Overclocking in theory..

Hi guys,

All of us know that better cooling on your CPU and raising the operating voltage allows you to raise processor's clock frequency which makes it run faster. All this is called overclocking.

I wonder how many of us really know what happens when we cool down the silicon as cold as possible and why does this allow us to run the CPU faster.

Why I started this thread is because many IT sites have recently wrote news about our liquid helium experiment and many people have posted comments like: "Why does is need to be that cold? Isn't keeping the temperature at 0C enough?". Some people have even tried to explain the theory behind all of this but so far I haven't seen one very good and definitive answer. How deep we have to go? Atom level?

Please post if you have a good knowledge about this topic or know good articles, documents or books about topics related to overclocking or methods / laws of physics. Maybe we can combine a simple but still fairly in-depth answer what is overclocking so that everyone can easily understand why we are cooling our CPUs to extremely cold temperatures.

Let me know what you think?

I've always thought that cold temperatures was just a mean to an end. The colder the object, the more volts it could tolerate. I know thats over simplifying it but whatever. Did you guys have anymore success using helium over nitrogen with the relatively same volts? I'm not educated enough to go in depth, though I am trying to go through the steps to do so, but that's always just been my outlook.

In fairness Sampsa I'm more impressed that the AMD still functioned as a cpu at those temperatures! A few things to note:

• Lower temperatures = lower resistivity. Lower resistance allows for more current to pass along a given cross section of copper
• Switching Frequencies. When a transistor switches there are losses involved in the form of heat etc so as you up the frequency you up the heat load from switching and heat = increase in resistance resulting in more losses
• Exotic materials properties at cryogenic temperatures. I've no idea what AMD are currently using in their silicon lithography processes these days but the electrical behavior of materials at cryogenic temperatures can be remarkably different from that at room temperature.

Another thing is as we increase these frequencies into the GHz the usual digital on/off square wave basically turns into an analogue wave as the ability to charge/discharge the capacitive load fast enough to get a straight edge just cannot to achieved. I'd say the circuits ability to distinguish this level change will also dictate an upper limit to frequencies.

Johnny, great start. I'm sure I will be able to get some details from straight from AMD and I hope we get some comments from Intel side.

Quote:

---

Originally Posted by filmbot

Did you guys have anymore success using helium over nitrogen with the relatively same volts?

---

Yes, with same vcore switching from LN2 to LH helped to raise the clock frequency 150 - 200 MHz. So just getting the silicon colder helps.

Johnnys answer was awesome:

# Lower temperatures = lower resistivity. Lower resistance allows for more current to pass along a given cross section of copper
# Switching Frequencies. When a transistor switches there are losses involved in the form of heat etc so as you up the frequency you up the heat load from switching and heat = increase in resistance resulting in more losses.

Less resistance allow more current and more clear signal. Higher switching speed allow to do more operations / calculations in certain time (higher clockspeed).

This is the simple answer, but I would like to hear really good and in depth answer from some guru.. like DRwho or some engineer who works with these parts.

I've always wondered if (excluding cold bugs) if there is not a best temperature at which to overclock a given processor at a given voltage. I know typically the colder the better but maybe there is a sweet spot, this may be something worth investigating with these AMD chips.

A very cool idea. Im no engineer or guru by any means, but I also thought that extreme cold limited electron migration.

Quote:

---

Originally Posted by fiveprime

I've always wondered if (excluding cold bugs) if there is not a best temperature at which to overclock a given processor at a given voltage. I know typically the colder the better but maybe there is a sweet spot, this may be something worth investigating with these AMD chips.

---

AMD sweetspot is 1°K :p:

I work in a team at the materials lab here in the uni aand at our meeting today brought this up as an end discussion :) here's what we decided:

• limiting factors of making a transistor switch faster basically boil down to your drive strength and the capacitance.

• The drive strength we can adjust with the Vcc aka vcore

• The capacitance goes down with transistor size so the smaller the process the theoretically faster the transistor can be charged/discharged

When temperature was brought in it proved quite interesting. From a bulk view the lower the temperature the less electrons will be shooting off in random directions colliding and generally not flowing smoothly (resistance). Again the faster the switch time the bigger the current spike and the greater the thermal losses for that switch so excess heat will be an issue, one that cooling would take care of. Also we can get technical an invoke some transistor equations to relate kT to drive strength but it's a bit obscure to directly relate it to increasing the frequency. I'll hopefully get the equation rewritten in terms of temperature soonish :)

Johnny has right.
Some more:
The main guiltys are noises. When bus or transistor is working faster then normal operation speeds, there is much more noises.
There is something like heat noise. It happens when temp. are high, and it is lower when we lower temps.
Eg. when we adjust some voltage we are trying to avoid this noises, because there is higher potential, and heat noises causes by heat (because of current and resistance which generate heat) are relatively lower, and then devices can understand where is logical 1 and 0 (when noises are present, its difficult to recognise where is 1 and 0.

Quote:

---

Originally Posted by Kean

Johnny has right.
Some more:
The main guiltys are noises. When bus or transistor is working faster then normal operation speeds, there is much more noises.
There is something like heat noise. It happens when temp. are high, and it is lower when we lower temps.
Eg. when we adjust some voltage we are trying to avoid this noises, because there is higher potential, and heat noises causes by heat (because of current and resistance which generate heat) are relatively lower, and then devices can understand where is logical 1 and 0 (when noises are present, its difficult to recognise where is 1 and 0.

---

Thermal noise - that was the term I was looking for! :D

good catch Kean :up:

A few years ago i wrote up a wiki on overclocking thermodynamics and electrical properties due to the massive amount of questions from my 8GHz project seen over here: http://overclockingwiki.org/index.ph...iquid_Nitrogen

WorldRecord Overclocking 101

Xelmon wrote: "I could see voltage limiting, because mobos can only do so much.

The heat, @ -15C though, I really cant see it. Care to explain?"
[edit] Temperature

-15c only feels cold in relation to our normal body temperature, the real point at which something is cold is absolute zero or -273C anything above that is simply heat. and most materials begin to show signs of superconducting properties around absolute zero. superconducting is the ability to transmit large amounts of power at very high voltage with out the risk of burning up the lines or having any migration of the electricity (jumping out of the cable into near by objects)

so the colder you get a CPU in theory the more voltage you can pump into it with minimal damage to the circuits.

for example if you were to apply 1.9 Volts to a CPU (lets say a P4 631) with stock cooling the CPU would overheat very fast by overcoming the thermal capacity of the stock cooler. your CPU has a device in it (a thermal couple or diode) that will sense when the core temperature exceeds its safe zone (about +100C) this will tell the CPU to shut down in order to prevent fatal damage.

So why not put a really big copper cooler of dewm on it and pump 1.9 Volts through her?

If you manage to keep the CPU below its thermal threshold using water cooling or Air (lets say we have her at 1.9 Volts and its a nice 30C) an evil bugger known as Electron Migration will come into play.
[edit] Electron Migration

is something Chip manufacturers have been battling since the dawn of time. When you have a very small and very complex electrical component such as a CPU both heat and voltage are directly responsible for instability and damage to the component.

When we look a the very inner working of a CPU it is similar to the traces of a motherboard but on a Scale of (lets stick with the 631) 65nm meaning the transistors that open and close billions of times per second in effect computing lines of code (so you can have your Uber Interweb and... stuff) are actually "65 Nano Meters" wide! this means really really small. well because they are so small there is also not much distance between the "traces" and transistors.
[edit] Thermal Electrical Resistance

When you heat something up you are in effect adding to (in this case) the electrical resistance (meaning electricity has to push harder to get to the other side) so when we think of a CPU if you were at a theoretical (lets forget "minimal kinetic energy" for a moment) absolute zero (-273c) electricity would have no problem moving through the "maze of traces" in a CPU and if we were at say +100C electricity would have to work a lot harder to make it through the "maze".

With electricity being the lazy creature it is, its always looking for the easy way out. (remember those two words Electron Migration?) If two traces are very close to each other in the core of a CPU and one is in essence "Closer to the maze exit" and Mr. electricity is working hard to run through at +100C he will simply cheat and migrate (Jump) over the insulator (wall) to the other trace this causes the CPU to destabilize resulting in a flawed computation ending in a blue screen or shutdown.

Imagine there are hundreds of millions of transistors in a CPU core that is about 1x1cm x 1mm thick and when the migration starts there are usually thousands jumping over causing fatal errors and extreme instability.

The damage electron migration can cause is the death of your CPU or a permanent return of corrupted information. This happens when a Higher voltage trace jumps into a near by lower voltage zone due to its lower electrical resistance and the result is a fried portion of the chip rendering it useless...
[edit] Hardware

If we want to say make our CPU (631) run at 8GHZ we have to take into account every thing listed above and a few more things like our budget Sad !st off we know we need to keep it cool, so we get a nice Team NexGen Rev 2.5 Copper Liquid Nitrogen Pot and a frosty 160L Dewar of LN2 from the local A-L compressed gas company. we solve the problem of Water Condensation with nice Neoprene tape and heating elements. We have a good ole NexGen Full modded ASUS COMMANDO board capable of supplying more voltage than needed. we have a CPU that is a good steeping and can take us to 8GHZ and the Micron D9GKX that can handle at least 1200MHZ at 4-4-4-12 @ 2.7V we have a nice PSU that can push 800W of clean power. a crappy GPU that is PCI (less stress on the FSB) and pulls minimal power from the board.

so now we know why we need the CPU cold but we don't know why we need the extra voltage
[edit] Voltage

The CPU speed is a Frequency! measured in Hz, or kHz (kilohertz, 10^3 Hz), MHz (megahertz, 10^6 Hz), GHz (gigahertz, 10^9 Hz) & THz (terahertz, 10^12 Hz) (thanks wiki). (Hertz not the car place, but rather cycles per second) now in order to have a CPU operate at a frequency and be stable enough for use we have to match a voltage (voltage is acts as the stabilizer for the CPU speed) to carry the signal through the CPU and successfully compute and output information. Intel does a good job of this when developing faster chips they spend a great deal of resources finding the optimal voltage for a set frequency by testing possible environmental temperatures (home, office, server room....) so if we are to increase the frequency of a CPU we need to in crease the voltage to stabilize the faster signal but with the increase in voltage we have an increase of heat in the CPU and heat will destabilize the CPU with electron migration. when you double the Voltage of a CPU you quadruple the heat generated.

This is where the LN2 comes into play, If we want a speed of 8GHz or 266% of the stock 3GHz we need a lot of voltage. and to prevent the CPU from heating up and causing electron migration we cool it to about -150c (LN2 is -196c) this allows the CPU to operate with less electrical resistance and hopefully prevent migration when running such a high voltage like 1.9 Volts.


The colder you can get the CPU the more voltage it can take allowing a higher clock frequency.

Quote:

---

Originally Posted by TheKarmakazi

A very cool idea. Im no engineer or guru by any means, but I also thought that extreme cold limited electron migration.



AMD sweetspot is 1°K :p:

---

below 17K silicon starts to become a superconductor, So i would say to have an effective cooling solution/overclock around 17K would be prime :)

I wrote this on another forum a while ago, an extremely rough conceptual overview relating temperatures and voltage to the rise and fall times of transistors charging and discharging the parasitic capacitances at a given node. The standard model is to approximate a transistor as a resistor and a capacitor. This resistivity is inversely proportional to electronic mobilty, D*q/KT, where D and K are physical constant, q is electronic charge and T is of course temperature. So, in a nutshell, lower temperatures -> lower resistivity -> quicker charging/discharging of capacitances -> lesser delay in producing a "1" or "0" -> higher stable frequency.

Great info V2-V3 & Gautam ! :up:

and I was j/k about the 1°K. Team Finland did have their cpu at a frosty 31°K with the LHe though.

Quote:

---

Originally Posted by V2-V3

The damage electron migration can cause is the death of your CPU or a permanent return of corrupted information. This happens when a Higher voltage trace jumps into a near by lower voltage zone due to its lower electrical resistance and the result is a fried portion of the chip rendering it useless...

---

IMHO electromigration is almost completely irrelevant to the discussion at hand. Sure, colder temperatures keep the transistor gates intact for longer, but that has little to do with why it allows them to switch faster. Though, semiconductor physics is my weak spot.

Tighter lattice due to shirnkage from the cold? Just simply lowers resistance, or what?

This might be slightly off-topic and I apologize in advance, I completley agree with the preceeding posts but how about the power coming from wall outlet?? What kind of power are we feeding our gear in the first place?? I used to live in Perth Oztralia and the old power grid was horrendous, spiking up and down, (215V daytime, 245-257V @ night) Being a bit of an audiophile I installed 2 dedicated 20A lines run on their own phase (old 3 phases house) and the improvement in sound quality was very noticable, even to my Ozzie pals that thought that much stereo gear was crazy in the first place. We always listened at night. Zero background noise, deep open sound stage.. imagery like no tomorrow, etc. etc. Then I bought a swanky power conditioner and took it to another level as well. I'm now living in Belgium, power supply is more stable but noisy. I now have 2 25 Amp dedicated lines for the stereo and I'm thinking of yanking out my Power Conditioner and plugging the mother board into it and start with nice, clean unrestricted power. It's just like putting better fuel in your car!! This should eliminated even more potential scatter, jitter, etc. and allow for better current to the board. I think I've even seen a few power conditioners, stabilizers in some of the high end overclocking pictures.

Quote:

---

Originally Posted by Gautam

.... This resistivity is inversely proportional to electronic mobilty, D*q/KT, where D and K are physical constant, q is electronic charge and T is of course temperature....

---

Yep that's the equation I was looking for Gautam :up:

Yes...there are a couple of long equations which most computer/electrical engineering students are forced to become intimate with for the different modes of operation of a transistor. :p: (I would rather not recount them here, but you can probably find them by looking up "unified model" if you really wish) Suffice it to say, the important ones in question are directly proportional to electron mobility.

wow guys this is an xtremely interesting thread, this is definitely the reason why I chose to switch to EE a week ago and I cant wait to get to some higher level classes, thats for sure.

You won't be able to wait till they're over once you get to them. ;)

Quote:

---

Originally Posted by SNiiPE_DoGG

wow guys this is an xtremely interesting thread, this is definitely the reason why I chose to swtich to EE a week ago and I cant wait to get to some higher level classes, thats for sure.

---

soo troy NY does that mean you go to RPI? :D



on a more relevant note one of my professors actually brought this stuff up in class asking if anyone had heard about the cpu being run under liquid helium. He was actually curious about how it ran at such a low temperature. Now i wish i had asked him more about it.

Quote:

---

Originally Posted by Wiggy McShades

soo troy NY does that mean you go to RPI? :D



on a more relevant note one of my professors actually brought this stuff up in class asking if anyone had heard about the cpu being run under liquid helium. He was actually curious about how it ran at such a low temperature. Now i wish i had asked him more about it.

---

yes I am currently a freshman @ RPI :) next semester is when it gets interesting though.... right now I'm just taking chem and phys 1100 :\ I know the subjects completely and fully but my HS sucked so bad I couldn't take the AP's to credit out. :rolleyes:

Quote:

---

Originally Posted by SNiiPE_DoGG

yes I am currently a freshman @ RPI :) next semester is when it gets interesting though.... right now I'm just taking chem and phys 1100 :\ I know the subjects completely and fully but my HS sucked so bad I couldn't take the AP's to credit out. :rolleyes:

---

I almost went there. I'm at BU computer systems engineering :rocker:...... EE is for girls :p:

Quote:

---

Originally Posted by Wiggy McShades

on a more relevant note one of my professors actually brought this stuff up in class asking if anyone had heard about the cpu being run under liquid helium. He was actually curious about how it ran at such a low temperature. Now i wish i had asked him more about it.

---

Cool! He was interested about "the Experience". You can tell him to contact us personally :D

Thanks Gautum and other guys about good info! I know this is exactly what Sampsa is looking for :up:

Quote:

---

Originally Posted by Wiggy McShades

I almost went there. I'm at BU computer systems engineering :rocker:...... EE is for girls :p:

---

Haha too bad I suck at programming.... or I would still be CSE. thing is that I'm an extremely visual person and software and I don't always get along.

Snipe, im with you there, programming just doesnt seem to work whenever i dot i!
This is great to hear though, I had heard that the lower temps allow greater frequencies at the same voltages (and I have hoped, considering the highest on air I have got with my e7300 is in my sig, and that took 1.7v!)
And so I was worried that there wasnt much point in using DICE with this cpu, (would be my first time)
However, reading stuff like this makes me realise that not only would I be able to use 1.8, 1.9 volts without it being too hot, due to the lack of heat, I will be able to get higher frequences at the volts I have already tried, (must admit, have tried 1.8v!)

George

Quote:

---

Originally Posted by fiveprime

I've always wondered if (excluding cold bugs) if there is not a best temperature at which to overclock a given processor at a given voltage. I know typically the colder the better but maybe there is a sweet spot, this may be something worth investigating with these AMD chips.

---

So far with Phenom II X4 it looks like the colder the better. We saw -242 Celsius measured from base of F1 EE copper container (1 - 2 mm from bottom surface):

http://plaza.fi/s/f/editor/images/X-...5515956225.jpg

http://plaza.fi/s/f/editor/images/X-...7475236927.jpg

and around -230 measured from AMD Pete Hardman's container where thermal sensor is milled right in the middle of bottom surface so it is basically touching CPU heatspreader. Under load temperature was -225 Celsius:

http://plaza.fi/s/f/editor/images/pete_sensor.jpg

Since the system was running just fine at these temperatures I wouldn't be surprised if Phenom II X4 could work at -250 and -260C range. Next step is to design more sophisticated solution to use liquid helium to cool down the CPU to even lower temperatures than we achieved in Las Vegas.

Quote:

---

Originally Posted by Sampsa

Since the system was running just fine at these temperatures I wouldn't be surprised if Phenom II X4 could work at -250 and -260C range. Next step is to design more sophisticated solution to use liquid helium to cool down the CPU to even lower temperatures than we achieved in Las Vegas.

---

what do you think that would be? is it possible to set up a closed loop of pressurized liquid helium to make it colder?

Quote:

---

Originally Posted by SNiiPE_DoGG

what do you think that would be? is it possible to set up a closed loop of pressurized liquid helium to make it colder?

---

Yes or "double container" where is inner pot filled with liquid helium and some kind of lid adding pressure and outer pot filled with liquid nitrogen.

This explains a whole lot thanks! Always wondered why temperatues mattered so much when clocking, I only knew it did matter!

Nice discussion :up:, going to XOC Q&A section and sticky :yepp:

Quote:

---

Originally Posted by cadaveca

Tighter lattice due to shirnkage from the cold? Just simply lowers resistance, or what?

---

I suck at chemistry, hopefully someone else can pipe in with details at the molecular level. I only learned up to the equations. :p: :sofa:

it seems you gave the answers, yes two main thing we have for overclocking,

1.increasing voltage

2.cooling down

1. is definetly for keeping 1-0 level when you increase clock frequency, and 2. is for higher electrical conductivity of semiconductors under cold which allows transistors to switch faster with lower potential difference.

but if we came to coldbug again, that is usually complicated,but here my theory on this is that;
yes, lower temp means higher electrical conductivity but that is the case for semiconductors only, if we look at metals or other materials side it is reverse, when you heat you got higher conductivity, so what i want to come with is that somehow metal molecular side of the chip may cause coldbug since its conductivy is effected by extreme cold. yea this is theory, some may prove or not by giving how the metal parts of chip acts :rolleyes:

Gautum,
How does th increase of voltage help reduce the high low delay, and therefore allow higher clockspeeds, is it just because more volts will cause the transistors to switch more quickly?

There are two basic reasons why increased voltage lets the caps get charged faster and the transistors switch quicker. First of all, the increased supply voltage increases the drive current across the transistor, so it charges it's own output cap quicker.

Secondly, all voltages are increased, so the input voltage is increased as well. In my crappy diagram I idealized the input into a perfect square pulse, but in reality, the input is coming from another transistor's output itself, so the input also rises and falls for some given period of time, like below, so increasing the input voltage at the gate allows the transistor to start switching faster as well.

but is it really the voltage increase, or the current increase that allows faster switching?

Or is the voltage a sneaky way of bypassing current limits?

Well higher voltage comes back to drive strength.

There are minimum setup and hold times for stable clock switching.

There are "on" and "off" thresholds for a clock signal. E.g.... 67% (or more) of max amplitude is "on" for the rising clock and 70% of max amplitude is "off" for the falling clock

Higher voltage to the clock gives it a higher maximum amplitude.

rising and falling edge clock transitions are not vertical lines with respect to time- they have a slight slope- a very steep one, but a slope all the same. They take a finite time to change

Making the max amplitude bigger but keeping the low-high transition time the same means that the signal will pass the "on" threshold faster and will be a more stable valid clock.


There is also a mean time to metastability which has a temperature dependence, as well as reliance on specific device constants. Metastability drawn as a clock means the signal meets itself coming back and cant decide if its on or off.....so falls over.
I should add some diagrams to show this but im too tired... if no-one beats me to it, I might give it a shot tomorrow.

It would be interesting to design a chip that would not compare on and off with set levels, but by the change from the previous step, this could possibly add more overclocking head room, although you would have to set the change threshold to a specific level because I would imagine that it would be unreliable after a point.

Quote:

---

Originally Posted by cadaveca

but is it really the voltage increase, or the current increase that allows faster switching?

Or is the voltage a sneaky way of bypassing current limits?

---

As far as parasitic capacitances need charging, increasing one or the other while holding the other constant would make it happen quicker.

From the standpoint of the transistor, voltage in fact is at first directly proportional, and then a square factor for the drive current as it changes operation modes. You can check this out if you want, slides 5-8 are relevant.

Yeah, I understand that, I guess my question would be more directed at what point is it current, rather than voltage, that is needed, but then, I suppose that would depend on the resistance of the transistor. Cooling the cpu lowers the resistance, and by my mind, by better aligning the crystal's lattice, allowing for higher clocks at the same voltage.


Due to variations in the lattice structure, the current, and voltage, required for a specifc clock speed changes from chip to chip.



:shrug:.

So who's go the transistors that require less current to run...AMD or Intel? Given AMD specs Phemon 940 .85-1.5v, I would think AMD...but I dunno, becuase the work they do is slightly different.

What if instead of a defined voltage for "on" and a defined voltage for "off, the switch would be based on the change in voltage between each cycle, this could possibly allow for "looser" cycles and improve clockability.

That's known as sense amplification and is basically necessary in RAM to be practical.

awesome Thread guys !!! Very nice explenations. This answer some questions that keep me awake @ night.... :up:

Reducing temperature has different effects on the conductivity of different materials in the CPU. It will increase the conductivity of the copper and silicon, but decrease the conductivity of the silicon dioxide dielectric. The latter reduces current leakage and helps the system tolerate higher voltage.

When will new semiconduter materials be required? In other words, are we at the boundaries yet of silicon? It seems as Guatam explained it, there will eventually be a limit of how fast the CMOS switches between PMOS and NMOS generating the 1-0 signal neccessary. Yet with AMD recently pushing past 7ghz on "old" technology, it seems like it is more in how effecient the rest of the system is and not neccessarily all on the switching of the signal inside the processor itself?

What did Intel pull off with SB that allowed the discharge of electron to happen quicker at "normal" operating temperature? I.E. 5ghz possible without increasing electron conductivity greatly through temperature?