Sign in with
Sign up | Sign in
Your question

Strategy for overclocking a CPU to its limit! Will this work?

Last response: in Overclocking
Share
September 8, 2008 12:18:49 AM

Hi,

I will be buying a new semi-gaming rig in 2 weeks and I just wanted to see what's the fastest way to (safely) get the most juice out of an overclock.

I've read the great guide that was stickied about overclocking C2D, but I want to know if this could be a faster (and more brute) way to find the limits of the CPU in a system.

I'd greatly appreciate any comments and suggestions (and reasonable flaming ;p) for the plan below...


- Part A) Finding the maximum vcore by CPU temp monitoring (assuming temp is mostly dependent on vcore)
1) Pump vcore to the max (e.g. 1.3625 V for E7200) & keep CPU at stock speed
2) Start the torture test on Prime95
3) Monitor the CPU temp
4) If the CPU temp exceeds its limit, reboot & lower vcore
5) Return to 2) until the CPU temp on max load is acceptable, then fix system at this vcore

- Part B) Finding the maximum clock speed for the current vcore
1) Increase clock speed to as high as machine is bootable
2) Start the torture test on Prime95 (include error check)
3) If errors pop up in less than ~24 hours, then reduce clock speed
4) If nothing goes wrong, then you got it! :D 
5) Double check temperature to make sure it's not cooking the chip...
a b à CPUs
a c 86 K Overclocking
September 8, 2008 12:34:49 AM

Usually you take your time little by little by tweaking till you find it.
You have to play with NB voltages, memory timings, multipliers etc. It's a journey.
a b à CPUs
a b K Overclocking
September 8, 2008 5:20:51 AM

pcxxy,

Welcome to Tom's. Nice summary, but unfortunately, you've over-simplified the process. As Conumdrum has pointed out, there is no overclocking "EASY" button!

Please read graysky's Overclocking Guide at the ^TOP^ of this Forum: HOWTO: Overclock C2Q (Quads) and C2D (Duals) - Guide v1.6.1 - http://www.tomshardware.com/forum/240001-29-howto-overc...

He's already taken a great deal of time, and made more than a significant effort to explain all of this in great detail for the benefit of everyone.

If you know something that graysky hasn't already included in his Guide, then please feel free to enlighten us. If you've worked through his Guide and haven't been successful, then we'll be happy to help you.

If you have questions about temperatures, then check out the Temperature Guide below in my signature.

Comp :sol: 
Related resources
September 8, 2008 8:44:24 AM

hmm, although conundrum, you could pump up vcore till it's stable, then pump RAM till its stable, then pump NB voltage till its stable, then continue to part B. memory timings would be on auto, and tweaked lastly to the max.

I'm not seeing anything wrong with that?
September 8, 2008 9:19:58 AM

There other area of concern is a Core 2 at 2.4ghz 1.5 volts still generates less heat than a Core 2 at 3.2ghz 1.5 volts.
a b à CPUs
a b K Overclocking
September 8, 2008 9:50:29 AM

That method will probably get you no further with overclocking than I get, which is not far. Then again, I don't aim for stability. There are lots of other things to take into account, memory timings as previously mentioned, but possibly also chipset quirks, especially if you are lucky enough to be using something like P965.
September 8, 2008 10:19:20 AM

I over clock to the next chip level, just to get that chip that costs twice what I paid for mine.
September 8, 2008 11:08:34 AM

oh yeah jdocs... zomg what have i been smokin :??: 

lol randomizer you don't care about stability? :lol: 
a b à CPUs
a b K Overclocking
September 8, 2008 11:44:44 AM

V3NOM said:
lol randomizer you don't care about stability? :lol: 

I'm running an E6600 at 2.7GHz now, anything more doesn't really net me any more performance so I can't justify the time spent overclocking "properly". I did hit 4GHz, it was unstable enough to crash Abit uGuru but run CPU-Z and Paint. That was at 1.725V on air. Pretty much my overclocks are to get nice CPU-Z screenshots, and to save time I stick to the following simple rule:

Moar powa! :bounce: 

This is why I'm sticking to 65nm CPUs ;) 

Note to OP: Don't try 1.725V on air with that E7200 if you are indeed using one. In fact, don't try it on water either.
September 8, 2008 11:55:11 AM

LOL!
a b à CPUs
a b K Overclocking
September 8, 2008 12:34:02 PM

LOL! Yep. There are a LOT of people out there bragging about their big overclocks, which wouldn't run a program stable for more 30 seconds if your mother's life depended on it. It's a fact.

To the OP, your method may indeed work to get you started, but overclocking is not an exact science. Every motherboard, every processor, every stick of memory although binned the same, is slightly different and will set-up some unique cirmcumstances to overcome and tweak through to get the very highest speeds possible, stable.
September 8, 2008 12:41:39 PM

thx for the welcomes & suggestions

Of course I have read and learnt tons from graysky's guide (how could I not? ;P). However it seems to me that he stressed more on minimizing the core voltage rather than getting the most out of the cpu.

In his guide, he aims to find the lowest working voltages for a given clock speed, and you can see that it's obvious that he's totally running out of patience towards the end with his settings. :D 

As far as I know, the CPU temp (btw, thx for your great guide too computronix) is one of the dominant factor that limits how far you can overclock (besides things like the limit of the CPU, ram, other things). So I was thinking why not we start from there?

My plan was to find the outermost limits, then back down to a stable level.

Most likely, I will also need to do voltage minimizing after finding the right clock speed.... but anyways... ;p

@Jdoc: thanks for helping me confirm that heat is clock speed depending. I think I can implement some steps in my overclocking strategy so that this is taken care of too.
a b à CPUs
a b K Overclocking
September 8, 2008 12:45:50 PM

pcxxy said:
My plan was to find the outermost limits, then back down to a stable level.

Well the outermost limits are pretty far, and dangerous. You should probably decide how far you want to go for now and work on that. With your chip, 3.2-3.4GHz is where you'll see massive increases in voltage needed for diminishing returns on speed.

pcxxy said:
Most likely, I will also need to do voltage minimizing after finding the right clock speed.... but anyways... ;p

Pretty much. The lower the better for any given speed, as long as it's stable.
September 8, 2008 12:52:11 PM

hmm, just wondering... how dangerous is 1.3625 V on a E7200 at stock speed? I planned to use an aftermarket cooler (OCZ Vandetta 2)... and by 'dangerous' did you mean the temperatures will go nuts? or there's a chance that these chips can't take the specified voltages?
a b à CPUs
a b K Overclocking
September 8, 2008 11:57:55 PM

Why would you run 1.3625V at stock speed? Don't even change the voltage at stock. There is no reason to overvolt at stock speeds.

By dangerous, I mean electromigration killing your chip.
September 9, 2008 6:35:14 AM

electromigration :o 
a b à CPUs
a b K Overclocking
September 9, 2008 6:58:54 AM

Guys,

Please read post #19 (shinigamiX) from the ever-eloquent enlightenments of the legendary JumpnJack - Electromigration: http://www.tomshardware.com/forum/217287-28-better-unde...

... in doing a search I found a 'term paper' on a college server by some student it appears that actually did a great job summarizing the whole electromigratin thing, I really like this paper as it does a good job going into basic detail without a huge technical overhead:
http://www.eng.uwaterloo.ca/~asult [... said:
30vlsi.pdf

Particularly, see figure 2-1 on page 12, it essentially summarizes electromigrationhttp://en.wikipedia.org/wiki/Electromigration in a simple picture. I will refer to this PDF in this reply.

Other articles on electromigration to get the intresest going (some are subscription based, but a library would also have them):
http://pmos.upc.es/blues/publicati [...] 7_7_97.PDF
http://ieeexplore.ieee.org/search/ [...] er=1044338
http://www.mrs.org/s_mrs/sec_subsc [...] ion=detail

And my favorite:
http://ieeexplore.ieee.org/xpl/fre [...] er=1493069
This one concerns ESD induced electromigration failures as latent ESD failures that can shorten the lifetime to as little as a few weeks.

Ok, now to the answer ----

The answer is actually simpler than you think.... the short of it is, increasing frequency also increases the total current through the device, hence, the metal lines will experience higher current density and higher electromigration degradation.

Here is the explanation .... those who have watch me post know I am keen on the td=CV/I, where td is gate delay, C is total capacitance, V of course is voltage, and I is current or Idsat, drive current. This is a fundamental equation - http://en.wikipedia.org/wiki/Equation - describing the max switching speed of a device. However, in this form and this way of thinking td is the dependent variable and is a function of C, V, and I -- all three of which are design parameters, what happens after we have optimized the process and nothing changes any longer --- then we rearrange this equation and, in this case, let's look at I as the independent variable:

I = CV/td or CV*(1/td)

But 1/td is one over time which is frequncy, f. So ---

I = CV*f

Thus, since C is fixed by the oxides, wires, and transistors in the CPU, V is dialed by you the user, and f is set by the clock generator, then I is a direct function of frequency AND voltage.

In the link above, electromigration lifetimes are modeled by Black's equation (see link above):

tf = A * (1/J)^n * EXP(Ea/kT)

A is material dependent, J is the current density which is I/unit area cross section of the wire, n is a emperically determined exponent, Ea is activation energy, k is the Boltzmann constant, and T is temperature. Key here is current density J as the current goes up so does the electromigration factor.

So really, the increase of electromigration with frequency is no more than an increase in current driven by the frequency generator. Pretty simple.

Side note: Validate my I = CV*f equation, recall that I also post many times that the equation for dynamic power is P=CV^2*f , well with a little algrebra check this out ---

P = I*V ===> fundamental electrical power equation.

Substitute the expression for current as a function of frequency from my argument above,

P = (CV*f)*V = C*V^2*f wow, now we see where the dynamic power equation comes from and that power really goes as a cube of the 'speed fundamental varibles --- 2 orders in voltage and 1 order in frequency.

EDIT: NOTE --- though volting and clocking up your CPU can increase the rate of electromigration, the time scale here is still VERY long. Pausert20 is correct, the heavier Cu atoms, the lower resistance and lower temperature of the wire as a result has pretty much eliminated electromgiration problems, they still exist but the lifetime is so long that it makes littel difference based on the turn over rate that we typically upgrade to.... (We meaning enthusiast).

TDDB is probably the most common failure mode if suiciding a chip. Notice that it has the same exponential form as electromigration, they are both first order affects.

Jack]... in doing a search I found a 'term paper' on a college server by some student it appears that actually did a great job summarizing the whole electromigratin thing, I really like this paper as it does a good job going into basic detail without a huge technical overhead:
http://www.eng.uwaterloo.ca/~asult [...] 30vlsi.pdf

Particularly, see figure 2-1 on page 12, it essentially summarizes electromigrationhttp://en.wikipedia.org/wiki/Electromigration in a simple picture. I will refer to this PDF in this reply.

Other articles on electromigration to get the intresest going (some are subscription based, but a library would also have them):
http://pmos.upc.es/blues/publicati [...] 7_7_97.PDF
http://ieeexplore.ieee.org/search/ [...] er=1044338
http://www.mrs.org/s_mrs/sec_subsc [...] ion=detail

And my favorite:
http://ieeexplore.ieee.org/xpl/fre [...] er=1493069
This one concerns ESD induced electromigration failures as latent ESD failures that can shorten the lifetime to as little as a few weeks.

Ok, now to the answer ----

The answer is actually simpler than you think.... the short of it is, increasing frequency also increases the total current through the device, hence, the metal lines will experience higher current density and higher electromigration degradation.

Here is the explanation .... those who have watch me post know I am keen on the td=CV/I, where td is gate delay, C is total capacitance, V of course is voltage, and I is current or Idsat, drive current. This is a fundamental equation - http://en.wikipedia.org/wiki/Equation - describing the max switching speed of a device. However, in this form and this way of thinking td is the dependent variable and is a function of C, V, and I -- all three of which are design parameters, what happens after we have optimized the process and nothing changes any longer --- then we rearrange this equation and, in this case, let's look at I as the independent variable:

I = CV/td or CV*(1/td)

But 1/td is one over time which is frequncy, f. So ---

I = CV*f

Thus, since C is fixed by the oxides, wires, and transistors in the CPU, V is dialed by you the user, and f is set by the clock generator, then I is a direct function of frequency AND voltage.

In the link above, electromigration lifetimes are modeled by Black's equation (see link above):

tf = A * (1/J)^n * EXP(Ea/kT)

A is material dependent, J is the current density which is I/unit area cross section of the wire, n is a emperically determined exponent, Ea is activation energy, k is the Boltzmann constant, and T is temperature. Key here is current density J as the current goes up so does the electromigration factor.

So really, the increase of electromigration with frequency is no more than an increase in current driven by the frequency generator. Pretty simple.

Side note: Validate my I = CV*f equation, recall that I also post many times that the equation for dynamic power is P=CV^2*f , well with a little algrebra check this out ---

P = I*V ===> fundamental electrical power equation.

Substitute the expression for current as a function of frequency from my argument above,

P = (CV*f)*V = C*V^2*f wow, now we see where the dynamic power equation comes from and that power really goes as a cube of the 'speed fundamental varibles --- 2 orders in voltage and 1 order in frequency.

EDIT: NOTE --- though volting and clocking up your CPU can increase the rate of electromigration, the time scale here is still VERY long. Pausert20 is correct, the heavier Cu atoms, the lower resistance and lower temperature of the wire as a result has pretty much eliminated electromgiration problems, they still exist but the lifetime is so long that it makes littel difference based on the turn over rate that we typically upgrade to.... (We meaning enthusiast).

TDDB is probably the most common failure mode if suiciding a chip. Notice that it has the same exponential form as electromigration, they are both first order affects.

Jack


I hope this answers everyone's questions, as this information pertains directly to 65 nanometer processors, but indirectly to High K Gate 45 nanometer processors, since the 45's are sill too new upon which to draw conclusions.

Comp :sol: 
September 9, 2008 7:09:57 AM

yeah i kinda figure what it's going to be... when the electrons skip across the cpu resulting in ridiculous instability?
September 9, 2008 1:02:55 PM

o... electromigration... it's like when the ****ans illegally step into another country... ;p

anyhows, you asked 'why overvolting'?

that's to determine how good the cooling of the system can reach, so you can get the highest clock speed that you can. it's not like i'm going to run it at 1.3625V at 2.56 GHz as my permanent settings... (unless maybe when it gets too cold in the winter ;p) - and from my understanding, these chips don't get fried instantly... no?

I know that there are other limitations to clock speeds, and diminishing returns of clock speed gains for each additional mV, but I just want to see how high the temperature gets...

maybe a better modification of my overclocking plan is to set the voltage at a moderately high value (1.3X V), and some randomly high clock speeds and see how hot/stable it gets... then go further or ease off depending on what you get....

but if i did that it would be just the same as the long journey that graysky is taking... where he obviously ran out of patience towards the end of his clocking journey (but i still love his guide)...

n_n
a b à CPUs
a b K Overclocking
September 9, 2008 11:19:38 PM

Test your cooling by doing a real overclock! Overclock first, worry about bad temps when you get to them ;) 
September 10, 2008 7:09:31 AM

damn right! 71 degrees under load on my C2D.... pffff no big deal!
September 10, 2008 2:05:50 PM

i like that plan! :) 

would you guys know whether the two sSpec of E7200 (SLAPC and SLAVN) make any difference?
September 11, 2008 7:51:30 AM

meh a 25cm fan cut into the side of my case should sort it out! 105cfm at 20dba FTW! :) 
September 21, 2008 7:35:06 PM

So I've bought my system (two in fact, one for my friend and I would be OC'ing for him soon)...

CPU: Intel C2D E7200 (SLAVN)
HSF: OCZ Vendetta 2 (with stock fan at max speed)
Thermal Compound: MX-2
Motherboard: GA-EP45-DS3L (BIOS version F8)
RAM: 2 x 1 GB DDR2 800 MHz, running at 4-4-4-15 @ 2.0V (OCZ2P800R21G)
Video Card: Palit HD4850 512 MB (the dual slot non-reference version)
HDD: Seagate 500 GB, 32 MB cache, SATA2
Optical Drive: Pioneer DVR-216D
PSU: OCZ600W SXS
Case: Antec Three Hundred

I spent half a day finding a stable RAM voltage for the 4-4-4 timings and a stable CPU clock speed, and I managed to get a semi stable clock of 3.8 GHz with 9.5 x 400 MHz, at a BIOS CPU voltage of 1.3625 V, with core temperatures of 32//53C at during idle//load (by RealTemp).

Owing to Vdroop, the voltage that feeds into the CPU is around 1.312 V to 1.328 V depending on load. Since Vdroop is designed to protect the system, I won't try to get around it (nor will I go to higher voltages because I will be keeping this box for years).

At 3.8 GHz, the system is stable for at least 8 hours while idling, but does crash within a few hours under load.

Now I backed down to 9 x 400 MHz, and torture testing with Prime95 (small FFT) for about 5 hours with no errors (still running), and I will see if a lower voltage will keep my system happy. =)

Not bad for half a day of probing around eh? ^^

edit:
@1.3625V (BIOS) = 1.328-1.312V (CPU-Z) -> No errors after 9 hr of small FFT
@1.35V (BIOS) = 1.312-1.296V (CPU-Z), -> No errors after 10 hr of small FFT test
@1.325V (BIOS) = 1.296-1.28V (CPU-Z), -> still running... has been stable for 30min

load temps (RealTemp) are dropping from 53 to 47 (3 degrees due to lower voltage, 3 degrees due to ambient temp change)
September 22, 2008 7:22:18 PM

dang... failed after 3 & 4 hours @1.325V. Gonna pump it up some more ;) 
September 26, 2008 2:13:30 AM

well i just got my S1963 (S1283 was like double the price and only 5 degrees better plus the added size and weight :??: ) and it only dropped my cpu temp 10C with AS5 compared to the stock heatsink with stock paste so i'm a little disappointed but nevertheless i overclocked my E4400 (2.0GHz) to 2.9 which is a fair bit higher than 2.66 i got on the stock cooler... anyway that appears to be all i can go to! my motherboard just will not get to 300MHz FSB for some reason :(  and unfortunately, it's a Via and i can't change the vcore - its either automatic or it doesn't change at all :(  i have a feeling its the mobo holding me back unfortunately not the cpu or ram.
September 29, 2008 1:48:22 AM

sounds good.

so i've found my minimum stable voltage for running at 3.6 GHz :) 

unfortunately, prime95 kept giving me errors in large FFT and blend torture test after a few hours. i know it's not my ram coz i ran memtest86+ and it passed 23 cycles. i know it's not my harddrive coz i've done a surface scan and there're no bad sectors. i've even tried upping the northbridge (edit: MCH) voltage from 1.1v to 1.3 but that didn't help either... would anyone know what's going on?

in case you guys wonder, i've tried oc'ing my friends computer and right off the bat i took it to 3.8ghz at 1.3625V (set in bios)! unfortunately the temps (by realtemp) were 72C so it looks like the HSF wasn't installed properly... but otherwise it looks like he's got a better chip xDD
a b à CPUs
a b K Overclocking
September 29, 2008 4:42:31 AM

pcxxy said:
i've even tried upping the northbridge (ICH) voltage from 1.1v to 1.3 but that didn't help either

ICH is southbridge, MCH is northbridge ;) 
September 30, 2008 2:48:50 PM

whoops for my typo ;)  but would any of you gurus know what other reason can cause prime95 to get errors in large FFT tests?
October 1, 2008 7:10:43 AM

possibly ram? have you blended for a couple of hours?
October 3, 2008 3:06:04 AM

yeah I did blend too and it would give me errors anywhere from 1-6 hours... I've done memtest86+ for many hours though without errors.

i've read that i should trust the results from memtest more, but if my rams are ok then i can't figure out what's wrong... -.-;
October 3, 2008 8:30:20 AM

>.< if blend fails then it's just not stable imo... what's the voltages, speed and latencies of your RAM?
October 3, 2008 1:19:37 PM

i'm running stock timings for my ram which is 4-4-4-15, and i haven't played around with subtimings (although in the motherboard for memory performance, i chose the 'extreme' profile instead of standard or turbo... (default is 'turbo', but i should test it with standard at least once too...')

the voltage i'm currently using is 2.00V (default voltage is 1.9-2.1V but up to 2.2V is fine). This is the minimum required voltage to be free from errors in memtest. I've put it to 2.10V too but I still had errors in prime95.

In case you don't want to scroll up, my FSB is 400MHz and I'm running in synchronous mode, so that's not even overclocking my sticks.

Anyways, i'm going to run it at lower timings and performance profile and see if things are ok first, and see what's going on.

update: in the BIOS, at 5-5-5-15, 2.00V, "standard" ram performance enhancement profile, i get errors at with blend tests after an hour or so when using fft length of 896k.

i should probably try one ram at a time this weekend... -.-;
!