 Mar 2010 20:15:35 -0500, Paul wrote:

It still doesn't
tell us though, whether the motherboard supports EPP.

Does the re-setting (from presumably the default of AUTO) of:

DRAM Frequency -- 1066MHz
Memory Over Voltage -- 2.100000V

not indicate that it *does* support EPP?

(The supplier of this PC say they did not make any changes to the
BIOS settings - though I am pressing them on this point, as it was
they who fitted the 1066MHz RAM..)

I could probably suggest ways you could test for support
of EPP yourself, but at this point, why mess up your
system ?

If there was a method that would not irretrievably mess up my
system, I might be curious enough to try it!

Regards,

On thinking about it, I don't see any easy way to verify whether the
BIOS supports EPP.

I've been searching for evidence of EPP support, and generally
it seems to get mentioned in adverts for Nvidia chipset boards.
I suppose this is only natural, as Nvidia and Corsair were the
first two to support EPP. Other than that, I don't seem to see
it listed as a feature in the feature section of the manuals.

The thing is, if Vdimm is set to "Auto" in the BIOS, you have no
way of reading out the value it is using. You'd need a utility
of some sort, running in Windows, to figure it out. Or, you
could use a multimeter, and (somehow) figure out where to probe
to measure Vdimm.

My motherboard has a whole bunch of voltages that can be adjusted,
but only seems to have four voltages measured on the hardware
monitor. So I don't have a way to measure the actual Vdimm
delivered, using just the motherboard. I think I'd need
a multimeter, if the BIOS was set to "Auto", to figure out
what it was actually using.

The information exists. There is a hardware register somewhere,
that is used to control Vdimm, and that means there should be
a way to read it out. But if no one designs a utility to do that,
then it is going to be pretty hard to figure out.

Hi Paul, Maurice. Sorry to butt in on your conversation. I think it very
likely that the mobo supports the feature. Unforunately it's hard to find
out on Google as EPP also applies to parallel ports and most of the search
strings I could complie ended up with hits that applied to this. Even the
reviews on overclocking sites I checked mentioned 64-step vDIMM adjustment
but nada about EPP w/r/t vDIMM.

I even went to ASUS' site and downloaded their manual for the mobo but it
isn't mentioned in there. However, as someone who's used a lot of Asus
boards in the past to build machines for gamers who insist on having hi-spec
RAM (even though they wouldn't know the difference from standard RAM...)
I've run into trouble with modules in the past that require
higher-than-default vDIMM.

I can't remember the exact details now but on one build the machine kept
crashing before I could even get into BIOS to raise the vDIMM. It was a C2Q
so it's not ancient history. Running Corsair RAM. I think that the SPD told
it to set the speed at a rate that it couldn't sustain at default vDIMM and
it wouldn't run long enough for me to change vDIMM. I ended up using a stick
of proper RAM to set the vDIMM to 2.1V, then shutting it down quickly and
putting the wank RAM back in. ('Wank ram' because IMO it's only good for
bragging rights, the money's usually much better spent elsewhere when
building a new machine when it comes to return on investment. Still, the
customer is always right, even when they tell me I'm wrong. :-\.)

I did write a rather scathing email to Asus at the time complaining about
the issue but I doubt that they changed to using EPP on the strength of just
the on email. It does seem that it's the only answer to Maurice's query and
IMO it's a good thing.

However it's always been my policy when building machines for myself (or
when speccing them myself) to always run RAM that only uses the default
vDIMM. I've seen too many sticks of 'enthusiast RAM' die just out of
warranty to ever want it in a machine of mine, or a machine that I've built
and am likely to be supporting unless it's been asked for and my
protestations to the contrary have been over-ridden.
--
Cheers,
Shaun.

Ticking away the moments that make up a dull day...

~misfit~ wrote:
Somewhere on teh intarwebs Paul wrote:
Maurice Batey wrote:
On Thu, 04 Mar 2010 20:15:35 -0500, Paul wrote:

It still doesn't
tell us though, whether the motherboard supports EPP.
Does the re-setting (from presumably the default of AUTO) of:

DRAM Frequency -- 1066MHz
Memory Over Voltage -- 2.100000V

not indicate that it *does* support EPP?

(The supplier of this PC say they did not make any changes to the
BIOS settings - though I am pressing them on this point, as it was
they who fitted the 1066MHz RAM..)

I could probably suggest ways you could test for support
of EPP yourself, but at this point, why mess up your
system ?
If there was a method that would not irretrievably mess up my
system, I might be curious enough to try it!

Regards,
On thinking about it, I don't see any easy way to verify whether the
BIOS supports EPP.

I've been searching for evidence of EPP support, and generally
it seems to get mentioned in adverts for Nvidia chipset boards.
I suppose this is only natural, as Nvidia and Corsair were the
first two to support EPP. Other than that, I don't seem to see
it listed as a feature in the feature section of the manuals.

The thing is, if Vdimm is set to "Auto" in the BIOS, you have no
way of reading out the value it is using. You'd need a utility
of some sort, running in Windows, to figure it out. Or, you
could use a multimeter, and (somehow) figure out where to probe
to measure Vdimm.

My motherboard has a whole bunch of voltages that can be adjusted,
but only seems to have four voltages measured on the hardware
monitor. So I don't have a way to measure the actual Vdimm
delivered, using just the motherboard. I think I'd need
a multimeter, if the BIOS was set to "Auto", to figure out
what it was actually using.

The information exists. There is a hardware register somewhere,
that is used to control Vdimm, and that means there should be
a way to read it out. But if no one designs a utility to do that,
then it is going to be pretty hard to figure out.

Hi Paul, Maurice. Sorry to butt in on your conversation. I think it very
likely that the mobo supports the feature. Unforunately it's hard to find
out on Google as EPP also applies to parallel ports and most of the search
strings I could complie ended up with hits that applied to this. Even the
reviews on overclocking sites I checked mentioned 64-step vDIMM adjustment
but nada about EPP w/r/t vDIMM.

I even went to ASUS' site and downloaded their manual for the mobo but it
isn't mentioned in there. However, as someone who's used a lot of Asus
boards in the past to build machines for gamers who insist on having hi-spec
RAM (even though they wouldn't know the difference from standard RAM...)
I've run into trouble with modules in the past that require
higher-than-default vDIMM.

I can't remember the exact details now but on one build the machine kept
crashing before I could even get into BIOS to raise the vDIMM. It was a C2Q
so it's not ancient history. Running Corsair RAM. I think that the SPD told
it to set the speed at a rate that it couldn't sustain at default vDIMM and
it wouldn't run long enough for me to change vDIMM. I ended up using a stick
of proper RAM to set the vDIMM to 2.1V, then shutting it down quickly and
putting the wank RAM back in. ('Wank ram' because IMO it's only good for
bragging rights, the money's usually much better spent elsewhere when
building a new machine when it comes to return on investment. Still, the
customer is always right, even when they tell me I'm wrong. :-\.)

I did write a rather scathing email to Asus at the time complaining about
the issue but I doubt that they changed to using EPP on the strength of just
the on email. It does seem that it's the only answer to Maurice's query and
IMO it's a good thing.

However it's always been my policy when building machines for myself (or
when speccing them myself) to always run RAM that only uses the default
vDIMM. I've seen too many sticks of 'enthusiast RAM' die just out of
warranty to ever want it in a machine of mine, or a machine that I've built
and am likely to be supporting unless it's been asked for and my
protestations to the contrary have been over-ridden.

That was common on some of the older boards. You had to tell a person
to get a stick of "slow" RAM, so you could get the BIOS running, and
make any adjustments to the settings. If the CMOS battery ever dies,
and causes the settings to be lost, the same bootstrap procedure
with the "slow" RAM might be needed all over again. It's kind of a curse.

I expect the solution to this, was for Asus and the other companies, to
run a bit more Vdimm without telling anyone. That is the quickest way to
solve it. Some DIMMs come without the top speed setting stored in the SPD,
and that is a second way to solve it (from the DIMM maker's end of things).

I think all the RAM I have in the house here, is low voltage stuff. So far,
I haven't been stuck getting a board to POST.

And the companies making the memory, can actually make it without huge
Vdimm boost. Look at what happened to DDR3 when the Intel Core i7 came
along. Plenty of 1.65V RAM showed up, able to meet the Intel max spec.

Paul

On Mon, 08 Mar 2010 11:53:37 +1300, ~misfit~ wrote:

I think it very likely that the mobo supports the feature.

Seems the only explanation. (I'm stll awaiting a reply from ASUS
on this...) Thank you for your input, Shaun - much appreciated....

I even went to ASUS' site and downloaded their manual for the mobo
but it isn't mentioned in there.

That's what I did and found (or rather, didn't find!).

... it's always been my policy when building machines for myself
(or when speccing them myself) to always run RAM that only uses
the default vDIMM.

With hindsight, that would be my policy, too - but when I ordered
the PC I didn't realise that the higher frequency RAM needed to run
at 2.1V rather than the usual 1.8V.
The Corsair RAM does come with a 'lifetime waranty', which sounds
comforting...

So far, I am not aware of any problems. It was the accidental glimpse
of the Memory Over Voltage "2.1V" setting in *red* that started me
asking questions!

Perhaps I should do a check with Memtest/Prime95?
--
/\/\aurice
(Replace "nomail.afraid" by "bcs" to reply by email)

On Mon, 08 Mar 2010 16:50:10 +0000, I wrote:

Perhaps I should do a check with Memtest/Prime95?

Well, I installed memtest86+b V4.00 and let it loose. OK for 3-4 mins,
then the foot of the screen went red with a list of error reports.

Then quiet again, then more red - as if it found problems in a cluster.

After 4 hours it had found 320 errors, and I closed it down for the night.

So, although the system appeared to have been working well for 2 months, the
RAM is suspect, so I guess I have to return the PC for a warranty repair (and I
shall try to get the RAM replaced by normal-speed RAM...).

Is there an alternative to returning it, I wonder?

Would it be worth trying some adjustment? For example, increasing/decreasing
the Memory Over Voltage from 2.1V?

Another thing I noticed in the Ai Tweaker menu settings was the only other
setting that was not 'Auto': FSB Frequency, which is set at 333. Does that
look correct for 1066MHz RAM? I have no idea of its relaionship with the other
settings...

--
/\/\aurice
(Replace "nomail.afraid" by "bcs" to reply by email)

Maurice Batey wrote:
On Mon, 08 Mar 2010 16:50:10 +0000, I wrote:

Perhaps I should do a check with Memtest/Prime95?

Well, I installed memtest86+b V4.00 and let it loose. OK for 3-4 mins,
then the foot of the screen went red with a list of error reports.

Then quiet again, then more red - as if it found problems in a cluster.

After 4 hours it had found 320 errors, and I closed it down for the night.

So, although the system appeared to have been working well for 2 months, the
RAM is suspect, so I guess I have to return the PC for a warranty repair (and I
shall try to get the RAM replaced by normal-speed RAM...).

Is there an alternative to returning it, I wonder?

Would it be worth trying some adjustment? For example, increasing/decreasing
the Memory Over Voltage from 2.1V?

Another thing I noticed in the Ai Tweaker menu settings was the only other
setting that was not 'Auto': FSB Frequency, which is set at 333. Does that
look correct for 1066MHz RAM? I have no idea of its relaionship with the other
settings...


FSB Frequency is the input clock to the CPU. That clock, times its internal
multiplier, gives the core frequency. I don't recollect what your processor is,
so I can't make a concrete example.

If the processor is FSB1333, that is the bus transfer rate. The Intel bus is
"quad pumped" and that means it transfers four items per clock cycle. If
we take your 333MHz number and multiply it by four, that is where FSB1333
comes from.

The AI Tweaker was set to manual, so they could get at the DRAM setting.

You have a "DRAM Frequency" setting, and that probably reads DDR2-1066
right now.

In the chipset section is the "DRAM Timing Control" setting. Setting
that to [Manual] exposes the memory timings.

tCL 5
tRCD 5
tRP 5
tRAS 15

As an experiment, you could try changing those to 6-6-6-18, then
run memtest again. That is loosening the timing and making it
easier for the memory to pass.

The purpose of doing that test, is to see if the fault will go
away or not. You can return those to their previous setting
(5-5-5-15 or all Auto, whatever they were using), when you're
finished testing.

I wouldn't adjust the Vdimm voltage, as you're already using 2.1V
and there is no point is adding to that.

Depending on what your testing shows, may help you decide whether
to return the system or not. I don't think relaxing the timing
is going to make that much difference to your overall performance.
If the memory errors are still there, then it is either the RAM
at fault, or some other setting is off a bit.

Another setting in the DRAM Timing Control, would be Command Rate.
1T is tight, while 2T is loose. Left at Auto, I expect the BIOS
can probably figure out the correct value to use. You can verify
in CPUZ, what was used.

I've had several batches of RAM fail, so having a RAM failure
of some sort wouldn't be unusual. The last module I had fail,
one chip died completely and delivered garbage for its entire
contents. That makes plenty of scrolling errors in memtest.

Paul

On Tue, 09 Mar 2010 09:02:35 -0500, Paul wrote:

FSB Frequency is the input clock to the CPU. That clock, times its
internal multiplier, gives the core frequency. I don't recollect
what your processor is, so I can't make a concrete example.

Intel Core 2 Quad Q8300 2.50GHz 1333MHz FSB/4MB cache

You have a "DRAM Frequency" setting, and that probably reads
DDR2-1066 right now.

Yes.

In the chipset section is the "DRAM Timing Control" setting. Setting
that to [Manual] exposes the memory timings.

tCL 5
tRCD 5
tRP 5
tRAS 15

As an experiment, you could try changing those to 6-6-6-18, then
run memtest again. That is loosening the timing and making it
easier for the memory to pass.

Sounds an interesting thing to do, but as this PC is only 2 months
old and still under warranty, rather than tinker in areas that I'm
not familiar with I think I'll lean on the suppliers to take it back
for 'repair'.
(Of course, they may suggest a similar adjustment first...)

What I find odd is that - despite the plethora of memtest errors, the
system continues to perform meticulously.
But I'm backing up more often...

Many thanks once again, Paul. Much appreciated....
--
/\/\aurice
(Replace "nomail.afraid" by "bcs" to reply by email)

Maurice Batey wrote:
On Tue, 09 Mar 2010 09:02:35 -0500, Paul wrote:

FSB Frequency is the input clock to the CPU. That clock, times its
internal multiplier, gives the core frequency. I don't recollect
what your processor is, so I can't make a concrete example.

Intel Core 2 Quad Q8300 2.50GHz 1333MHz FSB/4MB cache

You have a "DRAM Frequency" setting, and that probably reads
DDR2-1066 right now.

Yes.

In the chipset section is the "DRAM Timing Control" setting. Setting
that to [Manual] exposes the memory timings.

tCL 5
tRCD 5
tRP 5
tRAS 15

As an experiment, you could try changing those to 6-6-6-18, then
run memtest again. That is loosening the timing and making it
easier for the memory to pass.

Sounds an interesting thing to do, but as this PC is only 2 months
old and still under warranty, rather than tinker in areas that I'm
not familiar with I think I'll lean on the suppliers to take it back
for 'repair'.
(Of course, they may suggest a similar adjustment first...)

What I find odd is that - despite the plethora of memtest errors, the
system continues to perform meticulously.
But I'm backing up more often...

Many thanks once again, Paul. Much appreciated....

If you want confirmation of what memtest86+ is finding, you should
give Prime95 (stress test option) from mersenne.org/freesoft a try. If
your memory was really as fault free as you propose, then you should be
able to run that for four hours without a problem. It will start
a thread per core, and a thread will stop on the first error it
encounters.

On my other systems, I found it to be a pretty good indicator of
system health.

I've had that program stop in as short a time as 0.5 seconds.

I have one 3D game demo here, that if there are memory errors, some
of the bots start to misbehave. They go off to one side of the map
and just stand there. So I have seen verification, that what the
other tests see, is really there. The funny thing is, the game
hardly ever crashes, but you do see bizarre behavior when the
memory is not completely stable. Since I put together the DDR2
based system I'm on now, I've stopped seeing those things.

If you let a computer sit idle in the desktop, it might take
100 hours of observation before there is a symptom. Testing with
a stress program, shortens that interval considerably. One of the
reasons for doing all the test cases, when a system is new, is to
make sure that any initial "merchandise returns" can be executed
before it is too late. You don't want to leave it sitting idle
for 10000 hours, only to find a problem the first time the
machine sees a bit of stress. Better to thrash the machine
while it is in the return period. The computer is, after all,
designed to run at 100% load, and the cooling system and
power supply should have the capacity to do that. Otherwise,
the system builder didn't do their job.

Paul

On Tue, 09 Mar 2010 11:16:25 -0500, Paul wrote:

f you want confirmation of what memtest86+ is finding, you should
give Prime95 (stress test option) from mersenne.org/freesoft a try.

No - I believe it! W.r.t. Prime95, I was inclined to wait until the
memory problem is sorted out, as running it before then
would presumably inevitably come across problems probably related to
the dodgy RAM, so would not tell me anything new (or would they?).
Does Prime95 try to diagnose a situation, or does it just throw
its hands in the air and say "Something wrong"?

If your memory was really as fault free as you propose,

But I don't propose that! If memtest says the memory is rubbish,
then rubbish it is...
(The only thing that puzzles me there is why the system doesn't
display symptoms when it's not idling.)

Better to thrash the machine while it is in the return period.

Yes, indeed.

The computer is, after all,
designed to run at 100% load, and the cooling system and
power supply should have the capacity to do that.

I had a '600W Quiet Quad Rail PSU' & 120mm case fan installed,
plus a 'super quiet 22dba triple copper heatpipe CPU cooler'.

--
/\/\aurice
(Replace "nomail.afraid" by "bcs" to reply by email)

On Tue, 09 Mar 2010 10:53:37 +0000, I wrote:

Well, I installed memtest86+b V4.00 and let it loose. OK for 3-4 mins,
then the foot of the screen went red with a list of error reports.

For pictures of the memtest screen (and BIOS Ai Tweaker menu) see:

http://www.maurice99.ukfsn.org/memtest1.jpg
http://www.maurice99.ukfsn.org/memtest2.jpg
http://www.maurice99.ukfsn.org/memtest3.jpg
http://www.maurice99.ukfsn.org/ai-tweaker.jpg

I have also run 64-bit Prime95 on Windows 7 (Home Premium).
After 5 minutes or so several of the 5 tasks it had launched
reported 'hardware error - see stress.txt', but I was unable to find
such a file. Where does it hide it?

--
/\/\aurice
(Replace "nomail.afraid" by "bcs" to reply by email)

Maurice Batey wrote:
On Tue, 09 Mar 2010 11:16:25 -0500, Paul wrote:

f you want confirmation of what memtest86+ is finding, you should
give Prime95 (stress test option) from mersenne.org/freesoft a try.

No - I believe it! W.r.t. Prime95, I was inclined to wait until the
memory problem is sorted out, as running it before then
would presumably inevitably come across problems probably related to
the dodgy RAM, so would not tell me anything new (or would they?).
Does Prime95 try to diagnose a situation, or does it just throw
its hands in the air and say "Something wrong"?

If your memory was really as fault free as you propose,

But I don't propose that! If memtest says the memory is rubbish,
then rubbish it is...
(The only thing that puzzles me there is why the system doesn't
display symptoms when it's not idling.)

Better to thrash the machine while it is in the return period.

Yes, indeed.

The computer is, after all,
designed to run at 100% load, and the cooling system and
power supply should have the capacity to do that.

I had a '600W Quiet Quad Rail PSU' & 120mm case fan installed,
plus a 'super quiet 22dba triple copper heatpipe CPU cooler'.


Prime95 is not a diagnostic and will not provide the level of
detail that memtest86+ does. It is an "acceptance test", in that
if none of the execution threads die, you would then feel fairly
confident to use it for ordinary activities.

One of the limitations of Prime95, is it doesn't test all of your
memory. So in some ways, it is more of a "system test", rather than
being a tool to guarantee that each and every memory location is
function. And this is why, as a test, it is complementary to
the usage of memtest86+. You really need to use both of them,
to get some level of test coverage.

Your memory failure patterns are really strange! You'll notice
some of them, look like the memory locations never got written
with the test pattern in the first place. It isn't like a single
bit is at fault in each detected error.

One of the reasons for me suggesting Prime95, is as a cross-check
that this is not a problem with the memtest86+ program. After all,
each time they change a program like that, they can introduce new
errors. But since both Prime95 fails and so does memtest86+, then
there is likely a real problem there.

In CPUZ memory tab, is your command rate 2T ? Based on your memtest
errors, I might try a 2T setting for the command rate, if the BIOS
hasn't already selected that. Maybe if the memory was running 1T,
that would account for some locations simply not getting written.

Paul