[felinusz]

INDEX

Part 1: Preamble: Voltage, The Magic Ingredient - Below
Part 2: What Does Overvolting Actually Do? - CLICK HERE
Part 3: Overvoltage and "Safety" - CLICK HERE
Part 4: Thorough and "Safe" Overvolting Technique - CLICK HERE


PREAMBLE: VOLTAGE, THE MAGIC INGREDIENT

Voltage is the ingredient that grants greater stable MHz potential to our transistor based computer hardware. Our RAM, processors, video cards, and motherboard chipsets all see gains from overvoltage. Most of us have overvolted hardware in the past; we know and accept the fact that overvoltage helps us to overclock, when used within reason. We also know that overvoltage can damage hardware, or diminish hardware lifespan. For the uninitiated, overvolting is the act of increasing component voltage past manufacturer specification.

But what does voltage actually do? How much is really too much? What are some dangers to Overvolting? Are there numbers set in stone, or is each piece of hardware different? How can we accurately and safely determine safe overvoltage limits for our expensive hardware, without killing it?

These questions all have answers which will help us to overclock safely and with excellent results, by giving us a greater understanding of how our hardware works, and what its limits are.

[felinusz]

WHAT DOES OVERVOLTING ACTUALLY DO?
HOW DOES OVERVOLTING HELP US OVERCLOCK?

To begin, it is best to explain exactly what overvoltage accomplishes, and how voltage works in the first place. If you want to get right into the theoretical overclocking side of things, and aren’t curious at all about how it all works, feel free to skip this very lengthy bit – or come back to it later on.

Overvolting does not increase overclocking potential by giving our hardware "more juice" or "more fuel". It increases overclocking potential by altering signal strength.

Our computers use a language of 1s and 0s – the binary language of computer processes. Physically speaking, these 1s and 0s occur through voltage highs and voltage lows, two signals representing a 0 (voltage low), and a 1 (voltage high). 0V typically represents voltage low, referred to as VSS. Voltage high is typically referred to as VCC, VDD, or VCORE, and is a variable voltage, dependant on the specific piece of hardware in question, and the transistor type used within that hardware (For example, a 90nm Athlon64 3200+ has a stock VCORE of 1.5V. A 90nm intel 540J 3200 MHz Pentium 4 has a stock VCORE of 1.4V).

So, 0V (voltage low, or VSS) is treated as a 0. A voltage close to Voltage high (VCC/VDD/VCORE), is treated as a 1. Our transistor-based hardware is essentially a massive grid of constantly switching voltages, representing logic 1s and logic 0s – the binary language in a nutshell.

The point of importance to us as overclockers here, is the "If our processor sees a voltage close to Voltage high (VCC/VDD/VCORE), it treats it as a 1." bit. Because of various resistances, our hardware’s transistors must have a tolerance for voltage high – the exact value of VCORE/VDD/VCC is rarely seen. From now on, I am going to refer to voltage high as VCORE, for simplicity’s sake.

Lets use an example to explain this tolerance, where we have a transistor voltage high (VCORE) of 1.4V, and a tolerance of ~5%. If the stock VCORE is 1.4V, and a signal of ~1.35V is seen, it will be regarded as a 1 (the tolerance allows for a ~5% loss of voltage high – or a minimum of ~1.33V). The really interesting bit for us, is that when the tolerance threshold is exceeded, our hardware starts to mess up. In our little example (with a ~5% tolerance, and a 1.4V transistor), if a voltage of 1.25V were to be seen, it would likely be regarded as a 0 instead of the 1 it was supposed to be – tolerance has been exceeded. Our signal strength (the signal itself being the voltage) has weakened enough that the tolerance is exceeded, and our hardware makes a mistake – stability is compromised! Most modern transistors have a tolerance ranging from ~2% through ~10%.

Overclocking our hardware can throw our voltage signals out of tolerance, and cause problems when a 1 (VCORE, voltage high) is mistaken for a 0 (VSS, voltage low).

The best way to explain how this happens when we overclock, through use of imagery, is through the use of a runner. This runner is running back and forth on a 100 foot track. He can either be at one end of the track (Voltage high - VCORE), or at the other end of the track (Voltage low - VSS). But, our runner cannot immediately switch from one end of the track to the other (and likewise, our VSS cannot switch to VCORE instantly). There is a transitional period where our runner is partway between the different ends of the track. The runner is rapidly running from one end of the track to the other (this is the same as our signal switching from VSS to VCORE), and although he is quite quick, there is still a delay between each end of the track (there is also a delay when our voltage signal switches from VSS to VCORE).

The rate at which he is expected to get from one end of the track to the other in ten minutes is our frequency, similar to the frequency at which our hardware operates. At a "stock", un-overclocked frequency, the runner is easily capable of making it to either end of the track in time. The frequency of ‘on’/’off’ signals – voltage high and voltage low signals - representing 1s and 0s, is how fast our hardware can ‘think’ and process. When we overclock, we increase the frequency at which the runner needs to make it from one end of the track to the other (we increase the frequency at which our signal needs to switch from VSS to VCORE), and we shorten the amount of transitionary time allowed for the runner to make it to the other end of the track (when we overclock the frequency, we shorten the amount of transition time allowed for VSS to switch to VCORE).

We increase the frequency (overclocking), and we get to the point where it is impossible for our runner to completely make it to each end of the track in the amount of transition time that he is given. The runner is simply not given enough time to make transit from one end of the track to the other, given the extreme frequency rate expected of him. Now our transistor tolerance comes in. The runner only really needs to make it to the 95 foot mark on the track in order for his run to be registered as a 1 (A 5% tolerance). Increasing the frequency slightly (and as such shortening the transition time the runner is given), the runner is still able to make it to the 95 foot mark, before he has to head back towards the other end of the track in order to meet his frequency schedule. But when we increase the frequency too much, the runner cannot even make it to the 95 foot mark before he has to turn back towards the other end of the track to meet his frequency schedule (VSS cannot switch to VCORE completely within the transitionary time allowed). Our runner only makes it to, say, the 90 foot mark, and is no longer within the tolerance – his run is counted as a 0 instead of as the 1 it is supposed to be. This represents an overclocked an unstable processor.

In our transistor based hardware, it takes time for the VSS voltage low (0V) to switch to our VCORE voltage high. When we overclock our frequency, we shorten the length of time available for that transition to take place. When there is inadequate time for VSS to change to VCORE, the signal (the signal itself being voltage) doesn’t make it all the way to VCORE – at a certain point our transistor’s voltage high signal tolerance is exceeded, and the VCORE signal is not strong enough to be registered as a 1 anymore. Instability occurs as a result – our 1s are being mistaken for 0s, and the computer cannot make sense of it.

Overvolting can alleviate this problem. The issue lies in the amount of time that it takes for the signal to change from VSS to VCORE – the signal can’t switch quickly enough to reach a strength recognizeable as a voltage high (VCORE) by our transistors. When we increase our voltage high value (overvolting), we force the signal/voltage to reach a higher voltage high, but in the same amount of time as before. We stretch out the ‘range of motion’ (the difference between VSS and VCORE), but we leave the transition time alone. The result is that it takes considerably less time for the signal to switch from VSS to a VCORE that is within transistor tolerance – this accommodates our faster switching frequency, and keeps our overclocked signal switching frequency strong (stable) and within transistor tolerance.

Lets go back to our runners. Now we have two runners, and shall compare them in order to explain how overvolting alleviates the problem of inadequate signal transition time – which directly leads to instability. In this example, the measurement ‘feet’ represents voltage signal strength, and the measurement ‘time’ represents the transition period between VSS and VCORE. Runner #1 can run 100 feet in 20 seconds (Stock VCORE), Runner #2 can run 110 feet in 20 seconds (VCORE overvolted by 10%). Runner #1 can run 5 feet per second. Runner #2 can run 5.5 feet per second. Our overvolted signal switches slightly faster than our stock signal, in the same time period, just as runner #2 can run further in the same time period. Please keep in mind that this is a simple example for explanation, and that signal switching speed is not so linear as "10% overvolt=10% signal switching speed increase".

At a stock frequency, we have 20 seconds to get within transistor tolerance, say it’s 5%, or 95 feet. Both runners make it. Now we overclock our frequency, and shorten the amount of time in which the runners have to get within our transistor tolerance of 95 feet. We increase the frequency, and shorten the transition time to 18 seconds. Runner #2, running at 5.5 feet per second, makes it to the 99 foot mark in 18 seconds, he is within tolerance, and his run is correctly counted as a 1. Runner #1 isn’t fast enough. He makes it to the 90 foot mark in 18 seconds, and is not within the tolerance. Runner #1’s run is incorrectly counted as a 0, and we have instability. In this manner, increasing signal switching speed to directly increase signal strength, overvolting allow us to increase our frequency (overclocking), without compromising stability when we also shorten the signal switching transition period.

That about sums up how voltage works in our transistor based hardware, as well as the effect that overvoltage has on increasing our hardware’s stable overclock potential.

[felinusz]

OVERVOLTAGE, AND "SAFETY"

This section summarizes the dangers of overvolting.

One remarkably common misconception about overvoltage, is that there is a number set in stone for each particular component, a number which dictates the "safe" overvolt that one can use with any specific piece of hardware. This is not true.

It is partially valid however. Specific types of hardware are almost always similar in their limits, granted no manufacturing defects. However, each piece of hardware is still unique to small, varying degrees. Being able to find your individual hardware’s unique "safe" voltage limit is useful, when squeezing every last reliable speed step-up out of our of your hardware.

When we overvolt, we rely on the integrity of our hardware. We assume that it is of high quality, without manufacturing defects, and that it is capable of withstanding a large overvolt without sustaining damage. Most of the time, this is the case. There are always exceptions. It is speculated that hardware which is damaged through overvolting, already contained small defects or weaknesses that were only made apparent, or made worse, when the hardware becomes strained through overclocking, and overvolting. There is no accurate way to tell if your specific piece of hardware has such flaws. Sometimes, we get unlucky. For the most part though, ‘healthy’ hardware is capable of running smoothly for years and years with a considerable overvolt in place.

Overvolting is dangerous, and inherently risky. As soon as you overvolt your computer hardware, you void your hardware’s warranty, and run the risk of sudden hardware death. That risk is small if you are cautious and smart with your overvolting, but it is always present nonetheless. The words overvolt and safe should really never be used in the same sentence. For this guide, I will refrain from saying safe, and will use "safe" instead – there are always risks, make no mistake.

Speaking from a personal perspective, I have overvolted almost every single piece of transistor-based and overclockable hardware that I have ever owned. I have never had a hardware death as a result of overvoltage (although I have killed more than my fair share of computer hardware...), and barring bad luck, you shouldn’t either if you take your time, and are careful.

More voltage has the nasty side effect of creating heat. Our transistor based computer hardware does not do well with high temperatures, and can sustain damage from lengthy exposure to high temperatures. The "safe" temperature to run our hardware at is theoretically directly related to how much voltage you are giving your hardware – the larger the overvolt, the cooler you should be keeping your hardware in theory. There is certainly no number or ratio set in stone for this – common sense is your only reliable ally – "if it’s stable it’s safe" is an excellent guideline. Stability is a reliable and consistant standard to use for the measurement of safety, which is not a "variable" that is measureable directly.

Overvolting also has the effect of increasing the rate of electromigration within our hardware - both high current density and high levels of heat will accellerate the rate of electromigration (and both of these variables are directly increased through overvoltage).

Electromigration increases the resistance of the metal interconnects (miniscule wires and contacts - the conductive signal pathways inside of our transistor-based processors and memory) within our transistor-based hardware, which can mess up processor operation. Remember the transistor tolerances mentioned in the first section (If you didn't read the first section, don't sweat it)? When signal resistances within our transistor-based hardware are increased, our signal strength can fall outside of our hardware's transistor VDD tolerence as a result. Electromigration can also eventually cause interconnects to break entirely - and a broken interconnect pretty much means a dead piece of hardware (you cannot send a voltage high signal down a broken interconnect!). Electromigration is a long-term issue, an increased rate of electromigration means a shorter lifespan for your hardware. Electromigration is not a huge issue with modern hardware due to the high quality of the materials used in the construction of (most) modern hardware - although it apparantly does still occur, albeit at a slower rate. A very detailed and highly technical explanation of the electromigration effect can be found HERE, (thanks Gnufsh for the excellent resource!).

When overvolting, it is important that one has adequate cooling on their hardware. A good rule of thumb is "voltage doesn’t kill hardware, heat kills hardware" (and likewise, "overclocking doesn't kill hardware, heat kills hardware"). This simple rule, of course, is only applicable if one overvolts their hardware within reason, "reason" being actively paying attention to what your hardware tells you through its responses to overvoltage.

All the doom and gloom done with, lets move on to our last and most interesting section of the guide, Overvolting Technique in Overclocking.

[felinusz]

THOROUGH AND "SAFE" OVERVOLTING TECHNIQUE

"1.8V is the maximum safe 24/7 overvolt for your processor."

I see this, or comments like it, quite a lot. It is a common misconception that each specific type of hardware has a set-in-stone overvolt that one cannot safely exceed. This is very much untrue. No two processors are alike. Just as two ‘sibling’ processors coded one digit apart in their batch will overclock to different levels, those two ‘sibling’ processors will react differently to overvoltage.

Our hardware is unique, and needs to be treated uniquely, on a case-by-case basis. Going about this is actually quite easy, and very simple, if somewhat time consuming.

It’s all about two things. Stability, and Diminishing Returns. These two factors are the holy grails of overvolting. By paying attention to both, we can chart out and determine our specific and unique hardware’s unique "safe" 24/7 overvoltage limit.

Please keep in mind that the following is to be used as a guideline; those who are adventurous or suicidal, and those who are cautious, may respectively choose to adjust the following technique to suit their personal comfort levels. The following technique is independant of temperature and cooling – temperature is a variable which directly leads to both Diminishing returns and Instability, and as such it is taken into account, although indirectly. I am in no position to tell anyone specifically what is a "safe" temperature to run their hardware at. However, thorough use of the following guidelines will invariably leave you at a "safe" temperature by default.


First off, lets see how high we can overclock our piece of hardware, with some degree of overclock stability, using stock voltage – no overvolts just yet. Thorough stability testing is not really necessary at this point, as we are only doing some preliminary probing into our hardware’s capabilities. A quarter hour run of Prime95 or a similar stress testing program for a processor, a quarter hour run of 3DMark for a GPU/GDDR, or a quarter hour run of memtest86 test #5 for memory, is sufficient at this stage.

~ Scale your clock frequency higher in small steps. For modern processors, ~100 MHz steps are appropriate. For modern RAM, ~10 MHz steps are appropriate. For modern GPUs/GDDR, ~15 MHz steps are appropriate. The step size is not particularly important.
~ After each speed ‘step’, run a quick stability test to make sure that your overclock has some integrity.
~ When you get to the point where stability is compromised, begin to ‘fine-tune’ the overclock. Drop your speed step size, and find a rough stable overclock limit.
~ Write down the ‘final’ overclock, and the stock voltage used.

Now that we’ve done some initial probing, we can heat things up a little bit, and add some voltage. Overvolt the hardware in question by the minimum voltage increment available in the BIOS, likely 0.025V for processors, and 0.025V-0.1V for memory. If you are overvolting through use of a physical voltage modification, keep to tiny 0.025V overvolt steps. The smaller the voltage step, the more accurate our findings will be, the more time consuming the process.

~ Starting from the clockspeed we left off at after testing at stock voltages, ‘fine tune’ the frequency upwards in small steps, as before.
~ After each step, run a quick stability test to check for overclock integrity.
~ Continue untill you lose stability.
~ Once stability has been compromised, fine-tune the overclock to the absolute limit point where it can run with stability for 15 minutes.
~ Write down this rough ‘final’ overclock, and the voltage used.

We now continue with the above steps, incrementally increasing our overvoltage, and charting out the clockspeed gains which we see at each overvoltage step – go until you have completed four steps, including stock voltage. This will take some time, but it’s worth it.

After four voltage and clockspeed steps, it is time to start a graph. A piece of graph paper and a pencil, or graphing software, are all you need to do this. A chart with "VOLTAGE" for the horizontal X axis, and "CLOCKSPEED" for the vertical Y axis is appropriate. Use stock voltage as the voltage starting point, and the maximum rough stable overclock at stock voltage as your clockspeed starting point.

Chart out your results thus far. Can you see a curve yet? After four small incremental voltage and clockspeed ‘steps’, we start to get an idea of how our hardware is reacting to overvolts. Some hardware will already start to peak after four steps. Other hardware is just getting started, hungry for more. Every piece of hardware is different, which is why this graph is so important – on a piece of paper your hardware’s unique reactions to overvoltage are fully outlined.

Right now is where we need to start paying attention to our gains, looking for diminishing returns. This isn’t too difficult with a graph right in front of us! When your curve begins to taper off, and flatten out, diminishing stable MHz returns per mV overvolt have kicked in. This is a good point to stop overvolting, when looking for a "safe" 24/7 overvolt and overclock.

Keeping an eye on your graph as you go, continue upwards in small overvoltage and clockspeed steps, until such a plateau becomes apparent on your graph. At this point of diminishing return, we can 'fine tune' our overclock for stability, for 24/7 use. Below, I have attached a sample graph, outlining this peak with a fictional (and conveniently clear) example.

Going slightly past the point of diminishing returns is certainly not "wrong", although the risks of both long term and short term hardware damage increase significantly when once does so. For hardcore benchmarking, suicide screenshots, and crazy fun, the graph is somewhat irrelevant. The graph guideline is an excellent tool for finding a "safe" 24/7 overclock/overvolt for your unique hardware, it is not so useful for the benchmarker or record breaker.

If your hardware sees stable MHz gains from seemingly large overvolts without peaking, do not be afraid to continue. Your hardware will tell you when it has hit its limits – it will peak or become unstable. High temperatures will directly cause both of these situations.

Overvolting far past dimishing returns will take one to the point where the negative impact of increased temperature as a result of voltage, will outweigh any gains from overvoltage in the first place. The curve will start to head downwards. When one gets to this point, they have entered the ‘death zone’.


I hope that this guide will prove useful to some of you. Please remember that all of this is merely guideline, not law - not even approaching it. It is one possible method of finding "safe" overvoltage limits, one which I have found to be extremely effective, safe, and efficient through personal experience.

Keep in mind however, that this technique is thorough and "safe", in the sense that it is based off of your unique hardware’s reactions to overvoltage. I strongly believe that this technique for finding your hardware's overvolt limit is as good as any other out there.

Here is a sample gains graph, with a clear point of dimishing return.

[SolidxSnake]

wow.

nice guide..

STICKY!!!

[Gnufsh]

You don't mention electromigration, which kills cpus and is accelerated by higher voltage. It is less of a problem not that most processors use copper interconnects, rather than aluminum ones. Anyway, here is a good link (with pictures!):
http://www.csl.mete.metu.edu.tr/Elec...ation/emig.htm

[felinusz]

Thanks SolidxSnake - hope it's useful and informative

Gnufsh - cool, thanks for the link! Quite a thorough expanation on that page, that's a pretty serious site . I didn't know that electromigration was still a serious issue with modern processors though.

[subtotal]

stickey!

[RJARRRPCGP]

Originally Posted by Gnufsh

You don't mention electromigration, which kills cpus and is accelerated by higher voltage. It is less of a problem not that most processors use copper interconnects, rather than aluminum ones. Anyway, here is a good link (with pictures!):
http://www.csl.mete.metu.edu.tr/Elec...ation/emig.htm

I'm having a DNS error with that link.

[zexmarquies01]

felinusz, i have read many methods of overclocking. From small one paragraph explanations. to a couple paragraphs.

But your explanation, and method by far is the best i have ever read.

It easily tells people to go in steps, and carefully watch what they are doing. you explain about diminishing returns, and how and why thats a good place to stop.

I also purpose this thread to be made a sticky. Very informative, and Easy to understand. as soon as i'm done with this post, i'm going to PM a mod in this section, and ask them to sticky this.

while my own opinion on this forum of all these people is very small, I don't recall every purposing something to be posted as a sticky before. And i KNOW i have never PM'ed a mod at the idea either. And while my voice may not be much...the topic must be good for someone who never asks for anything be a sticky, for a thread to be done so. heh

[Gnufsh]

Originally Posted by felinusz

Thanks SolidxSnake - hope it's useful and informative

Gnufsh - cool, thanks for the link! Quite a thorough expanation on that page, that's a pretty serious site . I didn't know that electromigration was still a serious issue with modern processors though.

Like I said, it's less of an issue since the switch to copper interconnects, but I think it can still be a problem.

[Zulu-1]

very nice guide
STICKEY!!!!!!

[Electron Chaser]

felinusz, Excellent post!!!

and a very nice read.

This should be a sticky

[felinusz]

zexmarquies01

felinusz, i have read many methods of overclocking. From small one paragraph explanations. to a couple paragraphs.

But your explanation, and method by far is the best i have ever read.

It easily tells people to go in steps, and carefully watch what they are doing. you explain about diminishing returns, and how and why thats a good place to stop.

I also purpose this thread to be made a sticky. Very informative, and Easy to understand. as soon as i'm done with this post, i'm going to PM a mod in this section, and ask them to sticky this.

while my own opinion on this forum of all these people is very small, I don't recall every purposing something to be posted as a sticky before. And i KNOW i have never PM'ed a mod at the idea either. And while my voice may not be much...the topic must be good for someone who never asks for anything be a sticky, for a thread to be done so. heh

Wow, thanks a lot dude, that is really encouraging to hear. Guys, thank you for the kind words.


Gnufsh, I added Electromigration to the "Dangers" section, with the most excellent link that you posted

[dAvies.lOcker]

A very imformative read - thanks!

This has to be sticky material.... thanks again

[jack222]

Thanks for writing this up.

No need to even say this has to be a sticky

[donny_paycheck]

Wow, excellent post. This is sticky stuff for sure!

[felinusz]

Thanks a lot Donny!

[wfarid]

wow, this is a great sticky, i think maybe one of the simpliest and to the point stickies I've ever read. i learned something today, someone want to give me an apple?

note: macs aren't edible (although that'd be pretty cool if they were)

[fAlCoNNiAn]