New Build PC Spontaneously Reboots Sometimes Before OS Load

"Damaeus" wrote:

Paul:

The advertising mentions Intel. And possibly, because the modules
are equipped with optional Intel XMP profiles.

http://www.kingston.com/us/memory/hy...s/quad_channel

http://www.kingston.com/datasheets/K...1D3K4_16GX.pdf

JEDEC: DDR3-1333 CL9-9-9 @1.5V

Wow! I'll have to check the BIOS again, but automatic settings have the
voltage set to around 1.288 or something like that. Maybe too little
voltage is the problem.

Way too low. According to the Kingston, the *non-overclocked* voltage
is 1.5. For the OVERCLOCKED setup, Kingston says to use 1.65V. You
said before it was 1.4. Too low. Now it's even worse?

In news:alt.comp.hardware.pc-homebuilt, VanguardLH posted on
Fri, 11 Jan 2013 10:12:24 -0600 the following:

"Damaeus" wrote:

Wow! I'll have to check the BIOS again, but automatic settings have the
voltage set to around 1.288 or something like that. Maybe too little
voltage is the problem.

Way too low. According to the Kingston, the *non-overclocked* voltage
is 1.5. For the OVERCLOCKED setup, Kingston says to use 1.65V. You
said before it was 1.4. Too low. Now it's even worse?

That's what it was showing when the manual entry was at 1.4. When I
rebooted to the BIOS after setting it back on automatic, it was at 1.488.
I put up a few pages of the BIOS he

http://home.earthlink.net/~damaeus/crosshair.html

To be safe, I may just take it off automatic and put it on 1.5 volts.

What's also strange, but maybe it's normal, is that the voltage reading on
four settings was fluctuating. I watched for a few seconds in the Voltage
Monitor and got fluctuations in these readings:

VCore 1.284 - 1.296
CPU/NB 1.197 - 1.230
HT Voltage 1.204 - 1.210
NB Voltage 1.105 - 1.111

Damaeus

In news:alt.comp.hardware.pc-homebuilt, VanguardLH posted on
Fri, 11 Jan 2013 10:08:59 -0600 the following:

"Damaeus" wrote:

Boosting the speed to 2133 MHz and timing 11-12-11-30 increased
stability for about an hour and a half. Since then I've lowered the
speed back to automatic 1333 MHz and automatic 9-9-9-24.

I still got a reboot there after about 3.5 hours, but it's strange
how it just decided to stop rebooting so quickly with the same
timings I had at first. I then enabled an "optimize for 4 GB DIMMs",
which lengthens the timing a little to 10-12-11-28, and I can't be
sure yet until I run it for more than 3.5 hours, but it seems to have
increased the stability some more.

Here's what the settings look like with 4GB DIMM settings loaded:

http://home.earthlink.net/~damaeus/i...-crosshair.jpg

I've seen some memory brands that overclock their modules (by the spec
they sell at). You are led to believe the memory module is rated for
those specs but those are actually overclocked specs advertized for the
product so they can charge you more. When they claim the unit has been
tested at the overclocked values, that's not true. They have not been
"burned" for hours to test stability nor has every unit been tested but
just a sample.

Yes, I didn't realize the memory I bought was actually 1333 overclocked to
2133. And I didn't know it was just from a sample. I assumed that the
memory I actually got was in some testing gizmo and the ones found to run
at 2133 got one label, 1866s got another, 1600s got another, etc.... Well,
as long as it works well, I'm happy since I don't want to run them at
2133. I was kind of hoping to run them at 1866, though, since supposedly
the CPU will do that. If I had known about this kind of thing going on, I
would have done more research. But like I've said before, this computer
is still a whole lot better than my last one, so I'll still be happy with
it if it's as stable as my last one was.

http://www.kingston.com/dataSheets/K...1D3K4_16GX.pdf
"Each module kit has been tested to run at DDR3-2133 at a low latency
timing of 11-12-11 at 1.65V. The SPDs are programmed to JEDEC standard
latency DDR3-1333 timing of 9-9-9 at 1.5V."

Sounds fishy to me. They're saying that the modules have been
programmed with an SPD that has the memory run at a slower speed hence
the tighter timings yet they "test" then to run at the higher speed (and
looser timings). Sure sounds like they've taken DDR3-1333 modules and
are selling them as overclocked units. The chip maker says the product
is rated for 1333 at CAS9 but Kingston is overclocking that spec and
claiming (and hoping) the product works at 2133 CAS11. Try to get the
hardware working by not overclocking anything. You can play with
overclocking later where you decide if you want stability along with
continuous usability or speed at the expense of usability.

Well, yes. I never wanted to overclock anything. Yet at first, I
couldn't get it to stop rebooting with all automatically-detected
settings, running 1333 MHz and 9-9-9-24. I've rebooted a few times myself
today, but so far, I've now had it running without shutting it off for
eight hours without a *spontaneous* reboot. I don't know if my manual
reboots are resetting something or not. I'm going to stop doing that for
the rest of the day. I just had to go into the BIOS to get those screens.

Also, I have not had a reboot since disabling turbo mode. I noticed that
the CPU was getting "stuck" in turbo speeds. It's supposed to run at 3.5
GHz normally, but when CPU usage was almost nothing, it was still stuck at
3.8 GHz, but did come down from around 4.2 GHz. Well, I don't really care
about turbo mode, especially if it's going to get stuck there. And I have
those 4GB 10-12-11-28 DIMM timings set, and I'm more than happy with that
if it increases stability over 9-9-9-24.

I wonder if the memory needed to be "broken in"... run it at full 2133 MHz
for an hour and a half, then run them at normal speed. That's what
they've been through, and the reboots seem to be taking longer to happen,
assuming there's another one waiting to torture me. It's not a good
feeling to know that at any second, my screen could be showing me my BIOS
version against my will.

Is there decent airflow over the memory modules? Sometimes cables,
especially the flat IDE ones, are incorrectly oriented perpendicular
to the airflow instead of parallel to it. Of course, with your
external table fan blowing into the case, it's that airflow direction
you need to cool the memory cards.

Stupid me, I stuck my hand inside to check the air flow around the
thickest power cable and my fingertip accidentally touched the CPU fan.
It's a pretty powerful fan. But I cursed myself in the process. I'm
surprised that didn't make the screen display byte-trash.

While I was in there grabbing that screen shot, I noticed that the
voltage was set to manual at 1.4, and it was in yellow. I reset that
to automatic. I'm not sure yet how that'll affect stability.

These overclocked units at the 2133 clock require 1.65V. They're
overclocked. They need more power. They'll expell more heat. Even
at the module spec'd 1333 clock rate (what the chip manufacturer rated
for the module, not what Kingston decided to "test" and claim a higher
but overclocked rate), the modules are spec'ed at 1.5V. You said 1.4
was wrong but didn't mention what voltage the "automatic" setting
used.

Well, when I reset it to automatic, the voltage went to 1.488. Oddly,
even when I entered a manual setting of 1.65 volts at 2133 MHz, the actual
result wasn't *exactly* 1.65 volts, but just slightly less.

I remember once having to rework the cutouts. In every other case
where they were tough to get in, it just required a bit more jostling,
jiggling, and pushing to get the cutouts to align. Typically it's
just push in, push back, and click. The "sand" noise could be the
edge of the PCB for the cutouts scraping past the nubs in the slot
bracket.

Could be, but I don't want to have to take them out again until I build
another computer or buy new memory for whatever reason. Heh.

If these come with a folder-over heat plate, have you made sure they
are centered and not partially overlapping the side cutouts? Are
these just the cheap and worthless fold-over aluminum plates put on
just for looks or are they the ones that extend beyond the top of the
memory card with fins (which often results in problem in spacing with
other components)?

These don't have fins, just foldover plates with pretty graphics on them.
They look just like the ones you showed me at that link:

http://www.legitreviews.com/article/2030/1/

That's not why I bought them, though. I didn't even know what they were
going to look like. I just bought them because they were on the QVL for
this motherboard, and I could afford them, and they had free shipping, and
people were saying Kingston is a good brand. I didn't want OCZ because so
many people were saying that brand is not a good one. G.Skill was also on
the list, but I had never heard of them, and very few people were
suggesting that brand to others in the forums I searched. I didn't ask
anybody directly because there were already so many other people who had
asked before.

Anyway, these are Non-ECC memory cards. I wonder if I should disable ECC
mode in the BIOS for any kind of benefit or stability plus. I thought it
was strange that this motherboard can't even use ECC memory, and yet
there's an option for it in the BIOS.

The book says: "You may install 1GB, 2GB, 4GB, 8GB, and 16GB unbuffered
and non-ECC DDR3 DIMMs into the DIMM sockets."

From http://www.legitreviews.com/article/2030/1/, it looks like its
just the non-functional plate used for marketing glitz. Those just
slide over and are held by friction, don't have good contact since the
inside is painted or is unpolished, and no thermal paste is used plus
they provide little additional surface area for cooling than for the
chip tops themselves (and the lack of thermal paste and good surface
contact means poor heat transfer so the added surface area of the
plate is ineffective).

Why is it that I never hear about these things until I've already spent
money? Hopefully they'll still be fine running at 1333 MHz, which is all
I intend to run them at.

The holes in the mobo for the standoff screws might have a grounding
pad so, yes, you want the screws to contact them. Shorting happens
when flathead screws are used with oversized heads. The heads
obviously have to be bigger than the hole to hold the mobo but I've
seen some folks use really large headed screws which reached beyond
the grounding pad on the mobo. Also, not all mobo mounting points
have grounding pads. It's not the metallic standoffs that short
something away from the hole. It's using too-large headed screws.

The screws I used have a very tiny flange on them because they're also
made for a socket screwdriver, but nothing that extends past the little
circle of slight bulge around the screw hole. It doesn't even come close.

Did the mobo come with a packet of standoffs and screws? If so, and
if no phenolic washers were provided then you don't any assuming
you're also using the screws provided in the hardware kit. Or were
you salvaging the screws and maybe the standoffs out of a junk box?

The motherboard didn't come with any standoffs or screws. I wish it had
because I would have used them. I used the screws and standoffs I had in
my last build. It came with SATA cables and a USB cable, SLI connectors
for two or three video cards, stickers to label the cables, an RoG emblem
for the case, which I have stuck on the door on the case. (Yes, my case
has a door that hasn't been knocked off yet, even though it's about ten
years old.) But no screws.

Where there is a grounding pad around a standoff hole is where a screw
should contact it. That is what connects chassis ground to circuit
ground. Make sure the screw head does extend outside the grounding
contact. Not all mounting holes for the standoffs will have grounding
contacts. Make sure the metallic screw doesn't contact any foils,
components leads, or jumper wires next to those holes.

This one has no obvious silver rings around the holes like one motherboard
I had, but when installing the screws, there were little metal "dots" that
seemed to be peeking through the black layer that's on the motherboard for
looks. And the book doesn't say anything about needing to use washers.

Damaeus

Damaeus wrote:


Thanks. I'll give one of them a whirl. I don't know anymore what the
program runs under. I embarrassed myself in a web forum by referring to
it as DOS, and someone rather condescendingly told me that Memtest86 has
nothing to do with DOS.

Actually, it isn't DOS.

It is something pretty amazing, for modern computer users.

memtest86+ is a program that runs without any OS at all.

It bootstraps the way an OS might, the BIOS hands off to it
like it was an OS, but it's a dedicated program running
in Ring0. It has access to all hardware, never has permission
problems, because it owns the whole machine. As far as I know,
the low res graphics screen is a VESA access mode, and VESA
is a standard implemented in most all graphics cards. Only
a couple old Matrox cards, have busted VESA on them (the
screen repeats multiple times in VESA mode, and is hard to
read).

The disadvantage of running without an OS, is no threading
or scheduling would be provided by default. If a person
sought all the advantages an OS could provide, they'd need
a library of some sort with those functions in it. (At that
point, maybe the library would be running a form of microkernel.)
At best, all we can expect is that memtest86+ uses one processor
core, and only listens to user input, by polling occasionally
for it. The functionality is pared down, compared to running
with an OS.

Paul

Damaeus wrote:
In news:alt.comp.hardware.pc-homebuilt, VanguardLH posted on
Fri, 11 Jan 2013 10:12:24 -0600 the following:

"Damaeus" wrote:

Wow! I'll have to check the BIOS again, but automatic settings have the
voltage set to around 1.288 or something like that. Maybe too little
voltage is the problem.
Way too low. According to the Kingston, the *non-overclocked* voltage
is 1.5. For the OVERCLOCKED setup, Kingston says to use 1.65V. You
said before it was 1.4. Too low. Now it's even worse?

That's what it was showing when the manual entry was at 1.4. When I
rebooted to the BIOS after setting it back on automatic, it was at 1.488.
I put up a few pages of the BIOS he

http://home.earthlink.net/~damaeus/crosshair.html

To be safe, I may just take it off automatic and put it on 1.5 volts.

What's also strange, but maybe it's normal, is that the voltage reading on
four settings was fluctuating. I watched for a few seconds in the Voltage
Monitor and got fluctuations in these readings:

VCore 1.284 - 1.296
CPU/NB 1.197 - 1.230
HT Voltage 1.204 - 1.210
NB Voltage 1.105 - 1.111

Damaeus

Hardware monitor channels, have limited resolution. When viewed in the OS
level, the OS uses slightly better running average arithmetic, to make
the values look "smooth". The BIOS tends to show the quantization coming
from the actual analog to digital converter (ADC).

A typical cheesy hardware monitor, might measure 0 to 4.096 volts in
256 steps. That means roughly 16 millivolts per step. A regular multimeter
from the store, might have 2000 steps for comparison. The hardware monitor
is better than nothing, but is only roughly in the ballpark. It's
not "lab instrumentation quality". If something was way out of whack,
that's what those are for. To check to see if "something is burning".

Even a store bought multimeter, has fluctuation in the last digit.

Paul

"Damaeus" wrote:

Also, I have not had a reboot since disabling turbo mode. I noticed
that the CPU was getting "stuck" in turbo speeds. It's supposed to
run at 3.5 GHz normally, but when CPU usage was almost nothing, it
was still stuck at 3.8 GHz, but did come down from around 4.2 GHz.
Well, I don't really care about turbo mode, especially if it's going
to get stuck there. And I have those 4GB 10-12-11-28 DIMM timings
set, and I'm more than happy with that if it increases stability over
9-9-9-24.

You need to get rid of all the overclocking in the BIOS. Get to a
stable state first. Play around with overclocking later, if at all.

Isn't there an "optimal" config that puts the memory timings at 1333
CAS9 1.5V (as per the SPD defined by the chip maker and not Kingston's
hopeful overclocked values)?

http://dlcdnet.asus.com/pub/ASUS/mb/..._Formula-Z.pdf
page 3-3
Shows a button labelled "Defaults" for optimal settings. I'm assuming
that if this mobo lets you save a different set of defaults into the
EEPROM that either you haven't done that (i.e., altered the defaults) or
that saved config (flashed into the EEPROM, not saved in CMOS) isn't
used when you elect Defaults (i.e., you have to elect the saved config
to use that one). I would think Turbo mode would not be enabled when
using Defaults which I would expect to use optimal settings. Page 3-2
says "If the system becomes unstable after changing any BIOS setting,
load the default settings to ensure system compatibility and stability.
Select the *Load Optimized Defaults* item under the *Exit* menu or press
hotkey *F5*.

I wonder if the memory needed to be "broken in"... run it at full 2133
MHz for an hour and a half, then run them at normal speed. That's
what they've been through, and the reboots seem to be taking longer
to happen, assuming there's another one waiting to torture me. It's
not a good feeling to know that at any second, my screen could be
showing me my BIOS version against my will.

Burning in of electronics is to discover early any failing or weak
components. Things either break or stay the same during burn in. They
don't get better. The idea is to find failures before the device in put
into service.

http://en.wikipedia.org/wiki/Burn-in

Work on the hardware first. Don't bother even loading an OS until the
hardware is stable. The OS can't be stable if the hardware isn't. For
burn-in, you can use Prime95's Torture Test. You could customise a
BartPE image (http://www.nu2.nu/pebuilder/) or "Ultimate Boot CD for
Windows" (http://www.ubcd4win.com/index.htm) to add Prime95.

Well, when I reset it to automatic, the voltage went to 1.488. Oddly,
even when I entered a manual setting of 1.65 volts at 2133 MHz, the
actual result wasn't *exactly* 1.65 volts, but just slightly less.

Don't trust the voltages shown by the BIOS and monitored by sensors on
the mobo to be that accurate. That's why I mentioned using a multimeter
but even then measurements may only be close since rare few consumers
ever get their multimeters calibrated and also recalibrate them every
few years. They typically buy from the store and they sit forever in
the toolbox with an occasional exposure to sunlight too often with a
leaky old battery for the ohm function corroding the battery terminals.

Temperatures are "measured" by gradient changes in the components used
as sensors but their changes are not linear. Tables are used in the
Winbond chip to equate temperatures at certain points in the curve (the
table only has room for so many entries so temperature "readings" are
granular, not analog). There can also be a mismatch between the table
values in the Winbond chip and the curve for the sensor. Then there is
the hysteresis in sensors so that the reading shown in the BIOS may not
be the current temperature. At http://www.almico.com/sfhistory.php, you
can see all the updates in Speedfan trying to keep up with all the
various controller/monitor chips, versions, and their tables. Just
because mobo maker has a much smaller combo of chip and sensor to couple
together for temperature sensing doesn't mean they make the best choice
or there is a best choice but just a closest choice. There is also
intervening components, like maybe a 10K-ohm resistor between sensor and
chip so there's the variability (tolerance) in that component's value.
It's not like every mobo gets personalized tweaking to ensure the BIOS'
granularity is accurate. They design, they produce, they test, and then
they manufacture with perhaps sporadic sampling. They can't afford to
test every unit. That's not how manufacturing works.

Figure the BIOS readings are better than just seeing LEDS saying you're
close to the target voltage. A calibrated multimeter would be better.
A non-calibrated multimeter, especially years old and stored in the
garage, attic, car, or other non-temperature controlled environment,
could be a worse choice. Of course, have fun trying to find a spot
where you can measure the memory's voltage. Often you don't have access
to the backside of the mobo once it's mounted in the case.

Anyway, these are Non-ECC memory cards. I wonder if I should disable
ECC mode in the BIOS for any kind of benefit or stability plus. I
thought it was strange that this motherboard can't even use ECC
memory, and yet there's an option for it in the BIOS.

If you have non-ECC memory than you should not enable ECC in the BIOS.
You can't make the BIOS magically change the memory into a mode it
doesn't support. The BIOS should to match the memory. For non-ECC
memory, BIOS must be set to non-ECC. For ECC memory, you can choose
non-ECC or ECC mode -- but why pay the extra cost of ECC memory and not
use that feature?

http://en.wikipedia.org/wiki/ECC_memory

Non-ECC has become the norm for consumer-grade hardware, so much so that
often you won't even see it mentioned in the specs. If ECC isn't
mentioned, assume it's non-ECC. Did your BIOS really default to ECC
mode? Or was the consequence of using turbo or other overclock settings
in the BIOS? I'd be surprised if any BIOS linked ECC/non-ECC mode to
any timing configs.

On Friday, January 11, 2013 6:24:10 AM UTC-7, Damaeus wrote:

In news:alt.comp.hardware.pc-homebuilt, VanguardLH posted on

When you mounted the mobo to the case using the standoffs,
were they plastic standoffs onto which you push the mobo
or are they metallic threaded rods and you use screws
to hold down the mobo? If the metallic type, did you
use the phenolic insulating washers between the screws
heads and the mobo, or do the mobo have a nice grounding
pad around the hold through which the screw passes?
I have seen where metallic standoffs and metal screws
have caused shorts. In addition, I've found mobos that
had problems due to solder flakes or whiskers.

I have metal screw-in studs that go into a metal case.
I asked about using washers in my last build and someone
told me that those washers aren't needed because using
metal screws in metal studs "grounds" the motherboard to
the case. It looks like the printed circuits steer clear
of the mounting holes. If you think I should have
washers in this thing, I can get some and put them in.

These people are saying no washers are needed:

http://www.tomshardware.com/forum/24...ulated-washers

And this guy says no washers are needed because the
motherboard needs grounding:

http://www.hardwaresecrets.com/artic...herboard/145/2

That guy's wrong because motherboards are grounded through
the power connectors and rarely need additional grounding,
and you only want to ground grounds, not signals or voltages,
and there's no way to be sure about the standoffs (those
screw-in studs) or screws causing shorts unless you check
your particular motherboard. For example, here's a hole
that should get an insulator washer on this side:

http://farm3.static.flickr.com/2546/...aae2cb5b_m.jpg

Here's a hole that might need a washer:

http://farm4.static.flickr.com/3518/...9bdae2d8_m.jpg

And here's one that doesn't need one on this side:

http://farm4.static.flickr.com/3564/...88b58fc9_m.jpg

I had an ancient motherboard where the left side of one hole
had a ground plane and the right side had a +3.3V plane,
less than 1/8" apart. IOW don't rely on generalizations here.

On Friday, January 11, 2013 4:54:53 AM UTC-7, Damaeus wrote:

In news:alt.comp.hardware.pc-homebuilt, posted

I asked some guys in a web forum what to do and they said
to set the memory to run at its "rated" speed and timing.
So I set it to 2133 MHz and timing to 11-12-11-30, and
voltage to 1.65, which is what the specs are for this
memory.

But not for the memory chips because no chip maker rates
its visibly branded DDR3 for 1.65V ("visibly" means no
heatsinks covering the chips), and real 2133 MHz chips
are rarely found on 2133 MHz memory modules.

Kingston (or whoever makes these) supposedly tested it
and found that it would run at a maximum speed of 2133 MHz.
I don't want to run it that fast, of course. I got it
because I figured if it would run at 2133 MHz, then it
would run even better at slower speeds. Its standard
speed is 1333 MHz, which I'm happy with. I like the
overhead and I'd rather run memory capable of 2133 at 1333
instead of running memory capable of only 1333 at 1333. The
former is like having a car that'll go 213 miles an hour,
but I only drive it at 133. lol

Maybe I can get an exchange on this if it shows errors. I
researched the brand and Kingston was listed along with
Mushkin, Crucial and Corsair as good brands. But these
were web forums. I guess people were going by their
experiences.

Someone in a web forum suggested my battery backup might
be going bad.

That won't matter unless you have a true UPS, where the power
coming out of it is always from the battery. But most units
are actually just backup power sources that normally connect
the household AC directly to the outputs, through a relay,
so the condition of the battery won't matter, except maybe
when the backup supply is turned on -- many units won't
allow being turned on if the battery is really bad.

That was the only reason I changed the power source.
It seems fine and it passes its self-test.

BTW many batteries over about 3 years old will pass the
self-test but won't provide anything close to the rated
run time but instead only 30-60 seconds.

I don't think your PSU is bad, and most likely you're
experiencing bad memory (Kingston is made of no-name
or overclocked chips) or software, probably the latter.

Sigh... And someone on a web forum said that Kingston
makes their own memory. Sometimes I hate the internet.

Kingston buys whole finished silicon wafers and slices and
dices them on their own. I thought all their memory modules
were made that way, but I recently learned of this XbitLabs
review of some 2133 MHz Kingstons made from Hynix H9 speed
chips (1333 MHz):

http://www.xbitlabs.com/articles/mem...b_5.html#sect0

Companies that make memory modules but not memory chips tend
to have lower quality standards than the chip makers do, as
evidenced by their testing machines costing only thousands of
$$$ rather than millions of $$$.

On Friday, January 11, 2013 9:08:59 AM UTC-7, VanguardLH wrote:

I've seen some memory brands that overclock their modules
(by the spec they sell at). You are led to believe the
memory module is rated for those specs but those are
actually overclocked specs advertized for the product so
they can charge you more. When they claim the unit has
been tested at the overclocked values, that's not true.
They have not been "burned" for hours to test stability
nor has every unit been tested but just a sample.

http://www.kingston.com/dataSheets/K...1D3K4_16GX.pdf

"Each module kit has been tested to run at DDR3-2133 at a
low latency timing of 11-12-11 at 1.65V. The SPDs are
programmed to JEDEC standard latency DDR3-1333 timing of
9-9-9 at 1.5V."

Sounds fishy to me. They're saying that the modules have
been programmed with an SPD that has the memory run at a
slower speed hence the tighter timings yet they "test"
then to run at the higher speed (and looser timings).
Sure sounds like they've taken DDR3-1333 modules and
are selling them as overclocked units. The chip maker
says the product is rated for 1333 at CAS9 but Kingston
is overclocking that spec and claiming (and hoping) the
product works at 2133 CAS11.

I was told, in a message thread with my name in it, that
such overclocked modules are reliable enough for 99% of the customers and that factory testing done at room temperature
with just PCs or $2,000 testers is just as good as testing
done with the $4.5M machines used by chip manufacturers.
OTOH in other forums where most of the participants were engineers, almost everybody completely disagreed, sometimes angrily.