CPU & RAM burn-in guide

[MrSeanKon]

Here are CPU & RAM stretcher


Although I have opened a thread for supporting O/c Bible I don't want to be many pages long...
There I add/remove/edit my posts cos maintaining up to date but your feedback is appreciated.
In this thread I will describe step by step how I find MSFs using O/c Bible and my burn-in guide and more stuff.
Let's start!

System specifications:
CPU: A64 3200+ Venice core.
Default Vcore: 1.350 Volts (other pages say 1.400 anyway).
Default multiplier: 10X.
RAM: 1X512 MB Kingston HyperX PC4300.
PSU: Enermax 465 Watts.
Cooling: Hand made watercooling.
Mainboard: MSI K8N Neo2 nForce3 250 chipset.
O/S: Windows XP with SP1.
Burn-in program CPU stretcher (included in O/c Bible's package).

Step 1 --> Choosing multiplier
Usually Venices achieve around 2.7GHz or higher (OK it depends on stepping/cooling/Vcore etc)...
2700 MHz = 9 * 300 MHz --> my mainboard could not POST with FSB=300 MHz (I tried all HT settings).
So I choosed the default multiplier.

Step 2 --> Choosing BIOS settings
Hyper Transport speed should be around 1 GHz.
270 MHz * 4X --> 1080 MHz > 1 GHz --> I set HT=3X in BIOS.
The closest voltage in BIOS which gives default Vcore for my CPU is Vcore=1.350.
I measure Vcore with a multimeter (see O/c Bible's tutorials how if you don't know) and Vcore was 1.361 Volts.
As you can understand this mainboard overvoltages (a bit).
In addition I set relaxed timings in BIOS --> 3-4-4-8/2T and divider (RAM / FSB = 133 / 200).
I repeat here I must be 100% that my RAM is 100% stable at any FSB.

Some notes
1) RAM is stable around 310 FSB using 3-4-4-8/2T, Vdimm=2.82 (I tested on a previous Gigabyte K8NS Pro 754 socket).
2) RAM is one years old (at this moment) so it is not recommended to apply RAM burn in process.
3) My mainboard (visit MSI forum for more info) has a bug:
SATA1 + SATA2 are not locked.
Thus I could not enter in Windows using FSB > 239 MHz.
Moving my hard disk to SATA3 + banned SATA1+SATA2 in BIOS all are OK!
4) We must measure all voltages at full load and see if they are stable.
5) I put some extra small heatsinks on CPU MOSFETs cos they are hot!
6) I measure voltages at coil node.
7) We apply CPU burn in process when our CPU is new not after 2 months!
8) When achieve a MSF we don't change settings --> All change and not CPU stable settings makes our CPU a bad stepping! --> consider the proverb "All work and not play makes Jack a dull boy"..
9) This process is similar any CPU you may have.
See here for Intel burn-in (thread at XtremeSystems also)...

[MrSeanKon]

Step 3 --> Finding the first MSF
Of course I don't know anything for my CPU...
At this moment the highest 939 core is 2.4GHz so it is possible for my CPU to achieve 2.4GHz with default voltage.
Some 939 CPUs achieved higher speeds with default voltage; anyway I must started from a point.
Setting 10 X 240 in BIOS system could not enter in Windows...
This means that at 2.4GHz is 100% unstable....
I reduced FSB 5MHz and I try to run 8M SuperPi (see O/c Bible's thread for Mapping method).
Sometimes I use S&M (quick FPU test 75% CPU utilization) for Mapping method.
System crashed immediately.
"Xm MSF1 should be lower around 232 FSB" I thought....
Restart 10 X 232 and here is the result:



"Damn at 2310 MHz my CPU must pass 8M" I thought...
Really here is the photo:



Thus MSF1=2310 MHz and Vcore1=1.361 Volts.

[MrSeanKon]

Step 4 --> Finding the second MSF
Current version of O/c Bible has not an AMD 939 CPU speed estimator (desktop CPUs)..
On the other hand I cannot buy any AMD CPU...
Moreover each estimator needs time for developmenting equations etc...
That's why O/c Bible's package includes General CPU speed estimator --> a program for any CPU (AMD/Intel/desktop/mobile).

I increased Vcore (multimeter showed 1.413V)..

But you can cheat O/c Bible!
How??????????????
Well I am the creator --> of course I know anything for this program!
See this photo:



Choose AMD 754 socket cos it is the closest CPU to 939 (I reckon this is simple)...
Pressing Show Card2 button (the black calculator) combobox in Card2 has values from 1.45 up to 1.600 Volts...


The following photo shows that I edited combo box value (it was 1.45 and I put 1.361=Vcore1).
Pressing Show Card3 button I can choose in Group's 3 combo box my desired voltages!



So I put Vcore2=1.413 and pressing Calculate button (the laptop icon in Button panel group) O/c Bible shows that 2357 < MSF2 < 2397 MHz (6th region the most possible).



I tried 10 X 240 = 2400 MHz (almost 2397) and system failed SuperPi 8M.
Reboot 10 X 239 --> system passed!
Jesus what a waste time!
Thus MSF2=2390 MHz and Vcore2=1.413!

[MrSeanKon]

Step 5 --> Finding the third MSF
First of all we must see what is going on:
Let's analyze current data (O/c analyzer exponents):



Exponent n (only one cos one data pair) is:



This value is very good --> almost linear progress (see O/c Bible's FAQ)...
I increased to Vcore3=1.463 Volts (1.450 in BIOS).
As you can see voltage increases are small not abnormal increase is not recommended cos of it is too difficult to predict "very far" what is going on...
Normally (see OcForums O/c Bible's supported thread --> "Analyzing o/c in depth") the new n should be up to 0.908.
So let's estimate MSF3 using O/c analyzer find frequency/FSB:



But we have two pairs --> we can run General CPU speed estimator!
Cos of two pairs C.C. is always 100%.



O/c analyzer find frequency/FSB estimates MSF3=2467 MHz --> 10 X 247 = 2470 MHz.
This means that MSF3 should be smaller or equal to 2470 MHz.
General CPU speed estimator predicts from 2444 MHz (the smallest value) up to 2490 MHz (the highest value).
This means that 2440 < MSF3 < 2490 MHz.
The common range is (apply BAN method see O/c Bible's tutorials) --> [2440, 2470].
Reboot at 10 X 247 MHz --> system passed SuperPi 8M.
Reboot at 10 X 248 MHz --> system crashed SuperPi 8M.
Thus MSF3=2470 MHz, Vcore3=1.463 Volts!

[MrSeanKon]

Step 6 --> Finding the fourth MSF
We repeat the same procedure.
Analyzing current data (O/c analyzer exponents):



Exponents n1,n2 are:



Xm n2=0.927 > n1=0.908 that's why you see negative loss....
This means that this CPU loves Vcore!
But after a point n is expected to be slower and so on...

Anyway let's run General CPU speed estimator:



The Vcore4 was 1.510 Volts (1.500 in BIOS and +3.3% VID increase --> MSI settings)..
O/c analyzer find frequency/FSB:



BAN method now!
MSF4 <= 2544 (cos of O/c analyzer find frequency/FSB)...
2520 < MSF4 < 2570 MHz (cos of General CPU speed estimator --> note 2519 MHz is the smallest value almost 2520 = 10 X 252 MHz).
Thus 2520 MHz < MSF4 < 2540 MHz.
Reboot and 10 X 254 MHz --> system failed SuperPi 8M.
Reboot and 10 X 253 MHz --> system failed SuperPi 8M.
Reboot and 10 X 252 MHz --> system passed SuperPi 8M!
Thus MSF4=2520 MHz and Vcore4=1.510 Volts.

Note:
The second area of General CPU speed estimator has the smallest C.C. of all but 2519 MHz is more near to 2520 MHz (the actual MSF4).

[MrSeanKon]

Step 7 --> Finding the fifth MSF

O/c analyzer exponents:



n1,n2,n3 are:



Xm n3 < n2 and n3 < n1....
This is normal and we expect it!
Vcore5=1.562 Volts (1.500 in BIOS +5.0% VID increase).
General CPU speed estimator:



As you can see second area has the highest C.C. so we based on these results (BAN first estimated area and third)
This means that 2576 --> 10 * 258 = 2580 MHz < MSF5 < 2620 MHz.
O/c analyzer find frequency/FSB:



BAN method --> 2580 MHz < MSF5 < 2590 MHz.
Reboot 10 X 259 MHz --> system passed SuperPi 8M!
Reboot 10 X 260 MHz --> system failed SuperPi 8M...
What a waste time again!
Thus MSF5=2590 MHz and Vcore5=1.562 Volts!

[MrSeanKon]

Following the same procedure I filled this table:



Let's analyze all data:



Now CPU behaves normal for Vcores > 1.463 Volts (n becomes smaller and so on)...
Let's plot our data:



See where mouse pointer shows...
There curve is becoming hump backed more...
The total curve has banana shape!
That's why I call this "Banana convergence"...

[MrSeanKon]

Step 8 --> What it will be??
I reduce Vcore to 1.361Volts and I burned my CPU at 2330 MHz using CPU Stretcher.
You can see here at OcForums my procedure.
After that CPU can pass at 2330 MHz SuperPi 8M!
Thus +20MHz gain!
That's why I started burn-in process at 2340 MHz.
Here is a screen capture (this moment):



I wait your feedback/questions/comments/experiences.

[i found nemo]

nice program sir, gives pretty accurate mhz per v.core rating too! i reccomend this tool to someone who's new to overclocking and can fallow dirctions

[Xenom0rph]

Good job indeed, i find it very usefull. Congrats

[|3ourne]

good job !
sort of complicated but definitely a bloody nice detailed method.
I use a simpler method myself.
Step 1 : I use clockgen to find the lowest voltage the cpu can prime95 stably at stock speeds. I then let prime95 at lowest priority run for about 2/3 days continuously.

Step 2 : I then put the voltage back to default and slowly start raising the FSB. Once I find the fsb that prime95 fails at , I go down 1 mhz on fsb and run a program called Cpu Burnin for 24-48 hrs continously. And then slowly start going up again. This gives me about 5-10 more mhz at the same voltage that prime95 failed instantly.

Step 3 : Raise vcore one notch in bios and repeat the steps above. Its a tedious process but it benefits greatly for people who are on air cooling.

For eg. I couldnt prime95 stable at 2.35 ghz at stock volts when I first got my venice 3200+ , right now after burning it in , its completely stable at 2.4 ghz . My next step is to bump up the voltage in bios and start from there again. But 38C 100%load temp with the IHS on is completely worth it .

[iS.Terminator]

this is kool for overclocking noobs like me

[MrSeanKon]

nice program sir

I think you saw Athens2004's post who calls me "Sir"...

Good job indeed, i find it very usefull. Congrats

Thanks both of you for comments.
Keep on o/c!

this is tool for overclocking noobs like me

Well this program is for anyone.
Of course you should know some basics but you need time to take the most of it.
Believe me if you use this program for enough time you don't anyone; you don't need me!
You may read my FAQ (history etc)...
The main problem using this package is time; you have to be patient.

|3ourne:
Wow!
Thanks for info!
An alternative method for CPU burn-in!

[Ref]

nice program and thread, however thread title made me think of something different at first

[Skinny13]

sean
u said
1)RAM is stable around 310 FSB using 3-4-4-8/2T, Vdimm=2.82

ram is 333????

2) RAM is one years old (at this moment) so it is not recommended to apply RAM burn in process.

why so?

[MrSeanKon]

Thanks Ref well thread's title may mess someone but this is the name of this process.
Skinny13 my RAM is Kingston HyperX PC4300 (please read carefully my first post --> system specifications)..
This means it's DDR 533 not DDR 333.

Well I don't think more to burn my RAM cos it is old now.
Xm...
In addition a good for RAM burn in is MSDOS MemTest (OK it tests your RAM also)...

Note 310 MHz FSB 3-4-4-8/2T Vdimm=2.82 was using Gigabyte K8NS Pro 754 socket (I sold my 754 system yesterday and I have 939 now)...

[CD 1986]

Nice tool! I just had a go with it and it seems very accurate. I know my AMD 3200+ Venice doesn't respond well with more than 1.50v (at over 2700mhz) and using your method got me spot on results.

[MrSeanKon]

CD 1986 I saw that you have the same mainboard!
Wow!
Well do you use multimeter to measure voltages?
If no then MSI's monitor Vcore is -0.02 up to -0.03 Volts lower than actual.
Vdimms are OK Vdimm in BIOS is -0.01 lower than actual.
Well Vcore BIOS settings gives different values than excepted but MSI's monitor is enough accurate.
Well I saw your n is almost 0.5 (2540 --> 2770 MHz) so don't increase more your voltage at this moment.
Please be away to run CPU Stretcher near to unstable FSBs cos it is too dangerous (hard disk data corruptions).
Wow??????
You have stock heatsink and you achieve this speeds??
Wow you have a very good processor!
In addition keep in mind CPU Stretcher must run with a qualified PSU I don't suggest this evil program for medium quality PSUs...

[CD 1986]

Yep, same motherboard! Didn't use multi-meter i'm afraid, only MSI monitoring software; just wanted to do a quick test.

[MrSeanKon]

OK but O/c Bible's estimators are very sensitive if you enter non actual voltages then output is for garbage.
You can see how you can use a multimeter (follow the O/c Bible's supported thread here at XtremeSystems).
At least you can measure PSU's voltages easily.

Well I have to rest buddies!
Thanks all for your great comments.
I will login tomorrow!

[Ref]

Thanks Ref well thread's title may mess someone but this is the name of this process.

not that it matters but the name of the process is "burn-in"
"burn your CPU" means you put it in a fire or something
correct me if i am wrong ?

[davidw8818]

very nice guide, let me try to burn it this way, ill post bak with some results later

[MrSeanKon]

I changed thread's title as you can see.
Thanks for your suggestion (English is not my first language).

davidw8818 we are waiting your results.


Let's summarize my current CPU status:
Before burn in CPU passed 8M SuperPi at 2310 MHz, Vcore=1.361 and failed at 2320 MHz.
As you can see above I increase FSB (233 MHz) and I ran CPU Stretcher 4-6 hours each time.
I increase more FSB and I ran my system for 8.5 hours.
I went for ZZZ cos I was too tired but ZZZ=0 for my CPU and no rest!
See this photo:



Today I tried 2340 MHz 8M SuperPi and CPU passed!
Wow!



I tried 2350 MHz 8M but failed...
Note that CPUz shows variable Vcore --> multimeter is a must have tool.
Thus total gain after burn in (using Vcore=1.361) is 2340 - 2310 = 30 MHz.
I don't have enough time and patience thus I decided to stop burn in process using Vcore=1.361.
So I increased Vcore to 1.413 Volts.
My table (that's why it is very important to save data/notes) shows that MSF2=2400 MHz, Vcore2=1.413V.
Of course it is not smart to run again 8M there.
I ran 8M at 2410 MHz to see if I gain something.
Yes fellas it passed!



Therefore using Vcore=1.413 Volts current gain is 10 MHz.
I will continue the same process.
I will run my CPU at 2430 MHz for 8 hours (or more), ZZZ=0.
Next day ZZZ=0 at 2440 MHz and so on (until crash).


My recommendations
1) Don't burn too much your CPU!
Let adequate time for rest!
2) At least you must burn your CPU (each Vcore) for 24 hours.
You can split burn time (3 times 8 hours each burn or 4*6).
3) If you are patient burn it more hours e.g. 36 or 48 but I don't suggest you more (OK if you have enough free time burn it for days each step).


F.A.Q.
Q. Your CPU passed 8M at a MHz using b Vcore. This means that is rock stable.
A. Of course no!
See O/c Bible's thread for Mapping method.
To be more specific:
MSF=2340 MHz using Vcore=1.361 as you saw above.
If I decide to use this Vcore for permanent usage then I will reduce CPU speed at 2320 MHz or 2310 MHz and all are OK.
Q. What Vcore should I use?
A. Visit PcPer forums and OCAU (look my signs/threads there).
I opened two AMD Vcore polls and read posts etc...
Q. I have Intel CPU. What should I do?
A. The same process.
Keep in mind that Intel processors gain more MHz after burn in (in general).
See our chat at XtremeSystems for Intels (link is above).
But Intel processors are more sensitive than AMDs --> don't increase too much Intel Vcore...
Q. Why should I burn my CPU? What are advantages?
A. OK 30 or 50 MHz gain means nothing after burn in.
The most important is permanent Vcore.
If your CPU is stable after burn in at same MHz using -2% or more Vcore this is the difference.
CPU wattage follows the square formula --> Increasing Vcore +2% wattage becomes +4% higher, increasing Vcore +3% wattage becomes +9% (3% * 3%) and so on...
Q. Thanks Sir for this program I have achieved my target/goal. Should I use it more?
A. Yes!
Consider a bodybuilder!
Assume that your CPU is Prime95 (24 hours stable) or S&M long FPU test anyway your favourite stability program at a MHz.
Once day per week I suggest you exercise for (6-8 hours when you ZZZ or away from home) 2 - 4 MHz FSB higher (ZZZ=0-50, lowest priority-normal).
Just for keeping your gains (like bodybuilder or a 100m runner).

[Athens[2004]]

Nice work here John

How far goes your new 3200+ venice with water? My venice 3.000+ 0522 stops at 2.75 with 1.55 burn in rulez [in water , soon with subzero ]

[MrSeanKon]

As you can see my first table shows that MSF7=2650 MHz, Vcore7=1.663Volts (8M SuperPi)...
Xm let's try 32M at 2650MHz with same settings I thought..
Yeah CPU passed 32M there!



I tried 32M at 2660MHz but failed, 7 loops completed...
There no reason for worries!

What is going on with my RAM?
Well I was not 100% sure for stability but I ran 3-4-4-8/2T my RAM at FSBs <= 260MHz.
Vdimm was 2.62 Volts (Gigabyte K8NS Pro 754 socket)...
When I ran it with CL2.5 I increased Vdimm to 2.72 Volts (OK I remember this).
I prefer relaxed timings instead of stability problems and always I buy high specifications RAM (OK it costs more anyway)...
Damn now my RAM ran 3-4-4-8/2T but Vdimm was 2.51!!
And passes 32M at 265 FSB...
During 754 platform (I bought my 3200+ 754 13 months ago) I did not run any RAM burn-in programs.
Sometimes I tested CPU stability and that's all but I used to run CPU Stretcher 2 times per week.
Maybe CPU Stretcher improve my RAM?
I don't know!
This happening makes me happy of course!

Now my CPU is burning at 2450 MHz with Vcore2.
MSF2 was 2400 MHz (8M SuperPi).
Let's see what it will be!
I will stop increasing FSB using this Vcore until final crash!

Athens2004 I tried 2720 MHz yesterday and 32M passed but Vcore was 1.816!
I don't want to send for RMA my CPU now!