Does a CPU keep its processing speed after years of use? Is electron migration an issue that affects CPUs?

I want to know definitively if I can build computers using USED processors and expect the same performance as new processors of exactly the same make and type with exactly the same motherboard and drives.

I have searched for several days now online without finding a definitive answer or any studies that explain whether CPUs decline in speed after extended periods of use. I want to know if a used but functional CPU will provide the same performance as a new-from-the-box CPU.

Edit: To be clear, I am looking for a definitive scientific answer, supported by evidence, so that I can make informed decisions when building computers and refurbishing computers. I think this different enough from other questions asked that relate to this topic.

CPU processing speeds will decline as the temperature increases and heat transfer compound may need to be replaced, but aside from this will a CPU lose overall processing power or 'wear out' after extended use?

If so, what causes this? I have heard of electron migration in motherboards, but could this actually cause a CPU to slow down with use? From what I understand about electron migration, it is more likely to cause outright failure of the CPU rather than a gradual slowing of processing speed.

Matthew

Posted 2017-11-17T04:49:16.457

Reputation: 972

Old ones certainly 'feel' slower, especially when you are accustomed to using a new fast cpu. But I've never seen any benchmarking data that would suggest a trend with age. – Sir Adelaide – 2017-11-17T05:01:43.433

6The apparent decline in processor speed is mostly due to software bloat added with every program and OS update/upgrade. – sawdust – 2017-11-17T05:09:08.453

5Does friction increase on well-worn data paths? :) :) :) – Solar Mike – 2017-11-17T05:32:44.293

@SolarMike, actually YES :D see below. – Vladimir Cravero – 2017-11-17T08:58:15.373

NIce one, learn something every day... – Solar Mike – 2017-11-17T09:02:33.640

@SolarMike well, it isn't friction, but the answer explains it. As an aside, my Pentium 4 runs just as well as it always did. – Baldrickk – 2017-11-17T09:12:05.797

In the old days, disk fragmentation would slow a desktop down. Today, most systems have a scheduled defrag built in. – Walter Mitty – 2017-11-17T11:36:08.633

@WalterMitty also (as far as I know) modern file systems often don't need to do defragmentation as often as old file systems as they are better at planning where to store files. – mathreadler – 2017-11-17T11:55:56.360

@Baldrickk exactly. CPUs are designed so that EM doeas not affect performances, but theoretically it is possible. – Vladimir Cravero – 2017-11-17T13:18:29.667

@mathreadler I haven't kept track of that. It seems unlikely to me since, if the free space is fragmented, the OS has little choice about where to put new files. However, if the defrag happens behind the scenes, the use may be completely unaware of it. Also, disk drives with no moving parts may not be slowed down at all by fragmentation. – Walter Mitty – 2017-11-17T15:19:38.480

@WalterMitty yes if the free space is fragmented, but maybe it is better at avoiding that happening by planning in advance. – mathreadler – 2017-11-17T22:32:15.133

Answers

CPUs will not change speed with age. Their speed is determined by the clock speed and the physical design characteristics of the CPU.

CPUs do seem to get slower over time. This is caused by a variety of factors, but the main is simply that the things we use CPUs for is changing. For example, you might think you're just using your computer for the same "email and browsing" that you used to do. But email and browsing today are nothing like they used to be. For example, if you compare web pages today with web pages five years ago, they are completely different things.

David Schwartz

Posted 2017-11-17T04:49:16.457

Reputation: 58 310

1Just to add to David's answer: On top of the web pages in the past and today are completely different (loads much more resources for example) - the actual browser today and years ago are a completely different beast. In the past browser are just program to display (simple) HTML pages, now it can open a whole lot more than just pages. – Darius – 2017-11-17T06:30:45.677

Hi @DavidSchwartz would you please have a look at my answer below? You give some nice insights on the perceived speed, but it is possible for the max speed to be reduced by aging. – Vladimir Cravero – 2017-11-17T08:55:08.040

@VladimirCravero I don't think the original question was asking about maximum speed but about actual performance. – David Schwartz – 2017-11-17T09:00:41.497

1I fail to see how maximum speed does not count towards performance... – Vladimir Cravero – 2017-11-17T09:14:13.460

@VladimirCravero Say you're driving a car at 65 miles per hour and the car's maximum speed drops from 95 to 85. Will the performance of the car change? The maximum speed has no effect on performance, only the actual speed the CPU runs does. (If the maximum speed drops below the actual speed, the CPU will fail, not perform worse.) – David Schwartz – 2017-11-17T09:15:26.240

I see your point, but I would say that there are mechanisms that prevents the cpu from failing, slowing it down. One can be thermal throttling, the multiplier is sw or hw limited to a somewhat lower value and you get less performances. – Vladimir Cravero – 2017-11-17T11:04:00.763

I would add 2 examples: (1) HTTP vs HTTPS - stuff is moving to a more secure internet (slowly)... and it takes more processing power to secure things. (2) static vs dynamic web pages - MySpace vs Facebook. Myspace was no where near as dynamic as Facebook. Everything is Javascript, computed on the fly, database accesses, etc. – WernerCD – 2017-11-17T12:34:09.613

@DavidSchwartz and to your point... I wouldn't talk about changing speed... I would say cars of yesteryear carried people that weighed 175lbs... today, the average weight of people is 200lbs. Same engine. Same car. Carrying more weight. Of course it's going to go slower. Related: men weigh 15lbs more than 20 years ago, on average. IE: Computers carry more weight and, of course, go slower because of it

– WernerCD – 2017-11-17T12:38:37.703

@VladimirCravero Thermal/current throtting was explicitly excluded by the OP. Otherwise, there are no such methods so far as I know. – David Schwartz – 2017-11-17T20:30:54.893

Actually, CPUs and electronics components in general do age with time. It might happen that this aging phenomenon impacts the maximum clock speed of the CPU, but assessing the impact on a real case scenario would be next to impossible.

The main problem I see is electromigration. Basically, on the CPU you have tons of transistors connected by very short and thin metal traces. The electrons, when moving in the conductor (Al or Cu normally), can punch some of the lattice atoms. With time, the atom moves around and settle in places where they are not supposed to be. This phenomenon is similar to what happens with a river, that slowly transports materials from one place to another.

Electromigration can destroy a connection, and this would lead to a non functioning CPU, but before destruction, the connection gets thinner and thinner, so the series resistance increases. If the connection is the power supply for a certain block, increasing its resistance means that the voltage the block will see, will decrease. Roughly, less voltage = less speed, so you might get unreliable operation because of race condition between paths that now have different speed because of aging. This problem can be solved by reducing the clock speed. The actual clock frequency is not impacted by electromigration, because the clock is generated using a quartz crystal, which is extremely stable in any scenario.

This phenomenon is well known and studied, and it is taken in account when designing a chip. I do not expect any modern consumer processor to have such a problem, within its designed lifespan at least, but the possibility is there.

Vladimir Cravero

Posted 2017-11-17T04:49:16.457

Reputation: 298

I would also say that increased resistance in power paths can lead to temperature increase, that is mitigated by frequency reduction. – Vladimir Cravero – 2017-11-17T08:57:49.580

See, I was going to say something basic like if a CPU experiences frequent hot/cold cycles (so operating for long period of time doing intensive tasks, then switched off daily) you could expect to see metal fatigue in the connections at the microscopic level. But your answer is far more involved, I like it. – mickburkejnr – 2017-11-17T09:59:25.990

1I have heard of electron migration in motherboards, but could this actually cause a CPU to slow down with use? From what I understand about electron migration, it is more likely to cause outright failure of the CPU rather than a gradual slowing of processing speed. – Matthew – 2017-11-17T11:35:24.613

@Matthew No, It cannot. – Ramhound – 2017-11-17T11:40:08.837

If it is such an urgent matter, you should contact the manufacturer. I would say that EM effects are negligible, but given the fact that you do not have/provide data on previous usage of the used processors, everything is possible. @Matthew – Vladimir Cravero – 2017-11-22T00:06:08.537