increase voltage in ram damage?

[SirKronan]

Unless you can qualify this statement with something more than "happy" and "not happy", kindly stop blowing smoke to make yourself seem intelligent. Start talking about current density, carrier lifetime and channel length and I'll start listening.

This is correct, for all intents and purposes, but doesn't explain why.

Marginally correct, but very poorly worded. The P4 was designed, IIRC, to "reach 10GHz", thus the insanely long pipeline design. The premise that the clock speed will continue to rise with more voltage and cooling breaks down, however, due to electromigration. The increase voltage causes a square-relationship in the amount of current, and eventually, the metal will move to due the current. This will occur far before 10GHz, however.

If you've read these forums for more than six picoseconds, you'd know that "smart overclocking" is something that nearly everyone here is aware of and practices normally. I really don't see how this new "Intel smart overclocking" has much to do with memory, which is what the OP was wondering about, also which in case you didn't know is manufactured with a *completely* different process than an intel CPU.

This makes no sense at all. You're missing words .. and content. What is the new architecture limited by? The articles don't particularly say, at least not on the pages you linked. What is the manufacturing process limited by? The article doesn't mention that at all, and I highly doubt they would even know. How is the number of transistors related to function, process or design? You can group memory and CPU together *very* loosely by categorizing the number of transistors. After that, your comparison falls apart.

You can overclock anything as far as you want. It won't work right after a certain point but you won't break it. Extremely overvolting and poorly cooling your hardware will cause it damage. All semiconductors are directly affected by heat, voltage and power delivery; qed your premise is completely incorrect. For the purposes of overclocking, memory speed is limited by heat, voltage and design, in increasing order of relevance. Process size has very little to do with it, as memory hasn't gotten to 65nm yet so that comparison is utterly irrelevant. In fact, due to its very nature, process size for memory is nearly irrelevant. Most current memory I believe is at 110 or 90nm. There is no need for anything smaller as if yet.

Thank you for explaining what I was getting at, only way better and more eloquently than I ever could have.... That's what I like about this place. People smarter than me that I can learn from! You're explanation makes a LOT of sense and I have personally seen empirical data to back up what you've said. Thanks for helping clear this mess up on this thread, which is, as you said, about memory, not CPUs.

 

[imzjustplayin]

Unless you can qualify this statement with something more than "happy" and "not happy", kindly stop blowing smoke to make yourself seem intelligent. Start talking about current density, carrier lifetime and channel length and I'll start listening.


This is correct, for all intents and purposes, but doesn't explain why.



Marginally correct, but very poorly worded. The P4 was designed, IIRC, to "reach 10GHz", thus the insanely long pipeline design. The premise that the clock speed will continue to rise with more voltage and cooling breaks down, however, due to electromigration. The increase voltage causes a square-relationship in the amount of current, and eventually, the metal will move to due the current. This will occur far before 10GHz, however.



If you've read these forums for more than six picoseconds, you'd know that "smart overclocking" is something that nearly everyone here is aware of and practices normally. I really don't see how this new "Intel smart overclocking" has much to do with memory, which is what the OP was wondering about, also which in case you didn't know is manufactured with a *completely* different process than an intel CPU.



This makes no sense at all. You're missing words .. and content. What is the new architecture limited by? The articles don't particularly say, at least not on the pages you linked. What is the manufacturing process limited by? The article doesn't mention that at all, and I highly doubt they would even know. How is the number of transistors related to function, process or design? You can group memory and CPU together *very* loosely by categorizing the number of transistors. After that, your comparison falls apart.




You can overclock anything as far as you want. It won't work right after a certain point but you won't break it. Extremely overvolting and poorly cooling your hardware will cause it damage. All semiconductors are directly affected by heat, voltage and power delivery; qed your premise is completely incorrect. For the purposes of overclocking, memory speed is limited by heat, voltage and design, in increasing order of relevance. Process size has very little to do with it, as memory hasn't gotten to 65nm yet so that comparison is utterly irrelevant. In fact, due to its very nature, process size for memory is nearly irrelevant. Most current memory I believe is at 110 or 90nm. There is no need for anything smaller as if yet.

All that can be summed up by "smart overclocking". And no, I don't believe everybody on this forum performs "smart overclocking" like the article was talking about.

 

[(cf)Eclipse]

All that can be summed up by "smart overclocking". And no, I don't believe everybody on this forum performs "smart overclocking" like the article was talking about.

you might be surprised, but either way, the method you use to overclock doesn't determine how your hardware will respond to the voltage and temperatures it runs at... nice job at making another totally irrelevant counterpoint!

 

[imzjustplayin]

you might be surprised, but either way, the method you use to overclock doesn't determine how your hardware will respond to the voltage and temperatures it runs at... nice job at making another totally irrelevant counterpoint!

I don't follow, if you have to increase the voltage in a non linear fashion, then anything past that point is going to more than likely result in electromigration. I don't believe that running all silicon at any speed in voltage will result in electromigration, at least at the rate most people are thinking in. I believe that there is a set point for diminishing returns in overclocking (like the article was talking about) where past a certain point for every increase requires double the amount of power. On the inverse of the equation where you're "underclocking" I believe there is a point where there are diminishing returns where the life of the silicon is increased.

They're both curves, if you were to look at each curve you'd find the life expectancy dropping exponentially when overclocking and the life expectancy increasing exponentially when underclocking. Within each curve you can pinpoint an optimal point depending on what is most important to you which brings up my point about "smart overclocking".

 

[Mike Clements]

it's not something that has a "limit", where if you break it, you start hurting things. just keep both temp and voltage as low as you can within reason when overclocking and you'll be fine

Well, an accurate point about "smart overclocking" was made back on page 1, read the quote above. So, it would appear that Eclipse was correct all along.

 

[(cf)Eclipse]

I don't follow, if you have to increase the voltage in a non linear fashion, then anything past that point is going to more than likely result in electromigration.

electromigration occurs at any voltage, just the rate increases rapidly as voltage and current demands rise.

I don't believe that running all silicon at any speed in voltage will result in electromigration, at least at the rate most people are thinking in.

it depends on the speed and voltage, obviously, which is the point i've been trying to make all along. the more voltage and heat, the faster the chip deteriorates. end of story, regardless of how little or how much you increase the voltage. opposite is true for undervolting, but as you say...

I believe that there is a set point for diminishing returns in overclocking (like the article was talking about) where past a certain point for every increase requires double the amount of power. On the inverse of the equation where you're "underclocking" I believe there is a point where there are diminishing returns where the life of the silicon is increased.

you're absolutely correct, but not for the reasons i think you think. keep in mind that there are several things that deteriorate and ultimately fail over a period of time. even if you get one thing way down, something else will fail before it, making it difficult to get a really long lifespan. it's easy to kill things quickly though

also, with memory, not only do we have to worry about the transistor failing, the capacitor is also a factor, which is what i believe to the primary reason why DRAM chips fail so much more often than things like CPUs

They're both curves, if you were to look at each curve you'd find the life expectancy dropping exponentially when overclocking and the life expectancy increasing exponentially when underclocking. Within each curve you can pinpoint an optimal point depending on what is most important to you which brings up my point about "smart overclocking".

indeed, but as i said, i'm quite sure most people do that. hell, i even make charts relating top stable speed, temperature, voltage, and estimated power draw, somewhat like this:

you're preaching to the choir here when it comes to "smart overclocking", since i practically did this as a living for a while

 

[Rock&Roll]

I use extra voltage only when I'm trying to milk out the last few megahertz needed to push my CPU to it's limits. Even then, I have to decide whether those few extra megahertz are really going to make much improvement in performance to make it worth the risk of component damage.

Extra voltage hurts some IC's more than others, partially do to the extra heat that results from the voltage. Therefore, the least you can do is get some fans blowing on your ram. From there, extra voltage is up to you.

 

[GilmourD]

electromigration occurs at any voltage, just the rate increases rapidly as voltage and current demands rise.



it depends on the speed and voltage, obviously, which is the point i've been trying to make all along. the more voltage and heat, the faster the chip deteriorates. end of story, regardless of how little or how much you increase the voltage. opposite is true for undervolting, but as you say...



you're absolutely correct, but not for the reasons i think you think. keep in mind that there are several things that deteriorate and ultimately fail over a period of time. even if you get one thing way down, something else will fail before it, making it difficult to get a really long lifespan. it's easy to kill things quickly though

also, with memory, not only do we have to worry about the transistor failing, the capacitor is also a factor, which is what i believe to the primary reason why DRAM chips fail so much more often than things like CPUs


indeed, but as i said, i'm quite sure most people do that. hell, i even make charts relating top stable speed, temperature, voltage, and estimated power draw, somewhat like this:

you're preaching to the choir here when it comes to "smart overclocking", since i practically did this as a living for a while

I knew a graph actually created by you would appear at some point.

 

[mwarps]

I don't follow, if you have to increase the voltage in a non linear fashion, then anything past that point is going to more than likely result in electromigration. I don't believe that running all silicon at any speed in voltage will result in electromigration, at least at the rate most people are thinking in.

Wow, talk about blowing smoke. These two sentences don't even make sense. Electromigration has nothing to do with voltage, and also has nothing to do with silicon, either. It is an effect in metal interconnects, specifically at interfaces.

Electromigration, as a whole, will not effect memory. Period. the J is so low that it will never happen. You will literally blow every cap and short every fet before you need to worry about one instance of electromigration causing failure in properly designed memory.

I believe that there is a set point for diminishing returns in overclocking (like the article was talking about) where past a certain point for every increase requires double the amount of power. On the inverse of the equation where you're "underclocking" I believe there is a point where there are diminishing returns where the life of the silicon is increased.

Thank you for stating a fundamental premise of hardware enthusiasm and basic electronic theory. We are in awe of this knowledge. Except there are no concrete qualifications in the statement. What sort of increase would require a doubling in power? Clock speed? Voltage? Phonon density? If so how much?

They're both curves, if you were to look at each curve you'd find the life expectancy dropping exponentially when overclocking and the life expectancy increasing exponentially when underclocking. Within each curve you can pinpoint an optimal point depending on what is most important to you which brings up my point about "smart overclocking".

Please feel free to link to either of these magical curves. Please then explain the mathematical origins of them. Why is the like expectancy decreasing when overclocking? Is it electromigration? Is it trapped charge in the gate? Is it diffusion?
Also, doesn't matter how low you drop clock speed or voltage, unless it's off. The device will eventually die a standard death, and there are simple models for it. Arrhenius is really tired of the hand-waving. Boltzmann is pissed, and Kilby.. Well, Daddy is laughing his ass off.

There is no such thing as smart overclocking. You overclock through experimentation and through finding the limits of the system. Period. An overclocker does not turn down the level to "extend your hardware's lifetime", not anyone that I know of anyway, you clock to stability or for benches. (EDIT, let me qualify that.. either all overclocking is "smart" or no such thing exists. There is a pretty set, standard way of doing it, and the way that was outlined in the article was the way I learned it a looooong time ago from people in these forums)

Trying to validate some rough-shot article on Anand that has no real substance is not the best way to be spending time. There are smarter authors and better sites to read, and there is more accurate information to be had, some of it right here.

 

[mwarps]

electromigration occurs at any voltage, just the rate increases rapidly as current increases.

Fixed

it depends on the speed and voltage, obviously, which is the point i've been trying to make all along. the more voltage and heat, the faster the chip deteriorates. end of story, regardless of how little or how much you increase the voltage. opposite is true for undervolting, but as you say...

True, but current is the culprit and it is far more important than temperature for standard failure. If you want to talk overheating, you'll just melt a solder joint, or pop a wirebond. Heat only becomes and issue once you start cranking it really high, because your drive current increases at high heat (in mosfets)and then you start getting into problems. (current increases are due to basic phyics, and the mobility of the carriers also drop at high temps, but that's a discussion for another thread) Realistically, heat is a problem if it's a long-term issue, or if it is very high.. How high is very high? Depends on the chip.

you're absolutely correct, but not for the reasons i think you think. keep in mind that there are several things that deteriorate and ultimately fail over a period of time. even if you get one thing way down, something else will fail before it, making it difficult to get a really long lifespan. it's easy to kill things quickly though

Completely correct.

also, with memory, not only do we have to worry about the transistor failing, the capacitor is also a factor, which is what i believe to the primary reason why DRAM chips fail so much more often than things like CPUs

While I've never studied DRAM failure, I'd be willing to wager the major failure mode in memory is blown caps. FETs are REALLY hard to kill at low speed and low current.

 

[(cf)Eclipse]

thanks for the corrections (just an fyi, i'm planning on adding a microE minor in the spring, so don't be surprised if i randomly bug you with homework questions )

as for the failure mode of DRAM, the more i think about it, the more i agree. one thing that capacitors traditionally don't like is lots of fluctuating voltage... and guess what turning on and off bits does

 

[mwarps]

(just an fyi, i'm planning on adding a microE minor in the spring, so don't be surprised if i randomly bug you with homework questions )

Sweet ass sweet! I'm going to claim partial credit for that
Let me know what classes you're going to take