CPU & RAM burn-in guide

[MrSeanKon]

What would you recommend for burning in the new opteron cpus?

I don't have Opteron CPU but I use this method for two years (until this moment).
My samples as are Bartons, one XP, one 754 and two 939s...

I've got a 146 here stock voltage and stock vcore = 31° idle Temp on water. So it's running very hot, so what exactly should i try ?

You have to calculate your CPU wattage --> see in Bible's package CPU wattage (read manual first for link etc)...
Well 31C is not dangerous temp in idle.
First of all I don't know your cooling so what can I guess?
Ambient temperature?
I use thermistor and multimeter always LOL my mainboard shows 48C at full load but thermistor never show over 32 - 33C.
Anyway give us more info.

Have tried the S&M...damn it get my CPU under water to 50°C and my PWMIC at 65°C

Mate S&M is an AMD CPU killer if you run it at 100%....
The S&M author suggests 75% utilization.
See in S&M help file.

If It pass at normal settings for CPU and RAM time and full load CPU can I say it is 100% stable...and super_PI 32MB passed...?

If your CPU passed Normal FPU test (utilization 75 - 80%) then it is 32M stable.
I have not tested my RAM with S&M so I have no idea....
Keep in mind all systems do behave similar; I mean that for your system maybe OCCT is more sensitive instead of Prime95 but for my system not.
But in general I suggest StressPrime2004 instead of Prime95 cos for some AMDs Prime95 crashes immediately (I saw this many times for Bartons)...

Primoz if you can inform me some data I can help you more...
I mean if you have found other (MSF, Vcore) pairs.
If yes please post a table and run O/c analyzer exponents to see what is going on...
Your questions were OK cos I must prepare here a "Stability testing tutorial" in the future...

[ic3m4n2005]

As i said 31° idle and about 39° load, i'am on watercooling, triple rad with 3x120 fans on it. My 6800gt is in the circuit as well. My 3200+@2800@1,64v was 25° idle and 43° load just as a comparison. CPU wattage@stock is 67w as CBI tells me.

[MrSeanKon]

OK ic3m4n2005 thanks for info.
I think your temps are very good cos I have w/c also but not 2.8GHz...
Different CPUs and steppings....
So if you decide to keep 1.64 for permanent usage run CPU Stretcher at 2820 first (ZZZ=0, Normal Priority) until final crash....
You may increase your Vcore up to 1.700 Volts but I don't suggest you more.
As I see (both we are European) you live in Germany so temps are lower than Greece!

[ic3m4n2005]

Oh you misunderstood me
My old Venice was 2800mhz @ 1,64 volt, now i need help with my opteron 146. The problem is that it get's warmer as my venice at Stock volt and clock So that's my question, what should i exactly do

Edit: 2800Mhz seems to be stable @ def vcore with opteron

[Primoz78]

I don't have Opteron CPU but I use this method for two years (until this moment).

If your CPU passed Normal FPU test (utilization 75 - 80%) then it is 32M stable.
I have not tested my RAM with S&M so I have no idea....
Keep in mind all systems do behave similar; I mean that for your system maybe OCCT is more sensitive instead of Prime95 but for my system not.
But in general I suggest StressPrime2004 instead of Prime95 cos for some AMDs Prime95 crashes immediately (I saw this many times for Bartons)...

Primoz if you can inform me some data I can help you more...
I mean if you have found other (MSF, Vcore) pairs.
If yes please post a table and run O/c analyzer exponents to see what is going on...
Your questions were OK cos I must prepare here a "Stability testing tutorial" in the future...

Thx again! As I said I passed normal at 100% (not at 75%!) @ 2,65ghz but SP not I will make the table and will report you here...the strange thing is that at 1,552V is SP stable for 17hours withouth problem, but I cannot pass a single mhz more...even with 1,664V and ram dividers! I noticed that could be a MB problem...cause I have got an 3000+ venice 0517APAW that on my MB and same sistem as in sig can get 300*9 @ 1,456V but not an mhz more...tried out 1,65V but nothing (with TCCD rams even using dividers). Then I have sold this venice and it can do 2,8ghz and even more with >1,5V < 1,63V... . I have sent my MOBO on RMA and after a week of testings they said that it has some stability issues when stressed and OCed...but when they sent it to the official service they found notthing and I have get returned the same MOBO...I had several problems with BH-5 chips with this MOBO, had the Geil ONE W for a month and never get stable even at 200FSB...so gone for UTT CH-5 and it seems to be good now...except the CPU OCing issue... I have noticed that if my system is @ 2,65ghz or 2,7ghz and VCORE at least 1,63V my MB starts to "scream" (like the 160gb Hitachi disks 7k250 but louder)...when both cores are 100% utilized...Should be some electrical components (I do not know how they are called in english)


Regards,
Primoz

p.s. Sorry for making you borring with my "problems" people...

[MrSeanKon]

Regards ic3m4n2005.
I think you are OK cos your temps are very good.
If you are curious and impatient buy a thermistor and measure your CPU temperature at full load.
If not then see AMD crash temperature estimator.
When any AMD CPU has problems with temperature (near to its limits or bad stepping) o/c exponent n < 0.3.
For this reason I always suggest to everyone "Try to find your CPU limits, (MSFs, Vcores) not for many time --> Mapping method".

LOL here is my custom watercooling! (OK posts are in Greek) but see my post there to understand more.
LOL 24/7 my mainboard is outside!
LOL tank LOL it's a tanker = 15 lt!
But CPU temperature with 25C ambient temperature is 32-33C with programs such as XMPEG, DrDivX, Prime95 with S&M --> 3C more.

Sorry for making you borring with my "problems" people

No problem!
Damn this seems strange...
Cos StressPrime or Prime95 stretch both CPU + RAM.
And as you said your CPU passed 100% S&M --> it is rock stable...
Maybe your RAM is the problem OK you tried RAM dividers but low RAM frequency more and relax timings more.
Keep in mind some memory combinations amybe problematic --> system cannot POST --> you have to play in the beggining and increase Vdimm to maximum value don't afraid most RAM modules are very strong...
On the other hand I avoid Tagan PSUs cos my friend Kleanor had many problems with Tagan 480 Watts --> went for RMA.
Anyway I wait your table but if you can measure your voltages with multiplier is a good idea (PSU + Vcores + Vdimm) to see if they are stable or not.
O/c Bible's thread is here it has photos and instructions for this.
I don't care what MBM or CPUz or anything program shows...
Thermistor + multimeter RULEZ!

Primoz see this thread SoF has the same problem with Prime95...

[Primoz78]

As I said will report when new PSU will arrive for now thx! I will check the threads you gave me in the weekend.

Have a nice weekend

Regards,
Primoz

[Napoleonic]

how to use CPU stretcher correctly? what is "zzz" ? and shall we fill the "duration"?

[ic3m4n2005]

OK i've burned quite a bit last night
Started @ 2866mhz@stock volt. -> prime fails after 5 mins. Burned overnight with "Toast", 40° load the whole night. Result is ... nothing , how can i check results best ? SPI 4m or something like that ? My Voltages are stable, occt says so and i think 600w Powerstream should be enough

More recommendations are welcome

[MrSeanKon]

what is "zzz"?

ZZZ it is CPU utilization.
When ZZZ=0 --> your CPU will run at 100% (if it is single core or non Intel HT) and as more as you increase ZZZ CPU utilization becomes lower.
You can check this --> see % in Task Manager.

shall we fill the "duration"?

No you must not modify anything else which appears later (results).
Intel HT owners or AMD+Intel dual cores CPUs must run two threads for 100% CPU utilization.

how can i check results best ? SPI 4m or something like that ?

Don't confuse Mapping method --> a method for finding easily your first MSFs with stability testing.
As you saw my CPU was 4M stable at 2650 MHz --> I must run it 3*multiplier lower in the beggining.
At least you must be SuperPi 32M rock stable if you prefer this program.

OK i've burned quite a bit last night
Started @ 2866mhz@stock volt. -> prime fails after 5 mins. Burned overnight with "Toast", 40° load the whole night. Result is ... nothing

Mate CPU burn-in is something which needs patience and it will not give you results overnight.
As you said your are at 2866 MHz + stock voltage Prime95 stable for 5 minutes only.
OK no problem.
Use CPU Stretcher every night at 2866 first as I suggested above next day 2866 + multplier (same Vcore) next night 2866 + multiplier*2 and so on until final crash....
Then go back and see if you gain something with Prime95 but as you can see SoF has problems with Prime95 ---> use S&M or OCCT.

Sorry for fast questions I am too busy as I said above.

[Skinny13]

At least you must be SuperPi 32M rock stable if you prefer this program.

or maybe another program...if you have amd,right sean?

[MrSeanKon]

Skinny13 as you know I don't play games but many users test their systems playing games.
SuperPi 32M is a good stability program only for AMDs.
Especially it's more sensitive in RAM testing than MemTest (diskette MS-DOS bootable).

[Daved+]

I can't pass 293 at stock volts, I have burned in my CPU for like 300 hours with no progress so far. It's the CPU cache which fails at a faster speed.

[ic3m4n2005]

OK i'll try burning overnight till it crashes
Taost gives me more heat output than cpu stretcher, is it useful to get more heat ? It's about 2 degrees difference between the two programs.

[MrSeanKon]

I can't pass 293 at stock volts, I have burned in my CPU for like 300 hours with no progress so far.

1) Maybe your CPU does not "love" burn-in yeah it is possible.
2) Have you tried RAM dividers and relaxed timings?
A64s are strange CPUs and FPU power inside.

OK i'll try burning overnight till it crashes

I must to warn you again you must be very carefull cos of hard disk data corruptions.
Don't run this program if you are very unstable run it at semi-stable frequencies.

Taost gives me more heat output than cpu stretcher, is it useful to get more heat?

In my opinion burn-in should be performed with the smallest temperature.
If you need the maximu heat run S&M 100% CPU utilization.

It's about 2 degrees difference between the two programs.

For my system Toast arises 2.5C higher than CPU stretcher and other programs.
But running CPU stretcher at High Priority anything freezes and Toast at highest priority cannot freeze my system.

[MrSeanKon]

Until this moment CPU is 32M SuperPi stable at 2750 MHz.
I tried 32M at 2760 MHz but failed after 15 loops (18 minutes and 18 seconds computation time):



OK no problem let's try 16M:



Again crashed...
--> Focus <-- that computation time was only 4.5 minutes and 16M is not sensitive such as 32M.
Anyway 8M passed:



F.A.Q.

1) Is your RAM stable there? How you are sure?
Using RAM dividers (RAM/FSB=100/200 --> RAM was running at half speed of 276 --> 138 MHz only!) 32M failed again.

2)What's your schedule now?
8 - 12 hours burn-in every day, ZZZ=0 below normal priority, one thread for one week at 10 X 276, Vcore maximum (1.816 Volts).

3) How you are sure? Maybe your mainboard cannot go further....
LOL my mainboard achieves FSBs around 325 MHz FSB (see in previous page).

4) Ok then why don't you run 9 * 307 = 2763 MHz?
Cos my RAM cannot run 1:1 and 307 MHz and I avoid dividers.
I think (OK I have not tested again) RAM MSF is around 300 MHz (maximum Vdimm=2.86, 3-4-4-11/2T).

5) OK then run your system 9.5 * 290 MHz = 2755 MHz.
Decimal multipliers are not recommended for A64s.
See multiplier RAM effect program.

6) Why do you run CPU stretcher at Below Normal or Lowest priority?
Cos I have my PC for main activities not only for burn-in!
For example I can compress my videos for 3-4 hours smoothly (98 - 99% CPU utilization) or Maths for some minutes, C# programming for O/c Bible etc....
But when the main program does not need 100% CPU power --> CPU stretcher will do it for some seconds or more.
Thus Task Manager shows always 100% CPU utilization.

7) I run CPU stretcher at Highest priority and two threads. Time duration is not renewed.
Please visit our forum for latest announcements/bugs/tips and tricks/read my O/c guides etc...
As I said above I don't want to bother users and especially all Moderators/Administrators here.
My threads at OcTech.gr are written in English.

8) Of course burn-in has its limits.
Yes and I think it is obviously.
It is impossible after a couple of months to achieve 3GHz!
Any CPU has its limits and some CPUs love burn-in and other no.
So I don't know what is going on with your CPU I am not a magician!
You have to try!

9) OK man you are crazy please give us what's your estimation for the CPU limit?
Here are the some of first MSFs before burn-in using Mapping method (8M SuperPi):



Results:

[MrSeanKon]

-->Continued...<--

You understand from 2310 --> 2650 MHz (1.361 --> 1.663 Volts) n=0.685.
Thus from 1.361 --> 1.816 n < 0.685 --> visit Ocforums Bible's thread --> see "O/c Analysis in depth".
After first burn-in CPU is 32M stable at 2340 MHz + Vcore1=1.361.
Let's compute exponent n near to limit:



Results:



Users who are familiar with Basic Calculus understand now what exponent n means...
It is something like differential but the internal formula uses logarithms instead of function values.
Sorry I cannot explain more Maths topics cos I am not Mathematician and it is too hard, anyway if you don't understand very well this no problem follow my ideas and do not care anymore.
All people use many programs but we don't know the source code.
Thus from 1.663 --> 1.816 n < 0.366.
Remember that MSF1=2680 MHz + Vcore1=1.663 Volts (32M stable after burn-in).
O/c analyzer find frequency/FSB now:



WOW!
Program shows that my CPU limit will be 2770 MHz (32M SuperPi stable).
So I am running my CPU near to its limits!
This means something very simple!
If I want 32M SuperPi stable at 2780, 2790 and so on I must change my cooling system (from custom watercooling --> subzero) or immigrate in Canada/Alaska
All readers can understand now the importance of Mapping method.
It helps me and I know before my CPU limits without any suggestions.

10) After this analysis what you expect?
Xm if after one week my CPU can pass 32M at 2760 MHz I will continue for 2770 else I will stop and maintain gains (see one first page my recommendations).

11) Man this program is difficult...
As you know any captain have the ability to manage a tanker.
The majority of captains are not mechanical engineers or naval engineers.
They know the basics.
Thus it is not necessary to be MrSeanKon for running O/c Bible, anyway this depends on you.

12) What's your HT ratio?
It is 3X cos 3 * 276 = 828 and 4 * 276 = 1104 MHz.
Many people suggest HT <= 1GHz (1000 MHz).

13) Burn-in works for old processors?
Of course you must burn-in your CPU when it is fresh not after six months!
OK if your CPU is one month old it is probably to gain something.

14) What about RAM burn-in?
It works and gains are bigger.
Use Sysoft Sandra RAM burn-in and MemTest86.

15) Why don't you run StressPrime 2004 or Prime95?
LOL see this photo:



LOL my CPU runs at 2758.6 MHz and LOL StressPrime 2004 shows different speed.
LOL FSB is 14 MHz or 192 MHz?
If FSB > 265 MHz StressPrime 2004 shows anything different for actual speed/FSB and multiplier.
That's why I banned Prime95 + StressPrime 2004.

Please avoid quote my large posts cos include many photos.
If you want to quote something specific you should remove non necessary sentences and [img] commands.
Otherwise any reader will see an enormous quote and this is "terrible".
As you can see I have quoted many times other users post but I prefer multiple and small quotes.

I must stop!
Enough!
Later guys!

[ic3m4n2005]

OK ty so far, will read your posts on octech.gr and will try a few days more, hope it'll work for me too

[MrSeanKon]

I tried 32M and 16M at 2760 MHz but crashed again after 3 days of burn-in.
Thus I was right cos I said "the new MSF will be up to 2770 MHz".
Vcore is too high anyway I stop burn-in process now I will not continue more and just I maintain CPU gains.

[KingYubbo]

This burn-in method. It requires regular burning-in yes yes? or is it a once off big thing?

[FE4R]

FE4R you should give me more info:
1) What's your mainboard?
2) What's your cooling?
3) What Vcore you are thinking for permanent usage?

You did OK cos in the past your CPU passed Prime95 at 2.4GHz with Vcore=1.35.
But don't use ZZZ at this moment cos you are running near to previous MSF I mean 2420 - 2400 (the previous MSF) is not great difference....
Also be careful with priority don't set Above Normal or High Priority right now...
Find first how far you are CPU Stretcher stable (ZZZ=0, Normal Priority, Threads=1) and then if you want to go higher set ZZZ=20 - 50 and Lowest Priority...
Run this program at least for 6 hours per day but as you saw my recommendations don't run it for many hours!
You must relax your CPU for some hours!

Thanks for replying to me man, but I havent been on in awhile since I dont have any more time. I have stopped using CPU strecher for awhile too. I have alot of programming work to do so no more messing with comp -at least until christmas holidays since this is my work and game comp.

Anyways, I have the GA K8N Ultra 9 board, I am currently on stock cooling but my cooler should reach any time now, the G Power Cooler Pro (air cooling) from Giga-Byte (read alot of good reviews on it) and I am thinking about using it at a permanent vcore of at most 1.55V, preferable 1.45V.

[MrSeanKon]

.It requires regular burning-in yes yes?

What do you mean exactly?
CPU Stretcher don't increase temperatures like Toast but stretches ALU+FPU.

OK FE4R thanks for your post.
Keep on computing.

[Lithan]

He's asking if you have to keep doing it to keep your clock or if you just do it once, and once it's stable where you want it you are done burning in.

[MrSeanKon]

Thanks!
I was thinking unfortunatelly I must relax for a couple of days..

if you have to keep doing it to keep your clock

Yeah as I showed above my CPU 32M limit is 2750 MHz --> that's why I have to maintain the gains.

and once it's stable where you want it you are done burning in.

I performed this process at 2650 MHz (8M SuperPi stable Vcore=1.663) and after some days --> 32M stable at 2680 MHz (Vcore=1.663 also).

[ringer94]

Great OC burn-in guide SK - I want to try it on a new XP-M 2600+. I don't have a MV meter but I think I used 8ardvore to get an accurate Vcore voltage on an NF7-S V2. Might be a workable alternative for those who don't have voltmeters.