Risks of Overclocking

[Atheus]

Originally posted by: Dark Cupcake

Originally posted by: Atheus

Originally posted by: Dark Cupcake

Originally posted by: Atheus
If you're not raising the voltages you won't damage anything. Even if you are raising voltage, it's pretty difficult to do actual permanent damage. Your motherboard is not designed to fry processors.

If your not raising voltages, the extra clock speed will still damage the chip

Why? How? Do transistors have a certain number of 'switches' in their life and then they die?

Cpus are made using gates based on cmos transistors...

Yes, thanks, I do have a compsci degree y'know

What I don't have is an EE degree, so I'm not sure how the _individual_ transistors behave from a physical point of view.

I was just wondering if they degrade more simply by switching more often. I guess if you say a transistor can conduct X amount of current in it's life before dying, then either raising voltages _or_ raising clock speed would affect it in the same way - but that's not how they work is it? If my CPU can switch 1V for 10 years, that reasoning would mean it could switch 2V for 5 years, which of course isn't true...

This makes me think the current switched each time is _way_ more important than the number of times it it switched.

 

[jiffylube1024]

Originally posted by: Atheus

Why? How? Do transistors have a certain number of 'switches' in their life and then they die?

Current leakage and running excessive voltage compared to their rated spec is what causes them to fail prematurely.

As others have said, the obvious risk of overclocking is that you kill the component(s) that you are overclocking. As long as you know what you're doing, it's very difficult to "kill" components.

The general rule is to not increase the voltage over the rated spec by more than ~20% on air cooling, though it varies on architecture.

For example, based on my past experiences, it's safe to give a Northwood Pentium 4 (rated to 1.5-1.55V) up to ~1.65V, though I wouldn't go higher than 1.6V. 1.70V or more caused death within weeks for several users.

For athlonXP, they were very tolerant; I ran them up to 1.75V but many people ran them up to 1.85V safely.

Athlon64 90nm chips I'd say 1.5-1.625V is fine; Dual core A64's I'd say 1.5-1.55V tops...

Do your research into how much voltage you can give the chips, what kind of cooling to use and you should be fine (ie stock cooling generally sucks unless they're loaded with heatpipes and cooling fins).

 

[Caldenfor]

I am getting an E6400 and I will be putting a Thermalright Ultra-120 on it with a 120mm Yate Loon fan.

Dave

 

[cmdrdredd]

Originally posted by: Caldenfor
I am getting an E6400 and I will be putting a Thermalright Ultra-120 on it with a 120mm Yate Loon fan.

Dave

And you should get a minumum of 3.2Ghz out of it easily.

 

[TuxDave]

Originally posted by: Atheus
Yes, thanks, I do have a compsci degree y'know

What I don't have is an EE degree, so I'm not sure how the _individual_ transistors behave from a physical point of view.

I was just wondering if they degrade more simply by switching more often. I guess if you say a transistor can conduct X amount of current in it's life before dying, then either raising voltages _or_ raising clock speed would affect it in the same way - but that's not how they work is it? If my CPU can switch 1V for 10 years, that reasoning would mean it could switch 2V for 5 years, which of course isn't true...

This makes me think the current switched each time is _way_ more important than the number of times it it switched.

The last sentence puzzled me. The current flowing from (or to) a CMOS transistor will switch when the transistor is 'switched'. While transistors are described as switches that open and close, CMOS logic gates are composed of at least two transistors. One is attached to power and the other is connected to ground and when one is open the other is not. So when the gate switches, it either delivers current or it will sink current. So when a transistor switches, it's the same thing as saying the current switches.

Also, this does have an effect on the lifetime of the transistors. There is a theoretical lifetime due to electromigration. Given enough time, the current will pull atoms off of the vias connected to power, or the output and it'll break like a fuse.

 

[OCNewbie]

I have a question....... how about undervolting? If everything is stable, like Prime95 or Orthos, is it a problem to slightly OC and then undervolt? I have my Opty 165, 1.8ghz 1.35v stock, running at 2.0ghz 1.25v and is stable. Any problem with that? Runs cooler this way, that's why I do it.

 

[Nessism]

Originally posted by: OCNewbie
I have a question....... how about undervolting? If everything is stable, like Prime95 or Orthos, is it a problem to slightly OC and then undervolt? I have my Opty 165, 1.8ghz 1.35v stock, running at 2.0ghz 1.25v and is stable. Any problem with that? Runs cooler this way, that's why I do it.

Why bother? The chip will basically last forever when run at stock voltage so what's to gain?

 

[OCNewbie]

Keeps my room cooler, or maybe I'm just imagining things.

Oh, and it keeps my northbridge and surrounding chipsets cooler too.

 

[Kwint Sommer]

Originally posted by: OCNewbie
I have a question....... how about undervolting? If everything is stable, like Prime95 or Orthos, is it a problem to slightly OC and then undervolt? I have my Opty 165, 1.8ghz 1.35v stock, running at 2.0ghz 1.25v and is stable. Any problem with that? Runs cooler this way, that's why I do it.

Run the thing at as low a voltage as you can. Going below stock doesn't matter much unless it's a P4 Prescott but it will save you upwards of 20 cents a month in electricity and can add a year or two to the several decades a normal CPU will last.

Excessive heat will fry a CPU, if you are using a Core 2 Duo keep it under 50c and you'll never have a problem. In the long run voltage will eventually form various metal oxides out of the transistors in the CPU and in so doing kill it. Under stock conditions a well designed and built CPU should last more than a century. Extreme overclocking, between high voltages and more clock cycles, can bring that all the way down to a decade or two. So if you want your CPU to outlive you, don't overclock. If you plan to upgrade in a few years keep it cool and don't go crazy with the voltages and you will be fine.

 

[jiffylube1024]

Originally posted by: OCNewbie
Keeps my room cooler, or maybe I'm just imagining things.

You're imagining things, or you have a VERY tiny room.

Oh, and it keeps my northbridge and surrounding chipsets cooler too.

Perhaps because the CPU is emiting less heat, but it's less than you think.

For starters, the Athlon64/X2 architecture is very efficient - at idle it uses and dissipates less power (therefore heat) than even a Core 2 Duo. So it's only at load that it will really make a difference.

Second, unless you have a tiny case or very poor airflow, the heat produced by the CPU shouldn't affect the surrounding northbridge/southbridge very much; the main factors affecting heat output on the Northbridge/southbridge are their respective clockspeeds and voltages (why don't you try undervolting those too ).

If you can Prime95, etc. your CPU at 1.25V stably for hours then go for it. But it won't draw that much less power or produce that much less heat. But every little bit counts, I guess...

Excessive heat won't necessarily fry a CPU, it will barely even shorten the life of a CPU; certainly not the useful life of the CPU. If the CPU is running at >50C, then heat shouldn't matter at all. It's only if it's pushing the 60's and 70's (or above!) for hours on end that it can really make a difference.

The main killer of CPU speed is voltage, and it's not the stock voltage either. CPU's are rated to run >10 yrs at stock voltage. It's only when you exceed stock voltage by a sizeable margin (combined with heat, etc.) that really drops the life of a CPU.

But if you're not drastically overvolting your CPU, it should last 5+ years easy.

 

[jiffylube1024]

Originally posted by: Kwint Sommer
Extreme overclocking, between high voltages and more clock cycles, can bring that all the way down to a decade or two.

I don't think even the most miserly of us is planning to keep their CPU in use for more than a decade. Certainly not much more than that.

A decade ago was the end of the Pentium 1 era (Pentium II debuted in mid 1997); not many people are running Pentium 1's these days...