Overclocking Bible: Post your comments/questions here!

[MrSeanKon]

This program is based on Mathematical algorithms and it will be useful if you follow some instructions
It is something like calculator.
For example it predicts the next Maximum Stable Frequency of your CPU, RAM, chipset you can optimize your memory timmings, calculate the recommended PSU etc...
AMD and Intel users can use it.

P.S.
1) You must install Microsoft Framework otherwise program cannot run.
2) Programs are freeware (forever)

Post here questions for OcBible programming

Questions for CPU Stretcher

Questions for A64 suggester

[MrSeanKon]

Here is an explanation of BIOS beeps..

AMI BIOS:


Award BIOS:


Phoenix BIOS:

[MrSeanKon]

How we can turn on our PSU without connecting to mainboard?

First we need a semi naked wire:



So connect this wire between green and black ATX hole.



Tip:
This process is very useful in case you need to check if your PSU is dead.

[MrSeanKon]

First you need a free PSU molex (look this photo).



The Red probe of multimeter measures the positive voltage and the black the negative.
So connect these probes between yellow cable (+12V) and black cable (ground --> zero voltage)



Similar measure +5V (red - black):

[MrSeanKon]

This photo shows the voltage between red and yellow.
Note I put probes in inverse order that's why you see a negative value.



Here is 6pin molex.
Orange cable is +3.3Volts.
So orange-black gives you the answer "How is my +3.3 PSU voltage??"

[MrSeanKon]

And you want 3 digits precision?
Mutimeters measure voltages in [0, 2] area with 3 digits.
So if I want 3 digits for +5 PSU voltage?

Use a naked wire like before.
Then put it on bottom of a +1.5V battery.



Measure battery first:

[MrSeanKon]

Measure orange-battery positive pole



It was 1.961 Volts.
So +3.3V is 1.450+1.961=3.411Volts

Measure red-orange:



Thus +5V is 3.411+1.769=5.180 Volts!

Note:
This is not necessary if you have a high precision multimeter or an oscillator.
Both of them are expensive..

[MrSeanKon]

Left of CPU socket located some black chips known as MOSFETs.
Each of them has three pins.
The first is positive Vcore.
So put red probe on it warning your hand must be steady otherwise your mainboard will burn!



The black probe of multimeter we put on PC chassis for reference ground.
The value was 1.756V.

If we put the red probe on middle pin we measure the negative voltage.
The value was 0.012V.



Thus my Vcore is 1.756-0.012=1.744.

Tip:
If we cannot put probe on a MOSFET we can measure Vcore on coil.
Nodes are located on the back of mainboard.

[MrSeanKon]

Similarly we can measure Vdimm if we find the proper MOSFET around memory slots.
Of course on the back of mainboard like coil we can measure Vdimm too.



Tip:
You have to seek a bit to find the proper MOSFETs.
This photo shows my mainboard's chipset (gold heatsink) but the shown MOSFET does not give Vdd.



The MOSFET is here (it be shown with a blue tube)



White arrow shows a MOSFET where I measured Vcc_HT.
Note AGP slot is green.

[mike]

Awesome Post! Welcome to the board!

[Skinny13]

excellent thread mr sean

sticky vote from me since I haven't seen a thread like this in xs

[MrSeanKon]

Thank you mikeguava!
Although I am AMD user my program (OK some features do not work be patient) it works for Intel processors too.
Well sometimes a load such as an old hard disk or a fan is required for the PSU to boot for the method discussed above.

[gfa3]

stickey vote from me too.very usefull thread i've seen things i always wanted to do but never had a reference image or guide with fotos like these

[MrSeanKon]

System was:
CPU: AMD Barton 2800+.
Multiplier: 12.5X (locked CPU).
Mainboard: ABIT NF7-S v2.0 nForce2 chipset.
BIOS version: Normal (not modded).
Memory: 1 X 512 MB TwinMOS PC3200 CL2.5.
Timings: 2.5-4-4-8 (CL3.0 did not boot) divider was 50%.
Vdimm: 2.984 Volts cos I want 100% stable RAM.
PSU: Antec True Power 430 Watts.
Cooling: Air cooled by SLK 947U.
Voltages measured with a multimeter (at full load).

The following table shows all necessary data:



Step 1:
Set voltage around default (factory) value.
All Bartons have Vcore=1.650 so the closest setting in BIOS was 1.65.
Note this mainboard undervoltage Vcore a bit (1.639 < 1.65).

Step 2:
AMD launched the highest Barton 3200+ (2.2 GHz real speed).
Thus, it is very possible for this processor to reach this speed if cooling is good (not necessary watercooling).
2200 --> 12.5 X 176 --> I set 176 FSB in BIOS.
I do not use any programs for overclocking via Windows (on the fly).
Tip:
If cooling was something extreme e.g. Prometeia, Vapochill, phase change then user must seek for first MSF in an upper area.
For temperatures around -15C first MSF is almost 160 MHz more (2200 + 160 = 2360 MHz) and around -30C first MSF is 200 – 220 MHz more.
This analysis works fine for any AMD up to A64 754.
I reckon it works fine for 939s too (I have not tested A64 939).

Step three:
nForce2 chipset runs OK (nVidia specifications) at least at 200 MHz.
This means no worry if CPU runs < 12.5 X 200 = 2.5 GHz.

Step four:
Set divider and relax memory timings.
Of course some combinations maybe not work.
This point requires experience and you have to check some combinations.
Sometimes system hangs during startup and you hear beep codes.

Step five:
Choose a CPU stability program.
For example S&M quick test, Prime95 (up to ten minutes) or SuperPi 4M.
It is very important to understand that you must not spend much time at this step.
This method called “Mapping method” and you save time!
Mapping method works fine for any system (AMD/Intel).

Step six:
I started 12.5 X 176 MHz and the stability program was SuperPi 4M.
Of course, system passes this test.
Reboot and set 12.5 X 177 = 2212 MHz (passed).
12.5 X 178, 12.5 X 179, 12.5 X 180, 12.5 X 181, 12.5 X 182 passed.
Reboot and set 12.5 X 183 = 2288 MHz SuperPi 4M failed.
Thus, MSF is 2278 MHz or 12.5 X 182 as table shows.
This is the most important step for “Mapping method”.
As you can see, I tried eight combinations (176, 177, 178, 179, 180, 181, 182 and 183) for FSB and step was 1 MHz.
SuperPi 4M runs up to four minutes for Barton systems.
Therefore, I spent 4 * 8 = 32 minutes only (Jesus what a waste time)!

[MrSeanKon]

Step seven:
If your AMD is different from my samples you must find the second MSF like above.
In addition, you must run General CPU speed estimator.
Sorry I cannot buy any AMD CPU….
However, you can save more time at this point if you use your experience and O/c Analyzer find frequency FSB.
Set n = 1 (linear), 0.75, 0.5 (square) and 0.333 (cubic) and compute estimated MSFs:






Reboot system set Vcore = 1.693 (1.700 in BIOS) and 12.5 X 188 = 2354 MHz --> SuperPi 4M failed.
Reboot (Vcore = 1.693 always) and 12.5 X 187 = 2335 MHz --> SuperPi 4M failed.
Reboot, 12.5 X 185 = 2316 MHz --> SuperPi 4M passed!
It 100% clear which is the next MSF!
Reboot 12.5 X 186 (last chance) --> SuperPi 4M passed!
Thus MSF2 = 2327 MHz!
In general, FSB is not integer (2327 = 12.5 X 186.16).
Therefore, I spent 4 * 4 = 16 minutes only (God what a waste time)!

[MrSeanKon]

Step eight:
Now we are ready to run General CPU speed estimator.
We enter data as photo shows:



Status bar prints a warning.
Therefore, you must not enter higher voltage from 1.798.
Reboot system set Vcore = 1.798 (1.800 in BIOS) and 2450 = 12.5 X 196.
2450 MHz is the upper limit so at this point we start seeking from upper limit.
SuperPi 4M failed of course.
Reboot system (Vcore = 1.798 always) and 12.5 X 195 --> SuperPi 4M failed.
Reboot 12.5 X 194 --> SuperPi 4M failed.
Reboot 12.5 X 193 --> SuperPi 4M failed (damn)…
Reboot 12.5 X 192 --> SuperPi 4M passed! --> MSF3 = 2.4 GHz!
Remember that C.C. is always 100% at this point.

Step nine:
I continued like above.
If Vcore > +15% of default (factory value) then CPU work in breakdown region.
In general, it is difficult to estimate next MSFs because of exponent n changes rapidly.
1.899 is 1.899 / 1.650 --> 1.1509 --> 15.09% higher from default Barton Vcore.
Anyway, let’s try!
Nevertheless, it is very important to see what is going on.
O/c Analyzer find exponents give us many info!
We enter three pairs first:



Results are here:



As you can see, n was 0.657 and became 0.568.
This very normal because increase Vcore --> exponent n decreases.
Thus, MSF1 --> MSF4 has exponent up to 0.568!
Let’s run O/c Analyzer find frequency FSB again:



Estimated MSF is 2477 MHz.

[MrSeanKon]

Let’s run General CPU speed estimator now:



Well upper limit is 2508 MHz and lower limit is 2449 MHz.
O/c Analyzer find frequency FSB predicted 2477 MHz so it is very possible 2449 < MSF4 < 2477 MHz.
LOL BAN [2477, 2508] area to save time!
It is impossible from MSF1 --> MSF4 n > 0.568 I have tested many CPUs trust me but I am not a magician of course!
We expect n < 0.568 but we do not know the exact value.
Many phenomena follow Gaussian distribution so it is not wrong if we apply these ideas.
By the way, we cannot express this complex phenomenon using only Mathematics needs test.
Gaussian distribution determinates mean value which is (a + b) / 2.
(2449 + 2477) / 2 --> 2463 MHz.
Reboot system, Vcore = 1.899 speed 12.5 X 197 = 2463 MHz --> SuperPi 4M failed.
Reboot system, Vcore = 1.899 speed 12.5 X 196 = 2450 MHz --> SuperPi 4M passed --> MSF4 = 2450 MHz!
Two combinations only! Whoopee!

Step ten:
Next Vcore (see table above) was 2.020 Volts a very high voltage.
It is 2.02 / 1.65 --> +22.42% higher from default Vcore.
In this case n <= 0.333 (cubic) in many cases.
Let’s run O/c Analyzer exponents:




O/c Analyzer find frequency FSB:



Reboot system, Vcore = 2.020 (2.100 in BIOS --> undervoltage) speed 12.5 X 200 = 2504 MHz --> SuperPi 4M failed.
Reboot system, Vcore = 2.020 speed 12.5 X 199 = 2488 MHz --> SuperPi 4M failed.
Reboot system, Vcore = 2.020 speed 12.5 X 198 = 2475 MHz --> SuperPi 4M passed.
Nevertheless, as I said above in many cases FSB is a float number --> MSF5 --> 12.5 X 198 = 2475 MHz (in BIOS) --> 2483 MHz CPUz showed.

This is the end!
We finish buddies!
As you can understand I spend less than 2 hours!

Step eleven:
Now we choose a Vcore for permanent usage.
It depends on you (see polls ask your friends etc)…
Assume I choose Vcore3 = 1.798 Volts.
MSF3 was 2401 MHz but stress time was small too.
OK at this point I have to run a stability tester for more time.
I did not set 12.5 X 192 = 2.4 GHz.
It is obvious MSF3 computed for a small period.
In general final MSF (long period stretching) is 2 - 3 MHz lower.
I checked this system and it failed after 5 hours at 12.5 X 190 but at 12.5 X 189 worked fine (24 hours Prime95).
“Mapping method” is the basis of O/c Bible.
You save money and time of course.

[MrSeanKon]

Someone could ask:
Why do not use General CPU speed estimator at step ten?
Is it wrong?
The answer has relation with extrapolation and divergence.
Unfortunately, model function is not smooth.
If you compute differentials you see they change.
Similar computations performed by O/c Analyzer exponents.
Thus if % loss changes rapidly any numerical method is not accurate.
In this case, you must use piecewise functions or “Pade” type models but it is too hard to explain it here.
On the other hand, it is stupid for everyone to spend time in order computation error become small.
Let’s input the first 3 pairs:



Status bar prints warning message again.
Note that second model has the biggest C.C.
Really lower value of second model (2507 MHz) is nearer to result (2483 MHz) from others.
LOL let’s BAN first pair:



Again second model has the biggest C.C. etc….

LOL let’s BAN second pair:



Although status bar prints warning second model lower limit is near to MSF5.
Now let’s compute exponents:




Note MSF1 --> MSF3 loss = 13.6%, MSF1 --> MSF4 loss = 23.4% --> loss increase = 23.4 - 13.6 = 9.8%.
MSF1 --> MSF5 loss = 37.3 --> loss increase = 37.3 - 23.4 = 13.9% > 9.8%.

[MrSeanKon]

Of course we can compute partial exponents cos until now we have computed total n.
I mean exponents from MSF2 --> MSF3, MSF3 --> MSF4, MSF4 --> MSF5, MSF3 --> MSF5.

MSF3 --> MSF4 exponent is:



MSF4 --> MSF5 n is:



MSF3 --> MSF5 n is:



Here % loss is high.
That's the reason why we must BAN LOL some initial pairs in order computation error be as small as possible.

Conclusion
1) Do not increase voltage too much.
Step should be smaller than 6% of default Vcore.
Thus for A64 754 socket step should be 1.5 * 0.06 = 0.09V.
Which means if Vcore1= 1.489Volts the next Vcore2 should be 1.489 < Vcore2 < 1.579 = 1.489+0.09.
2) It is high recommended to seek first MSF setting Vcore1 around default Vcore.
As you saw table shows that Vcore1 was 1.639 (almost 1.650 --> Barton default Vcore).
3) RAM + chipset must be 100% stable otherwise all estimators fail.

[MrSeanKon]

what is this error?
clr error: 80004005
the program will now terminate

aja you must install the Microsoft Framework v1.1 (look first post for link).
OK read instructions sorry for my English it is not my first language.
We are planning for a web page in the future me and other members want to be this program useful for all o/cers around the world!

[aja]

aja you must install the Microsoft Framework v1.1 (look first post for link).
OK read instructions sorry for my English it is not my first language.
We are planning for a web page in the future me and other members want to be this program useful for all o/cers around the world!

this problem arose after i installed the framework 1.1 thing.....

but it';s working now, for some reason it was set to windows95 compatability mode

but disabled it and now it works fine

[Athens[2004]]

awesome Sir ,.

[rahssan]

Nice dude, nice, keep up the good work!

[MrSeanKon]

Thanks Athens2004 + rahssan!

The user aja means this:



The solution is simple:
Right click on Oc_Bible executable icon.
Then choose properties in Tab box and uncheck this option (see mouse arrow):



Program runs OK at Windows 2000 (if your system has the requirements --> see first post) but I do not care if it runs OK at older Windows versions.

[Overconfidence]

Damn, and I was making a site called "The Overclocker's Bible" =P.

I trust I'll still be able to use the name, though. >_>; Or not...

Awesome proggy though.