What is the maximum (safe) voltage for a socket 939 AMD 90nm Venice core?

[brigby]

What is the maximum (safe) voltage for a socket 939 AMD 90nm Venice core?

I'm somewhat of an overclock enthusiast and I've been plagued by this question for quite some time.

I've looked on many forums and nobody anywhere can seem to provide a reasonably infallible solution to the question. Most people claim to know but simply speculate.

I know that this forums is read by many people who have an intimate understanding of IC's and I was wondering if you could provide a more technical, valid, answer than the usual DOOD LIKE 1.7V ITS TEH MAXX.

Thanks.

 

[Matthias99]

What is the maximum (safe) voltage for a socket 939 AMD 90nm Venice core?

Stock.

Seriously, unless you want to know more about WHY overvolting is bad for your CPU, this is a question for the CPU/Overclocking forum. It's a pretty theoretical question anyway. There is no way to pin it down for a particular CPU, since voltage tolerance (like overclocking potential) can very widely from chip to chip, and you won't know how much is too much until you have pushed it too far (which, with overvolting, usually results in damage).

 

[brigby]

I would certainly be willing to entertain a discussion about why overvolting is bad for the CPU

 

[Ayah]

You'll probably want an EE to answer you but I can tell you that beyond 2.0V, it's all based on luck, since you'd need a hell of a cooling system to get there anyways.

 

[MrDudeMan]

Originally posted by: Ayah
You'll probably want an EE to answer you but I can tell you that beyond 2.0V, it's all based on luck, since you'd need a hell of a cooling system to get there anyways.

[rant]

:Q!!!!! i didnt think anyone was stupid enough to go much past 1.65. i overclock some, but there is no difference in anything a consumer will do with the extra speed since almost nothing at home will be limited by the CPU. its just bragging rights, but i dont even get why since engineers made the chip, not the person OCing it. the 'enthusiast' at home that OCs the processor is just switching a few options in the bios and they think they are a badass. that really annoys me.

[/rant]

 

[BrownTown]

yeah, i also think its funny that people think they are computer experts becasue they know how to open up the BIOS on a computer and change some numbers.

But to the overvolting point. Tehre are several reasons why over volting a CPU is bad. First off there is heat. The heat generated by a processor is from both static and dynamic loads. Basically the static load comes from the resistance of pushing currennt threw a processor (and because of leakage currents in the processor). The dynamic load is caused by the transisters requireing power to be switched on and off. Dynamic load is directly proportional to clockspeed which oif course explains why overclocking a processor produces mroe heat, and also wht intel proessors are much hotter. Anyways, both sources of heat are also proportional to the square of the voltage, so the more you increase the voltage the more heat is being produced.

Also, the increased voltage is bad for the very fine circuitry of the chip. I don't know what the breakdown voltage of siicoan is, but i'd image for such small distacnes there is a change that you can litterly burn microscopic holes threw the circuitry.

 

[pm]

We recently had a discussion about this in the OC forum:

http://forums.anandtech.com/messageview.aspx?catid=28&threadid=1773169

As far as the original question, the answer to this is "define 'safe'". Manufacturers choose a voltage for a shipping part that is the maximum possible that still meets the long-term reliability goals of the company for that product.

The maximum "safe" voltage is the stock voltage from the manufacturer, everything above that increases the risk of premature failure.

Clearly, EE's are well aware that you can clock a part faster if you increase the voltage. This is one of the first things you learn at college, and it's fundamental to the entire process development and product development. The problem with increased voltage is decreased reliability. So a part intended to last 7 years, will instead die much earlier. How much earlier is the next question that most people ask, and truly there are very few who could answer that... manufacturers figure out what will work for their reliability goals... they don't worry excessively (in my experience) about what happens if you exceed that specification. So, not even engineers very familiar with a product will have much more than an educated guess about what will happen if you increase the voltage by 20%.

Definitely there is a non-linear dependence on reliability with increased voltage - even accounting for no increases in heat... it's an exponential dependence. So increasing the voltage by 20% will not decrease the expected longevity of the part by 20%... the impact will be much greater. How much greater... I don't know.

If we want to have a discussion of what exactly causes an integrated circuit to fail, then I'd be more than happy to dive deeper into this.

 

[BrownTown]

however it should be noted that companies like AMD and Intel are very focused on reliabiliy and therefore build in a good deal of headroom so that their chips will work as advertised. Increasing the voltage slightly might still decrease life expectancy, but maybe only from 8 years to 6. And if you are a computer enthusiest that probably won't really concern you at all.

 

[bobsmith1492]

The formula for power dissipation in a logic chip is P = 2fCv^2, where f is the switching frequency, C is the capacitance of the input being driven and connecting wires, and v is the voltage. From there on.... yeah, it's mostly just how much power is dissipated, as that creates heat that increases the resistances of the wires, creating more voltage drop (==heat), and so on. I suppose arcing between wires would occur if you put in waay too much voltage, like 30 or so; I can't remember or find what the lowest arc voltage in air is.

 

[maluckey]

The fact that most overclockers are not interested in maximum voltage, just the maximum speed at which the chip can operate with stability, leads us all to an interesting question. The question is has anyone bothered to try and deliberately kill a batch of chips to find out how much voltage is needed to kill a chip?? Mathematically and theoretically are fine, but are not good enough when you must take into account tolerances in manufacturing and all.

Overclockers run out of speed before they run out of voltage in most every case. I myself have run a Mobile XP-2400 at 2.1 volts actual and at 203 x 13. That's on air cooling for stability testing for three days straight of P95 and S&M. That's a HUGE increase above stock for this 35W mobile chip. Two years later, this chip can still run at default voltage 24/7/265.

This question will remain unanswered until a large scale test is done and the tester is willing to kill the chips to find out.

 

[RaynorWolfcastle]

Originally posted by: pm
We recently had a discussion about this in the OC forum:

http://forums.anandtech.com/messageview.aspx?catid=28&threadid=1773169

As far as the original question, the answer to this is "define 'safe'". Manufacturers choose a voltage for a shipping part that is the maximum possible that still meets the long-term reliability goals of the company for that product.

The maximum "safe" voltage is the stock voltage from the manufacturer, everything above that increases the risk of premature failure.

Clearly, EE's are well aware that you can clock a part faster if you increase the voltage. This is one of the first things you learn at college, and it's fundamental to the entire process development and product development. The problem with increased voltage is decreased reliability. So a part intended to last 7 years, will instead die much earlier. How much earlier is the next question that most people ask, and truly there are very few who could answer that... manufacturers figure out what will work for their reliability goals... they don't worry excessively (in my experience) about what happens if you exceed that specification. So, not even engineers very familiar with a product will have much more than an educated guess about what will happen if you increase the voltage by 20%.

Definitely there is a non-linear dependence on reliability with increased voltage - even accounting for no increases in heat... it's an exponential dependence. So increasing the voltage by 20% will not decrease the expected longevity of the part by 20%... the impact will be much greater. How much greater... I don't know.

If we want to have a discussion of what exactly causes an integrated circuit to fail, then I'd be more than happy to dive deeper into this.

I'm not entirely familiar with all the processes that arise when you increase the voltage, but I'd imagine that the main detrimental effects (apart from heat) might be mostly related to increased hot-carrier/impact ionization effects and electromigration. Come to think of it, I'm sure the lifetime of the gate dielectric is also a function of voltage

 

[pm]

Originally posted by: RaynorWolfcastle
I'm not entirely familiar with all the processes that arise when you increase the voltage, but I'd imagine that the main detrimental effects (apart from heat) might be mostly related to increased hot-carrier/impact ionization effects and electromigration. Come to think of it, I'm sure the lifetime of the gate dielectric is also a function of voltage

Cutting my comments from the other thread:
There's a linear dependence on mean-time to fail (MTTF) on a CMOS part to temperature, there's a square dependence on voltage to MTTF.

In non-statistics speak, increasing the temperature a bit will make your chip a little more likely to die. Increasing the voltage a little bit will have a much bigger statistical liklihood to kill your chip.

In the "old days" - prior to 0.18um process technology - the dominant reliability failure mechanism was electromigration... although this depended a fair bit on the design and the design rules used by the manufacturer. In electromigration, higher temperatures and higher voltages reduce the average time to failure. Because electromigration is dependent on current density, increasing voltage is worse.

Once the industry switched to dual-damascene copper technology, electromigration failures were relegated back to design-related mistakes. Instead the dominant failure mechanisms became time-depenedent dielectric breakdown (TDDB), PMOS BTI (although this is pretty much accounted for in manufacturer burn-in), and NMOS hot-electron gate-impact ionization (NMOS hot-e). In all three of these small increases in voltage can result in large reductions in operational lifetime, and small increases in temperature, result in generally small reductions in operational lifetime - in fact, in the case of NMOS hot-e, it gets worse with lowered temperature.

All three of these are quantum mechanical effects. If anyone is really curious, I can go into more details to explain exactly what is happening. For more details, you can read through this (rather condensed, somewhat esoteric but fundamentally correct) notes page http://www.eie.polyu.edu.hk/~ensurya/lect_notes/Reli_Fail/Reli_Fail_notes.htm. Although the whole page is an interesting reference, the part relevant to this discussion starts with "Properties of Metal-Oxide Silicon (MOS) System". Or Google, "NMOS hot-electron", "PMOS Bias Temperature Instability" and/or "time-depenedent dielectric breakdown", I can also give more detailed IEEE journal publications too if anyone wants them.

 

[maluckey]

So we are still at the part of who nows EXACTLY how much voltage....

Sure, more voltage kills faster, and heat kills faster, but exactly how much before death of the chip. I can sit all day and run formulas of design specs, but manufacturing tolerances have to be figured in. Each batch will be slightly different. Some better, some worse. That's why manufacturers oftentimes speed bin the chips. The best get the higher ratings, the dogs get the lower.

 

[pm]

So we are still at the part of who nows EXACTLY how much voltage....

It's all statistics. A certain (extremely small) percentage of the parts don't handle the "stock" voltage and will fail in the field. If you had a vast number of parts (10 million, for example), and you were to graph voltage (y-axis) vs. mean-time-to-fail (x-axis), you should see a "normal" or "Gaussian distribution.

http://en.wikipedia.org/wiki/Image:Standard_deviation_diagram.png

Ideally you want to have all of the curve ahead of the reliability goal (no parts fail before the goal has been met). In reality, you will have most of the parts on the curve ahead of the goal and a very small number fail before they should.

Increasing the voltage by any amount at all, will shift the curve along the X-axis (time)such that all parts will fail sooner. Small increases will shift the curve by a larger amount than you would expect. Doubling the voltage, does not half the expected life. It is more along the lines of (for example) increasing by 20%, halves the lifetime.

So there is no "right answer" - there is no EXACT voltage - about the best you can do is talk about the median of the curve.


Patrick Mahoney
Microprocessor Design Engineer
Enterprise Processor Division
Intel Corp.
Fort Collins, CO

 

[interchange]

It's all very hard to calculate unless you know a lot of exact details about the process. Even still, it would be mostly empirical for a variety of reasons (chip hot spots due to design, process faults, quality of materials, etc.). Chip manufacturers spend billions on test to increase yields. They dig deep into the faults of their process and chip design. Like pm said, it's still all statistics. Initially, you may get a 5% yield. Over time, they will tweak the process and design and the material quality will not always be constant, hence having some good weeks and some bad weeks.

If you want some raw math trying to figure out at what voltage things go bad, I can dig up my EE textbook... But, you're really not going to come to an answer from that.

 

[maluckey]

So voltage increase does not linearly shorten lifespan? I had always thought that it was so...

Makes sense as far as my experience with extreme overvolting. I have yet to have killed a chip though I thought that I would. As far as lifetime of the chip, I have never found the lifetime of any of the abused CPU that I have run. Thay ALL still run at default voltages, though we know that the manufacturers have some chips that run at less than the average voltages. I never tried undervolting to see how much my best chips had for variance. Maybe they can now run at manufacturer default only, whereas before they could have been able to run at a lower voltage than manufacturers claims. I gues that I'll never know.

 

[interchange]

I would expect it to be very non-linear, though I have no math to back it up .

 

[sdifox]

I don't think you can peg a number since tolerance level vary from cpu to cpu.

 

[brigby]

wow. just checked up on this thread after the first busy few days of classes...it seems like i'd need to switch my major to fully understand these posts though

oh well, it's good thing i was warned to "check my ego at the door"

 

[theMan]

most people will say that the "max safe voltage" is 1.55v but, thats based off experience, not science. it just varies between cpu's and depends on how long you are keeping the chip. for example, if you only need the chip for a week, go ahead and use 1.9v, as long as you can cool it. if you need it for 5 years, probably stick with 1.5v