i7 2600K random BSOD problem solved

joemax

一般般會員
已加入
10/6/06
訊息
184
互動分數
0
點數
16
This is the english version of the same as
The original chinese version here

The issue with random BSOD happening to some overclocked Sandy Bridge
computers has bothered quite a few users. My own i7 2600K can be
pushed to clock at 5GHz in the P8P67 board and would pass superPI and
Intel Burn Test, and yet it could not survive a day without getting an
occasional Blue Screen Of Death. And this random BSOD would happen under
light load when surfing the net, running Media Player, or just editing.

After a few months of leisurely investigation, a recent occassion had
me heading down a rather unorthodox path. I am pretty sure I have
found a cure to overcome this instability problem, and you need to be
prepared that my solution leans mostly towards working on the enclosure,
the chassis. While it may look elaborate at first, I can assure you that
it is easy to understand and really is not difficult to do.
And it is highly rewarding. Highly. Rewarding.

I got the idea to this solution after I received my B3 version motherboard
in exchange for the faulted B2. That's early April this year (2011).
I did open air test at first and everything went well until the moment
I installed it back in the chassis.
The random BSOD returned. Problem remained.

Then it dawned on me that the key might have to do with the way the
computer is assembled. I might be looking at a grounding problem.
So I made some changes to how the computer is assembled inside the chassis.
And it worked.


There are four parts to this BSOD terminator project:

1) Apply the "single point ground" technique.
2) Use dynamic VID to set Vcore. Make use of Vdroop. Overclock away.
3) Ground the CPU heatsink to the motherboard ground plane.
4) Use a power line filter.



These four parts are listed in descending order of effectiveness.
The first item of single point ground is the most important one.
But each of the four has a role and I had all four of them implemented
before achieving my ultimate goal of minimal voltage, high overclock,
sleep awaken fine, and no BSOD, no lockup, no spontaneous reset.

Picture here shows my success system running LinX with AVX
together with the video card stress program of MSI Kombustor.
CPU, RAM, and video card are all seriously overclocked while
power management and thermal management are active in the BIOS
and in the OS.

i83411_SBjoemax01.jpg





(PART 1)
SINGLE POINT GROUNDING


Following the scheme of single point grounding, I picked the
power supply unit (PSU) to be the only component in
the chassis allowed to make electrical contact with the metal chassis.
All other components are prevented from touching the chassis
by lining with plastic tape and by mounting with screws fitted with
insulating washers and tubings.

The standard way of mounting the PSU with four screws provides its
ground contact to the chassis. No alteration here.

All other components aside from the PSU require isolation from the
chassis. The material used for my isolation mounting is gathered
in this picture below. The amber color tape is Polyimide or
Kapton tape. It is puncture and abrasion resistant. It is thin enough
to be set into tight spacings and will not make mounting any
component harder than usual. 3M makes Scotch vinyl electrical tape
with an adhesive layer that does not degrade to gooey gunk with time.
That's also good.

i83412_SBjoemax02.jpg



Shown above on the left is a demonstration of isolation mount of a dummy
board with two insulating pieces of fiber washer added to the standard
set of mounting screw and hex standoff. The motherboard is insulated from
the chassis in this manner.

The normal 8 or 9 pieces of hex standoffs that go on the motherboard
support tray is filed shorter. Their length in the spacer portion are
reduced by 1 mm. A piece of 1-mm thick insulating washer is set on the
shortened standoff and glued down. The insulated standoff will then
maintain the normal spacing of the motherboard atop its support tray.

i83413_SBjoemax03.jpg



Screws used for mounting other components require an insulating tubing
in addition to fiber washers. As shown by the screw on the left of the
picture below, I inserted a short section of black heatshrink tube for
use as the insulating tubing. This plastic tubing will prevent the screw
from touching the rim of screw hole openings in a frame or a bracket.
If any clearance hole is too tight for an insulated screw to pass through
with wiggle room, one would have to either widen the hole, or file off
the thread crest on section of the screws.

i83414_SBjoemax04.jpg



Some chassis designs feature screw-less mount for drives.
They provide rails for the drives. In a rail-mount design, the plastic
guide that attaches on the drive are insulating and works perfect as
isolation mount. The picture below shows the plastic guide that
came with my chassis.

i83415_SBjoemax05.jpg



Video card and other add-on cards that use an "L" bracket requires
insulation at the back frame. All edges along the opening of the
back frame are covered with a layer of kapton tape. I also apply a
layer to the L bracket. This gives me two assuring layers of insulation
at all places where the bracket meets the back frame. The screw for
anchoring the bracket has the fiber washer plus the insulating tubing.

i83416_SBjoemax06.jpg



The picture below shows insulation tape applied to the back frame.
Also shown is the taping all around the back I/O panel opening.
We need to break the contact between the back I/O panel and the chassis
such that the motherboard does not get connected to the chassis through
the I/O panel. The tape is thin enough that the snap-on
action of securing the I/O panel is not hindered.

i83417_SBjoemax07.jpg



If any of the DVI or HDMI socket and connector gets to be too close to
the frame, one may need to patch spots where mated connectors
may have their metal jacket making contact with the frame. The DVI metal
jackets are ground shields that must be isolated from the chassis as well.
Keep in mind that we adhere strictly to the "single point ground" rule.

It is best if one uses a multimeter to check while building the system.
Check the motherboard for isolation mount. Cables detached for this test.
Should get infinite resistance between the chassis and the board's ground.

And then connect the ATX12V supply cable and check that the resistance
goes to zero. We have here a motherboard-to-chassis ground connection
that's established through the supply cable and through the PSU
contacting the chassis.

Check with the multimeter on other components as they are installed.
Infinite resistance from their metal lid to the chassis without their
supply cable plugged in. And zero resistance after plugged in.

Remove the main power cord when doing this test. Safety first!! ;director;

Before we get to the CPU heatsink grounding, let's go over
the EFI BIOS setting of the Vcore voltage for "power miser" overclocking.



(PART 2)
SETTING THE VCORE


We are not doing Extreme overclocking here.
While we are pursuing the highest clock speed, we do insist on low power
consumption and low heat as well. Therefore we should make good use of
dynamic Vcore that changes with speed, and Load Line Calibration
at a level that would let Vcore voltage droops with loading.

Load Line Calibration LLC is set to Medium, or 25% level in the Asus BIOS.
A lower level than that would have big droop and would give up stability.
A higher level would have too much heat from the CPU under heavy loading.

i83418_SBjoemax08.jpg



And don't set the Vcore to a fix voltage. Use dynamic Vcore. Use
the +offset to set Vcore. This sets up a dynamic VID that boosts
the Vcore on the fly with CPU clock speed. We get lower power consumption
this way. And we are not giving up overclocking at all. Thanks to
a stabilized system with suppressed ground loop interference.

i83419_SBjoemax09.jpg



In some advanced BIOS design, a two tier dynamic VID is available.
This allows one to dispense a minimal amount in the +Offset setting,
and to provide the bulk of extra juice with the "Additional Turbo Voltage"
setting. Use this setting to provide the required higher voltage for
TurboBoost to go for heavy overclocking. I am seeing this two tier VID
implemented in Asus and Asrock boards.

i83420_SBjoemax10.jpg



For my setting of TurboBoost for all-cores to 4.7GHz (47X BCLK),
the +Offset is 0.015V and the Additional Turbo Voltage is 0.048V.
For TurboBoost to 4.8GHz (48X BCLK) all-cores, the +Offset is 0.020V
and Additional Turbo Voltage is 0.088V.
For 4.9GHz, +Offset is 0.025V, and the Additional Turbo Voltage is set
to 0.120V.

Since my 4.8GHz resulted in a core temperature of 88 degrees C
running LinX with room temp 24C, 4.8GHz and above is not for me to keep.
We do want the Sleep and Awaken functionality.
So "Enable PLL Overvoltage" is Disabled. Else Vcore can be lower still.
(still waiting for Intel to resolve this).

And power saving features in C3, C6, and EIST, are enabled in the BIOS.

Note the short term and long term power limits of 165 and 185 watts,
and the primary plane current limit of 150 amperes. Good for 4.7GHz.
If you find TurboBoost down clocking to 4.4GHz or less during any stress
test, you need those safety limits set higher. AIDA64 can help here.



Below, TurboBoost to 4.7GHz, single core SuperPi got 7.940 seconds to
complete 1M calculation. And 7 minutes 10.188 secs to complete 32M.
I use this to check for light load stability at high clock speed.

i83422_SBjoemax11.jpg




i83423_SBjoemax12.jpg



Below, TurboBoost to 4.8Ghz. Use 4 threads. Run LinX with the new Linpack
that applies AVX instruction set under WIN7 SP1. We hit 132 GFlops.
One hundred and thirty Giga floating operations per second with one chip.

Awesome. Just awesome. ;em03;


i83424_SBjoemax16.jpg



I started AIDA64 monitoring around the cycle time # 17 and the GFlops
droped a little. Note that power dissipation reached a maximum of
165 watts. (Set power limit in BIOS higher than this to avoid throttling
down.) The maximum temperature recorded 88 degree C.


All right. So let's get back to the construction project.



(PART 3)
GROUNDing THE CPU HEATSINK


The picture below shows a self-built grounding wire for the CPU heatsink.
We are going to connect the heatsink to the ground plane of the
motherboard with this wire. This 16 gauge (AWG16) wire is 6 inches long.

i83425_groundwire01.jpg


A pair of lugs are crimpped on to the two ends of the wire. One lug would
be attached to the spring-loaded screw that came with the heatsink.
On the other end, the second lug is for connection to the ground plane of
the motherboard. We need to use the insulated screw for connecting to the
motherboard ground. The screw needs to have both fiber washer plus
heatshrink tubing insulation. The washer and tubing would insulate
the lug from the screw.



This below shows where the motherboard ground plane connection will be
made. The mounting hole next to the ATX12V supply socket is picked for
grounding the CPU heatsink. The red fiber washer and the standoff behind
the motherboard are in place.

i83426_groundwire02.jpg



One end of the grounding wire is now attached to the motherboard with the
insulated screw. The lug makes contact with the solder bumps surronding
the hole and therefore gets connected to the ground plane of the
motherboard. At the same time, the washer and tubing prevent the lug
from touching the screw. And therefore the wire is not connected to the
chassis through this screw.

i83427_groundwire03.jpg



The other end of the wire is connected to the CPU heatsink.
The lug at this end can be seen as being firmly pressed down by the
spring-loaded screw. Note that one should pick a larger lug with
a loose fit in the screw pass the threaded section.
We want the spring to press on the lug which in turn would press
on the bracket that forces the heatsink on the CPU.
Never let the lug get stuck in the threaded section of the spring-loaded
screw. Letting this happen would ruin the pressure balance of the
heatsink assembly.

i83428_groundwire04.jpg



After we connect the heatsink to the ground plane using our specially
made grounding wire, let's test again with a multimeter. We should see
that the resistance between the heatsink and one of the tin cans of the
back I/O sockets to be zero. The tin cans are soldered to the
ground plane. They serve as a convenient spot to probe the ground plane.

Also, use the meter and test the path between the mounted CPU heatsink
and the chassis. With none of the supply cables connected to the
motherboard (remove them if cables are connected), and that the job of
Part 1 of "single point ground" is properly performed, the meter should
read infinitely high resistance indicating the path is indeed insulated.

However, on the other hand, if a low resistance is measured in between
the CPU heatsink and the chassis, then the reinforced backplate support
of the heatsink might be touching the tray behind the back of the
motherboard. Mine did just that. Time to use the roll of tape again.
Just apply a wide area of tape on the tray to block the contact.

If there is a big cut-out hole factory made to provide an access
port on the tray, and that it can clear the reinforced backplate of
the heatsink, we won't need any tape on the tray.

And if the chassis is all factory painted on the inside (black is the
fashion), we may not need to tape those frames and drive bays since paint
is reasonably insulating by itself. The insulated screws are still needed
though. I won't count on the paint not being chewed off as screws go
past it.



It's no small feat that you have read so much and so far down this
article. You should pat yourself on the back just for this reading
achievement alone.
;cheer2;


We are down to the last part on the list.




(PART 4)
POWER LINE FILTER



The power line filter should be really less of a concern for most people.
With the first three parts in place, the goal is just about 100 percent
achieved and the overclocked Sandy Bridge is 100 percent stable.
But just for presenting my story in its entirety, let me show you the
power line filter I use.


i83429_islatrol01.jpg



This is the filter brick that I plug my i7 2600K computer into.
Its job is to stop noise and transients on the power line from getting
into the PSU and hence the computer. Specialty filter brick like this
Islatrol unit is capable of filtering out the so-called common-mode noise
as well as the normal-mode noise. It does this far better than the
X and Y capacitors built into a PSU.


So how does one diagnose if the condition of the power line is affecting
the i7 2600K stable operation.

Simple. First run your overclocked system "by itself".
"By itself" means not connecting any external gadget to the computer.
We make an exception here for the monitor. One needs that connected to
see a BSOD.

No ethernet wire connection (wireless LAN is fine).
No powered speaker connection (non-powered headset is fine).
No eSata, NAS. No USB connection to powered equipments.
Just tune and test the isolated computer till it is stabilized.

Then start connecting externally.
If it then run into random BSOD, or freeze screen, or a LinX error,
the condition of the house power line need to be considered.
May be it is the ground wire routing in the power outlets.
May be a good power line filter will help.

The current model Islatrol LRIC+ 105 (made by Control Concepts) is in
my mind best for this job. It is even better than a UPS since it is
light and small and we won't hesitate taking light and small
to LAN parties.

My home wiring is a big fail as far as computer stability is concerned.
There is no ground wire in the wall outlet. The LAN router is a room
away, and my CAT-5 cable is 100 feet long. The perfectly stable
"by itself" Sandy Bridge would soon turn unstable when the
LAN cable is plugged in. No, it's not IE9's fault. It's the power line
providing power to a system that has two connected units seperated by a
long distance.

So I added this power line filter to my Sandy as a gift in return for
so much fun she has provided. With all four elements of BSOD terminator
coming together, and the fifth element being me TurboBoosting at 4.7GHz,
there has not been a single instance of BSOD, or lockup, or random reboot
for the past two months. None. As in terminated.... ;tongue;

Here's for entertainment a video showing how the i7 2600K got in and
out of stability by me switching between mounting screws with and without
insulating washers. It's played back at 2X speed.


Video showing my i7 2600K getting in and out of stability with mounting screw changes


My rig:

Chassis: Superchannel G90 9002
the chassis at Superchannel
Motherboard: Asus P8P67 B3 EFI BIOS 1305
CPU: i7 2600K batch# L042A969
Heatsink: Thermalright Ultra-120 Extreme
TIM compound: Antec Formula 7 Nano Diamond
Heatsink fan: ADDA AD1212UB-A7BGL 12cm 2500rpm max PWM
Video card: Gigabyte HD6850
DRAM: G.Skill F3-12800CL7-2GBRM x2
PSU: Seasonic X-560
C: Intel X25-V G2 40GB
D: WD 2500AAKS
E: iCute iSwap 203
G: Intel 320 G3 80GB
R: Toshiba-Samsung CDDVDW SN-S083B
Monitor: Samsung F2380
Power line filter: Control Concepts Islatrol LRI-105
Speakers: Creative Labs SBS 2.1 370
 
▌延伸閱讀