Voltage

Voltage can destroy; respect it.– Ed

Frankenstein Used Too Much Power And Look What Happened To Him

I’ve been getting emails from folks lately who don’t seem to have the slightest clue that jamming 20 or 30% more voltage than default might be bad for their CPU.

If you know that, accept that risk, and realize this could kill your processor or more at any given time, fine, this is not for you. This is for those who think there’s no risk at all.

There’s a number of ways CPUs fail. You can blow out those delicate little channels quickly by simply jamming too much electrical power through a circuit, or more slowly through electromigration.

In simpler terms, either punching a hole through something or wearing away at it until a hole does develop.

I’m Just Lowering My CPU’s Life From Ten to One Year, And I’m Only Going To Use It Six Months, So I’m Perfectly Safe

Wrong. Dead wrong.

If you hear something like “A CPU is built to last ten years,” the originating comment probably said, “The CPU is designed so that the average lifespan is ten years.”

If you build a ton of anything, it will have a failure distribution curve. Taking our CPU as an example, the average lifespan may well be ten years, but a few will last fifty years, and a few will last just fifty days.

If you grossly overvolt or let the CPU run too hot, what you are doing is changing the failure distribution curve so that the average failure will occur much sooner. What that means is that the proportion of CPUs that die within a couple months will be much higher than it would have been had you not done so.

So if you shove 2V into a PIV Northwood (and presuming this reduces the average life to just a year), you may increase the chance of it failing in the course of the course of six months from 0.1% to 20%.

Now Intel and AMD are very coy about these things, but under extreme conditions, they’ve lowered the average life of a processor down to months.

Electromigration is a complex phenomenon. Electrical flow is a big factor, and so is the temperature. To some degree, you can compensate for more voltage with more cooling, but adding a Delta 38 isn’t going to change the electromigration formula too much.

If I Don’t Know, It’s Not There

Somebody offers you a gun and a chance to play Russian Roulette. That person won’t tell you how many chambers are loaded; it could be all, it could be none, it could be some. Would you say to yourself, “Since I don’t know, I’m just going to assume the gun is empty.” You would call anybody who did that an idiot, right?

But that’s just what people so often do in this area. If they can’t quantify something, it isn’t there.

Let’s take that PIV Northwood. Do I know putting 2V in the thing reduces the average life to a year, or that 20% of them will fail in six months? Hell, no. The average life might be three years or three months. The failure rate after six months might be 2%, or it might be 92%.

All I do know is that 30% more voltage than normal is supposed to be bad for a CPU, and that I’ve noticed over the years that very high voltage often leads to dead processors. Not always. Often.

I know there’s some degree of risk, even if I don’t know exactly how big it is. That’s going to make me treat voltage with respect, and compare the reward with the risk.

Usually, CPUs reach a point beyond which they don’t want to go. If you have a Northwood processor that will go at 2400Mhz at 1.5V, but needs 1.85V to do 2500Mhz, that CPU is trying to tell you something.

That extra 100Mhz might give you 3% extra performance, which you’d probably not ever even notice in any meaningful way, but probably increases the likelihood of failure a whole lot more than 3%. Is it worth it? In most cases, no.

Now if you’re willing to take that risk and pay the price of that CPU or more blowing out or just not working one day, fine.

But if you’re not, then ignorance is not bliss, and it doesn’t matter what you think about it. If it’s going to blow out, it’s going to blow out no matter how much you say, “It can’t happen to me,” just as much as that gun will blow your brains out if there’s a bullet in the chamber.

There Is Middle Ground Between Something Harmless And Something Suicidal

Now if you buy a Northwood and find you need a slight voltage nudge to reach your target, all that I’ve said above doesn’t mean you should be in a state of sheer terror as a result, either. That’s just as silly as treating voltage like it’s mustard on a hot dog.

Even Intel and AMD at times increase the default voltage on their processors, so there is some safety margin. The idea behind overclocking and increased voltage is to prudently use it, not abuse it.

On the other hand, you can’t reduce your risk to zero. Electromigration occurs whenever you use your CPU. Eventually, if you used them long enough, every CPU will die due to it. Even at default voltage. No matter how good the cooling is.

So the only way to have zero risk of CPU failure is to never buy a computer, just as the only way to have a zero percent chance of being murdered is to never be born.

CPUs Are Just Like People

A few die right away. A few live a long, long time. Some die before the average life expectancy, some after. They can take a little abuse without harm, but usually not a lot. If abused a lot, they tend to die earlier, but not always. If treated well, they tend to die later, but not always.

Some climb Mount Everest, more won’t climb Everest but will climb a rockpile, others are afraid of bumps in the sidewalk. Some of the Mount Everest climbers never come back (usually those who didn’t prepare well enough for it), but more do and find the experience worth it, and some of them live longer than the stick-in-the-muds.

The only certainty is eventual death, but that doesn’t make life not worth living. All the rest is a matter of playing the odds, and judging the risks against the rewards.

Email Ed