The Truth About Processor "Degradation" ie. Overclocking :S

[DiGiTaL MoNkEY]

Found this part of an Intel processor article, quite a interesting read if you have some time..



Here is a snip of the obvious stuff...

Degradation - the process by which a CPU loses the ability to maintain an equivalent overclock, often sustainable through the use of increased core voltage levels - is usually regarded as a form of ongoing failure. This is much like saying your life is nothing more than your continual march towards death. While some might find this analogy rather poignant philosophically speaking, technically speaking it's a horrible way of modeling the life-cycle of a CPU. Consider this: silicon quality is often measured as a CPU's ability to reach and maintain a desired stable switching frequency all while requiring no more than the maximum specified process voltage (plus margin). If the voltage required to reach those speeds is a function of the CPU's remaining useful life, then why would each processor come with the same three-year warranty?

The answer is quite simple really. Each processor, regardless of silicon quality, is capable of sustained error-free operation while functioning within the bounds of the specified environmental tolerances (temperature, voltage, etc.), for a period of no less than the warranted lifetime when no more performance is demanded of it than its rated frequency will allow. In other words, rather than limit the useful lifetime of each processor, and to allow for a consistent warranty policy, processors are binned based on the highest achievable speed while applying no more than the process's maximum allowable voltage. When we get right down to it, this is the key to overclocking - running CPUs in excess of their rated specifications regardless of reliability guidelines.

As soon as you concede that overclocking by definition reduces the useful lifetime of any CPU, it becomes easier to justify its more extreme application. It also goes a long way to understanding why Intel has a strict "no overclocking" policy when it comes to retaining the product warranty. Too many people believe overclocking is "safe" as long as they don't increase their processor core voltage - not true. Frequency increases drive higher load temperatures, which reduces useful life. Conversely, better cooling may be a sound investment for those that are looking for longer, unfailing operation as this should provide more positive margin for an extended period of time.

Here is the full article with graphs and more details: http://www.anandtech.com/cpuchipsets...spx?i=3251&p=6

[SLATYE]

Some of that makes no sense. They say that CPUs are binned based on the maximum speed at the stock voltage - and then say that it's unsafe to run CPUs faster than the rated speed even if you stay at stock voltage. If the maximum speed at stock voltage is greater than the rated speed then clearly Intel aren't binning/labelling the CPUs based purely on the maximum speed at stock voltage - if they were, you'd expect the CPU to be rated at the maximum speed.

They've completely missed the biggest issue: demand. If Intel binned and labelled CPUs based only on the maximum clock speed at stock voltage, there'd be about ten times as many E8500s available as E8400s and E8200s. Almost all of the 45nm chips have no trouble running at 3.16GHz on stock voltage, so they'd all end up being binned as E8500s.

Of course, that's not what actually happens. Intel realises the problem and labels some of the E8500s as E8200s and E8400s, so that they can offer a decent range of CPUs. Demand for the E8200s and E8400s is far higher than demand for the E8500s, so it makes sense to sell some of those faster chips with a lower speed rating.

If you have an E8500 which is marked as an E8200 and overclock it to E8500 speeds on stock voltage, it will run hotter than an E8200 at stock speed. In fact, it'll run exactly as hot as an E8500 at stock speed. If this was a problem then Intel would have a shorter warranty on the E8500s (since they'd be expected to die sooner).

[DiGiTaL MoNkEY]

Interesting points SLATYE, reading through the article atm.

[chainbolt]

Originally Posted by SLATYE

Some of that makes no sense. They say that CPUs are binned based on the maximum speed at the stock voltage - and then say that it's unsafe to run CPUs faster than the rated speed even if you stay at stock voltage.

I must agree, that does not make sense at all.

Originally Posted by SLATYE

If the maximum speed at stock voltage is greater than the rated speed then clearly Intel aren't binning/labelling the CPUs based purely on the maximum speed at stock voltage - if they were, you'd expect the CPU to be rated at the maximum speed.

This is also true. And it is well known that in particular at the beginning of the life cycle, for marketing reasons, many samples are labeled/rated lower than they actually could run. This is because Intel/AMD wants to introduce higher speed grades (or more demanding quad core versions) at a later point of time. That's the reason why we got, for example, so many wonderful overclocking E6600 and E6700 in 2006, but later they did not overclock anymore.

[irR4tiOn4L]

Originally Posted by SLATYE

Some of that makes no sense. They say that CPUs are binned based on the maximum speed at the stock voltage - and then say that it's unsafe to run CPUs faster than the rated speed even if you stay at stock voltage. If the maximum speed at stock voltage is greater than the rated speed then clearly Intel aren't binning/labelling the CPUs based purely on the maximum speed at stock voltage - if they were, you'd expect the CPU to be rated at the maximum speed.

They've completely missed the biggest issue: demand. If Intel binned and labelled CPUs based only on the maximum clock speed at stock voltage, there'd be about ten times as many E8500s available as E8400s and E8200s. Almost all of the 45nm chips have no trouble running at 3.16GHz on stock voltage, so they'd all end up being binned as E8500s.

Of course, that's not what actually happens. Intel realises the problem and labels some of the E8500s as E8200s and E8400s, so that they can offer a decent range of CPUs. Demand for the E8200s and E8400s is far higher than demand for the E8500s, so it makes sense to sell some of those faster chips with a lower speed rating.

If you have an E8500 which is marked as an E8200 and overclock it to E8500 speeds on stock voltage, it will run hotter than an E8200 at stock speed. In fact, it'll run exactly as hot as an E8500 at stock speed. If this was a problem then Intel would have a shorter warranty on the E8500s (since they'd be expected to die sooner).

The more i think about it the more interesting your comment becomes. On the face of it, if most 45nm CPUs clock well, youd think Intel would make more money if it made its E8400/E8200s faster, or produced more E8500s instead of taking what is virtually an 'E8500' and badging it E8200 for a lot less money.

But your argument is a little different; Intel essentially takes a homogenous chip, but sells it at different price points in a hierarchy that is matched to their research of demand levels. Although some chips are sold for much less then they could be, if the hierarchy were disturbed it would destroy part of its sales. Theyve reached the 'optimum combination'

This has quite a few implications; the economy of scale needed to satisfy the entire market's hierarchy of demand is gigantic, but obviously Intel has it. Competition must also be pretty weak or compliant (also using this 'hierarchy of demand') in order for this to work, which it obviously is.

But once these conditions are met, were not actually in a situation where pricing conditions under perfect competition apply. In fact, those rules dont apply at all; instead of transferring the cost of manufacturing plus a markup that gets lower as you increase competition, we are instead pricing according to what the market is willing to bear, with manufacturing costs and markup as end results.

Although this is nothing new, its still startling to realise that companies like Intel are so big and such a monopoly/duopoly that they begin to break down the efficiency of the capitalist market.


OF course, it could just be that the CPU's really do differ in reliability/lifespan based on yield.

[metamorphosis]

I had a celeron 2 566mhz that used to run fine at 950mhz. Later it would only run stably at 900mhz (same MB). Later it stopped being able to do that and would only run stably at 800mhz. It's not stable at even that speed now. Degradation- yes.

[power]

Originally Posted by metamorphosis

I had a celeron 2 566mhz that used to run fine at 950mhz. Later it would only run stably at 900mhz (same MB). Later it stopped being able to do that and would only run stably at 800mhz. It's not stable at even that speed now. Degradation- yes.

this may not be the processor only it can also point to degradation of the mainboard or psu in the system


as for processor degradation i'm hardly surprised

[Lardman]

Reading this when I get home. :P

[metamorphosis]

Originally Posted by power

this may not be the processor only it can also point to degradation of the mainboard or psu in the system

Nah, both were changed later on, had no impact.
Cheers,
Matt

[TMM]

Carefactor=0. Most overclockers don't want to keep their CPU for 9 years, so it's a moot article. They also draw conclusions on current processors (i.e. pulling it out of their arse) when we don't know how they will react to overclocking.

Also, how is this the 'truth' ?. I don't see any facts, just hearsay. Using a graph to show made up data is pretty funny, imo.

FTR: my rig has an Opteron 170 @ 2.9, 1.4v which has not degraded at all (not even 10mhz at the same voltage) over a period of a year.

[Onyx]

Originally Posted by metamorphosis

I had a celeron 2 566mhz that used to run fine at 950mhz. Later it would only run stably at 900mhz (same MB). Later it stopped being able to do that and would only run stably at 800mhz. It's not stable at even that speed now. Degradation- yes.

My history of overclocked CPUs have all more or less had the same thing occurred - and I'm beginning to wonder whether there's some truth to it.

I had a P3 550E that would run comfortably at 733, it's last stable config was around 680-ish. Same with my P4 2.8C that was did 3.5, over time it developed stability problems that definitely weren't there in the beginning and had to be relegated to ~3.2. My current C2D chip, an E4300 - showed quite accelerated 'degradation' IMHO - as it was a cheap chip I didn't bother with aftermarket cooling and clocked the hell out of it. Folding full time, load temps would routinely reach 70-75degrees. It's only 10mths old, but has developed instability at the same 3Ghz speed it's been running since day 1. Even increased Vcore doesn't seem to help. I've subsequently had to 'clock it down to 2.7 to maintain stability. Which so far it's done so - touch wood.

[SLATYE]

Originally Posted by metamorphosis

I had a celeron 2 566mhz that used to run fine at 950mhz. Later it would only run stably at 900mhz (same MB). Later it stopped being able to do that and would only run stably at 800mhz. It's not stable at even that speed now. Degradation- yes.

But there is no way to check that a similar CPU would not do the same. Assuming that you ran it at stock voltage, I would expect that a CPU with the same core but rated at 900MHz would experience exactly the same degradation. The difference is that when it stopped working at stock speed, you'd say that it had 'died', while with the overclocked one you just say that it's 'degraded'.

Originally Posted by irR4tiOn4L

The more i think about it the more interesting your comment becomes. On the face of it, if most 45nm CPUs clock well, youd think Intel would make more money if it made its E8400/E8200s faster, or produced more E8500s instead of taking what is virtually an 'E8500' and badging it E8200 for a lot less money.

Intel have two things making them want to produce the slower chips:

• They do get some cores which can't run at the higher speed, just not many. They can either scrap these or sell them as lower-end chips. If they sell them as lower-end chips for a lower price than the high-end ones then they must ensure that they have a lot of them available (because demand for the cheap chips is always huge compared to demand for the fast ones). To do this they have to re-label some of the faster chips, because there simply aren't enough slow ones to keep up with demand.
• Future expansion. Intel wants to avoid spending money if they can. If they went right to the limits of the Core2 architecture immediately (I'd guess that this would put the top-end chips at about 4GHz) then they'd have nowhere to go. They'd have to spend a fortune bringing Nehalem forward so that they can make people upgrade. If they sell the relatively slow chips now (E8400 and similar) then they can release a slightly faster version later (E8600) with no extra work (all they've done it increase the multiplier). Quite a few people with E8400s will then upgrade to the E8600s, and this is essentially free money for Intel. I wouldn't be at all surprised if this is the reasoning behind the Q9x00 quad-cores being delayed (a lot of the people who have just bought E8400s will upgrade to the Q9x00s as soon as they're available). Once again, by delaying the faster chips (quad-cores), Intel can ensure that a lot of customers buy two CPUs! Clearly this only works for whatever company is on top at the moment. AMD have to release their very fastest CPUs just to try to keep a bit of the market share.
• There's no point releasing faster CPUs. At the moment, Intel probably has something like 90% of the enthusiast market, and they're gaining in other areas too. AMD's Phenom is too slow, too hot, and too hard to find. If Intel released a 4GHz CPU for $300 they'd probably end up with 95% of the enthusiast market (there'll always be some people sticking with AMD). That 5% market share gain just isn't worthwhile, when it'll mean that they don't have future upgrade options.

Originally Posted by irR4tiOn4L

But your argument is a little different; Intel essentially takes a homogenous chip, but sells it at different price points in a hierarchy that is matched to their research of demand levels. Although some chips are sold for much less then they could be, if the hierarchy were disturbed it would destroy part of its sales. Theyve reached the 'optimum combination'

Yes indeed. They've clearly done a lot of work here in figuring out exactly where the best price points are. AMD still puts a bit of pressure on them in the low-end market (note the new Celeron Dual Core E1200 to compete with the low-end A64 X2s) but in the high-end it's pretty much completely up to Intel.

I don't think it's true that "some chips are sold for much less than they could be". If Intel deleted the E8200 and E8400, selling just E8500s, they wouldn't actually sell many more E8500s than they do at the moment. For most people, the E8500 is just too expensive. Essentially, the E8500s which are now sold as E8200s are not selling for less than they could be sold - because if they were priced as E8500s, they wouldn't sell at all. By selling them at a lower speed Intel can at least make some money.

Originally Posted by irR4tiOn4L

This has quite a few implications; the economy of scale needed to satisfy the entire market's hierarchy of demand is gigantic, but obviously Intel has it. Competition must also be pretty weak or compliant (also using this 'hierarchy of demand') in order for this to work, which it obviously is.

Yes, the competition thing is critical. For consumers, the best situation is when the companies are fairly close. We saw a great example not long ago when AMD released the Athlon64. Previously Intel was easily in the lead - the Athlon XP was really getting pretty ancient. If AMD had not released the A64, the P4s would have survived a lot longer. As soon as the A64 was released Intel was forced to work a lot harder, and soon we got an excellent result: the Core 2 Duo.

When one company is really struggling (like AMD is now) then the other one can do exactly what was mentioned above - instead of releasing their fastest CPUs, they release them in a way which maximises profit.

Originally Posted by irR4tiOn4L

But once these conditions are met, were not actually in a situation where pricing conditions under perfect competition apply. In fact, those rules dont apply at all; instead of transferring the cost of manufacturing plus a markup that gets lower as you increase competition, we are instead pricing according to what the market is willing to bear, with manufacturing costs and markup as end results.

Exactly. I guess you could say that they're always aiming to sell at whatever price the market will bear, but the other manufacturer can decide what the market will bear. For example, the market will currently bear the Geforce 9600GT's ~$200 price tag. However, if AMD dropped the price on the Radeon HD 3870 to $100 then the market would most definitely not pay $200 for the 9600GT any more.

[one4spl]

Has anyone tested a uniform group of overclocked and non-overclocked CPUs and then tested their over-clockability after a few years?

What I'm trying to say is do overclocked CPUs really degrade all that much faster than non-overclocked ones? My guess is that we simply notice the OC'd ones because they start to fail at OC'd speeds.

[Dangerous_Dave]

Originally Posted by TMM

Carefactor=0. Most overclockers don't want to keep their CPU for 9 years, so it's a moot article. They also draw conclusions on current processors (i.e. pulling it out of their arse) when we don't know how they will react to overclocking.

Also, how is this the 'truth' ?. I don't see any facts, just hearsay. Using a graph to show made up data is pretty funny, imo.

FTR: my rig has an Opteron 170 @ 2.9, 1.4v which has not degraded at all (not even 10mhz at the same voltage) over a period of a year.

This is exactly my thoughts as well.
Been running an e6600 at 3.4ghz for the last 2 years and if it ever dies I'm just going to upgrade anyway.

[SLATYE]

Originally Posted by one4spl

What I'm trying to say is do overclocked CPUs really degrade all that much faster than non-overclocked ones? My guess is that we simply notice the OC'd ones because they start to fail at OC'd speeds.

That's an extremely good point, and I'd be very surprised if you weren't completely correct. Also, in a non-overclocked system a lot of people completely overlook the CPU as a possible cause of instability - they'll blame Windows, blame the PSU, blame the RAM, or blame the mainboard, but the CPU is rarely considered. When they've checked all of those and can't find the problem, they just throw away the system because it "wasn't working". As a result, you never hear about this "degradation" when it occurs in CPUs at stock speed.


Regarding the whole "it's irrelevant because we don't use the CPUs for that long anyway" argument: I completely agree. Realistically, after five years the system will be obsolete. Even if it's still in use, the mainboard is likely to die before the CPU, and it'll be very hard to find a replacement. Finding a decent Socket A mainboard now is virtually impossible, but it's been less than three years since the Athlon XP finished production.

Even if the CPU did die in maybe three years, who cares? It's not uncommon to pay ~$250 for a CPU and overclock it until it's faster than any currently available stock CPU. That includes the $1500+ Extreme Editions. Essentially, you're getting $1500 worth of performance for $250, a $1250 saving. When the CPU dies, you can build a whole new system with the money you saved there!

The only time when the CPU is likely to die 'too soon' is when you're running at pretty high voltage levels. We saw this with the Sudden Northwood Death Syndrome, where CPUs running at >1.7v would often last a matter of weeks.