187

This is a hypothetical question about how a CPU operates. If I purchase two identical CPUs, and use one long term (say one year), will it be identical in speed to the unused CPU? Will the number of clock cycles, latency of requests, etc on the used CPU be less than that of the unused CPU?

A supporting argument may be that mechanical devices degrade over time, While a CPU has no moving parts (other than the external fan), it does have circuits that can be damaged by heat, and voltage spikes. Lets say that after a year of intensive use, the circuits degrade and fewer electrons can pass since the pathway is narrower, etc.

Is this the nature of how a CPU operates, or is it simply working or broken, with no speed degradation in between?

17
  • 8
    I suppose that in theory, a CPU could run slower as it ages if the cooling mechanism doesn't cool as efficiently as it used to (maybe the fan breaks a little and can't reach top speed), I think some CPUs can automatically scale back their clock speed if they detect they are too hot. Note that this doesn't mean the CPU itself is aging into poor performance; in this scenario, replacing a bad fan would probably allow the CPU to run as fast as when it was new. I don't have any references to back this up though, but it seems plausible to me... Aug 1, 2012 at 20:47
  • 3
    @FrustratedWithFormsDesigner, I've seen a Dell notebook throttle its CPU severely because it thought it was getting too hot (mainly through bad design I believe). It's entirely feasible that the build up of dust over time can cause that too, but you're correct that it's not strictly age that causes it. Aug 1, 2012 at 21:55
  • 37
    What gets slower is the software. Aug 1, 2012 at 22:00
  • 20
    Here is a great IEEE article written specifically in regards to transistor aging I urge anyone interested in this topic to read. Aug 1, 2012 at 23:50
  • 2
    @JoãoPortela it's all relative. The CPU will run at the same speed/voltage until some transistors stop functioning correctly due to age. The only way to solve the problem at that point is to either slow the CPU down by reducing the clock speed, or increase the operating voltage (further aging the transistors on the CPU die). And of course, over time, the clock generation units in the CPU also become unstable, leading to more clock jitter. Aug 2, 2012 at 15:18

15 Answers 15

129

Is the performance of a CPU affected as it ages?
after a year of intensive use, the circuits degrade and fewer electrons can pass since the pathway is narrower, etc.

No,

Crystal oscillator

the speed of a CPU is determined by a crystal oscillator - so far as I know this is an external part for most CPUs

crystal oscillator Mobo with xtal

Picture from TechRepublic article

Crystals undergo slow gradual change of frequency with time, known as aging.

However, I suspect this is not a significant factor.

Drift with age is typically 4 ppm for the first year and 2 ppm per year for the life of the DT-26 crystal.

(from TI concerning an RTC IC but I believe this rate is similar for timing crystals in general)

CPU Semiconductor changes

Breakthrough posted a link to an IEEE article that describes the myriad of ways that semiconductors are affected over time.

It is possible therefore that the maximum clock speed the CPU is capable of will decrease over time. However in most cases this will not cause the CPU's theoretical maximum possible speed to fall, within a year, below the actual operating speed set by the crystal oscillator. Therefore a CPU that has been stored for a year will run at the same speed as an originally identical CPU that has been used continuously for a year.

CPU Thermal regulation

Many CPUs reduce their speed if their temperature exceeds a pre-set threshold. The main factors that might cause a one-year-old CPU to overheat are not to do with semiconductor degradation within the CPU itself. Therefore these factors have no bearing on the question as formulated.

It is unlikely that a given pair of identical CPUs will diverge in capability within one year sufficiently to trigger thermal issues that require one of them to run itself at a reduced speed. At least, I know of no evidence that this has occurred within one year on a device that is not considered a warranty failure due to manufacturing defect.

CPU Energy efficiency

Many computers, especally portable ones, are similarly designed to reduce energy consumption when idle. Again this is not really relevant to the question as stated.

16
  • 2
    What units are the ppm you reference? I'm familiar with that meaning "parts per million" which doesn't fit here.
    – Cajunluke
    Aug 1, 2012 at 22:21
  • 15
    I interpret it to mean a variation of +/- 0.0004 % of the nominal value in the first year and +/- 0.0002 % thereafter.
    – StarNamer
    Aug 1, 2012 at 23:23
  • 3
    > Clock frequency variations (be they up or down) are a common problem, but they will most probably never be noticeable from the user as slowdowns. You can see them in CPU-Z and similar programs. The frequency fluctuates a few MHz (mine always seem to be a couple below the rated speed), but like you said, it’s a small percentage of the total speed, so no normal human will notice its effects.
    – Synetech
    Aug 2, 2012 at 1:02
  • 2
    I like how this answer addresses the main issue with CPU speed: the clock. The other answer talk about other issues that may affect CPU speed but they are not the main issue that affect CPU speed. Aug 2, 2012 at 12:42
  • 8
    This has supporting evidence throughout the related comments. Breakthrough links to an IEEE article discussing transistor slowdown that suggests that these wear out over time. Then, as DanH mentions, "if the circuit slows down no one notices until errors start appearing due to the clock being 'faster' than the circuit." So, as you mentioned, crystal oscillator dictates the speed, and fluctuates an almost imperceptible amount. As long as the slowing transistors still respond fast enough to the speed set by the crystal oscillator, no slowdown would be measured after a period of time. Aug 2, 2012 at 13:48
77

In theory, no, a CPU should run at basically the same speed its entire life.


In practice, yes, CPUs get slower over time because of dust build-up on the heatsink, and because the lower-quality thermal paste that prebuilt computers are often shipped with will degrade or evaporate. These effects cause the CPU to overheat, at which point it will throttle its speed to prevent damage.

Cleaning the heatsink and reapplying the thermal paste should make it as good as new, though.


Note: if you're asking this due to having an old computer slow down, there are other reasons (usually dying hard-drives or popped capacitors) that old computers will slow down over time.

9
  • 4
    Very good answer. Theory is not reality.
    – Ugo
    Aug 2, 2012 at 7:58
  • 8
    True, I made my CPU faster by vacuuming the dust out of the fan.
    – MSalters
    Aug 2, 2012 at 8:13
  • 13
    @Ramhound: Sorry, but that's not true. Intel has been using their SpeedStep technology since second-generation Pentium III (circa 2000), while AMD had PowerNow! since 1999. I also clearly remember Pentium II's having CPU-throttling before AMD, before they had a fancy trademarked name for it. Aug 2, 2012 at 15:13
  • 1
    How do you explain laptops running hotter over years of constant use? that are cleaned regularly. Aug 2, 2012 at 19:01
  • 1
    @Tony: As I mentioned, the thermal paste likely needs to be reapplied (it could also be that the fan is dying/dead, but that should be easy to see/hear). Aug 2, 2012 at 19:15
37

Short answer, no a CPU will not get slower with age.

Slightly longer answer:

A CPU will work so long as all of the connections and transistors are working properly. While in a normal wire there might be movement that can make the connection intermittent, that is not the case on the CPU as:

  • the circuits are etched into the silicon
  • things are much smaller

If something does break, anything can happen: from bad math to the computer not starting up.

6
  • 18
    Downvoter care to comment?
    – soandos
    Aug 1, 2012 at 20:36
  • 1
    I'm not the downvoter, but it might be because you implied that CPUs are printed; they are actually etched. (Of course, that doesn't change the essential correctness of the answer, so I upvoted you.) I also submitted an edit to correct this.
    – Cajunluke
    Aug 1, 2012 at 22:20
  • 4
    @CajunLuke: Actually, the etching step is just one of many. You first put an anti-etch layer on top, then print the desired circuit on the layer, flush away the printed parts, and then etch the entire surface. Where the protective layer is gone, etching will create channels. in the layer below. The process is called "photolithography"
    – MSalters
    Aug 2, 2012 at 8:21
  • how do you explain fan speed increase and laptop temp rise after years of aging? when kept clean. Aug 2, 2012 at 19:03
  • 4
    Maybe thermal paste deterioration?
    – Spidey
    Aug 2, 2012 at 19:21
12

I would argue - that the essential heart of this matter - has far less to do with physical hardware - as it does with how our perceptions - and the relative performance of the software that we run - change over time.

In a world of 1's and 0's - there is very little that can happen, especially to the CPU - that would drastically (or even statistically) alter the machine's overall performance - other than a total failure.

This question caught my eye because I've recalled times in my life where I couldn't believe the machine I was using - was the same one that maybe only a few years before I thought was so fast - that I was now being tortured by what at that point seemed to be interminably slow.

On a brighter note - as Moore's lawyers have seemed to be on recess - software developers have made major improvements in recent years - that seem focus on fine-tuning performance vs. relying on brute power. It is no exaggeration when I say that my 8-Core Xenon 2.8 GHz Mac Pro seems 2X or 3X faster now than it did when purchased in 2008. These are meaningful and measurable differences that could only be due to massive improvements / optimizations on the software side.

What I'm saying is that the human mind / our perceptions / our expectations, combined with other more flexible aspects of the operating environment are exponentially more impactful than any variations from a factory spec - that you may be worried about.

4
  • This is interesting that you say your Mac performs better now than before. This indicates that software developers for Mac concentrates on improving performance to be able to get more out of the machine force, while PC developers for Windows just use the improved hardware power to make more fancy programs without thinking too much on performance. This is probably why a PC tends to slow down over time - not because the hardware degrades, but because the software demands more of the hardware...
    – awe
    Aug 2, 2012 at 7:14
  • I agree - with you - that the relative human perception of what is fast changes - with time as we repeatedly exceed limits - with faster hardware, note that there are ways around this, for instance by using standardized - benchmarks which do not account for software improvements or subjective bias.
    – Thomas
    Aug 2, 2012 at 7:41
  • 1
    "I would argue - that the essential heart of this matter - has far less to do with physical hardware - as it does with how our perceptions - and the relative performance of the software that we run - change over time." That's nice, but the OP is asking specifically from a hardware perspective. I do agree that the issue is hardly perceptible by a human being, but indeed transistor switching characteristics do drastically change over the lifespan of the semiconductor. Aug 2, 2012 at 15:36
  • @Breakthrough I'm certainly not an electrical engineer and to better understood what you are saying I found this article which is interesting.. But what does it basically says that without millions of dollars worth of equipment not even chipmakers really have their heads around this aging process. I guess it's just that in day-to-day life, this issue is never likely to be the root of anyone's actual problem, nor something that we can even contact with our human senses.
    – mralexgray
    Aug 2, 2012 at 16:26
7

If I purchase two identical CPUs, and use one long term (say one year), will it be identical in speed to the unused CPU?

Most likely, yes. The speed a CPU runs at is variable, and set by the end user (although usually set automatically as per the manufacturer's specifications). However, you might find that at the end of the first year, the unused CPU (assuming they were truly identical to begin with) overclocks better than the used CPU. This effect can be attributed to transistor aging, which you hinted at later in your question:

While a CPU has no moving parts (other than the external fan), it does have circuits that can be damaged by heat, and voltage spikes. Lets say that after a year of intensive use, the circuits degrade and fewer electrons can pass since the pathway is narrower, etc.

This is exactly the case, and is precisely what happens after a CPU is used.

Similar to a vehicle, there is some wear-and-tear on the conductors as electrons pass through them. Heat also affects the transistor aging, which is why the CPU die is designed for a particular range of operating temperatures. During operation, the electrons have to tunnel through some layers in the semiconductor materials, degrading them over time. This causes the switching speed of the individual transistors to increase over time, making them "slower".

However, as I said before, the CPU speed is set by the end user. It's a synchronous digital circuit, and will run as fast as you tell it to - even if the propogation delay exceeds the switching time, and the computer crashes. This is what will happen as a CPU ages. Over time, the various sub-units in the CPU will take longer and longer to finish their computations, leading to instability in the CPU.

This effect can be mitigated by slowing the clock speed down, making the CPU slower but compensating for the increased propagation delays. This effect can also be mitigated by increasing the CPU voltage (causing a reduced switching time for the transistors, allowing for a higher clock speed), but raising the CPU voltage will only cause the transistors to age faster.


This is why we say a processor gets slower as it ages - the processor becomes unstable at higher speeds, requiring you to lower the clock speed over time. The good news is that this effect is usually noticable on a timescale of years.

1
  • This is the best answer (IMO). Expensive hospital equipment (millions of dollars) auto scales down the processor to prolong the life of the machines, since they can't afford to pay millions of dollars for new equipment every few years. Personally, I have a Mac Pro mid-2012, and it is noticeably slower now than even last year. Tasks that used to take milliseconds now take seconds. It's very annoying. It makes me think computers may auto-scale down too as CPUs age. I can't support this claim -- it's just my observation.
    – James L.
    May 3, 2023 at 19:01
4

I am reminded of an effect seen in some early integrated circuits: When relatively high current densities were run through the gold wiring, there would actually be a physical migration of the gold similar to the meandering of a river over time. At corners the corner would slowly migrate outward (just like an oxbow bend in a river) making the wire thinner and longer (and also creating a risk that it would short out to an adjacent wire). This thinning/lengthening of the wires would surely affect the max clock speed of the circuit (if only very slightly).

Anymore, I believe that designers know how to control the manufacturing processes to prevent this specific effect (or at least make it immeasurably small). But, as noted in a comment above, there are several other effects.

However, there are two factors that make it reasonable to say "no, for all practical purposes" in answer to the original question:

  1. The vast majority of computer circuits are externally "clocked", most often with some sort of crystal-controlled oscillator. So if the circuit slows down no one notices until errors start appearing due to the clock being "faster" than the circuit.
  2. There are several effects (eg, metal "whiskers" growing on the circuits -- a serious current problem as lead is removed from circuits) that cause circuit failure long before circuit slow-down becomes significant or even measurable.
1
  • 1
    The wire thinning and becoming longer as you described sounds like the phenomenon of electromigration in the IEEE article above. You are right though, in that designers are building these tracks far enough apart that they wouldn't come into contact. Aug 2, 2012 at 13:38
4

This is not a full answer, but a presentation of a possible source of speed degradation (not as major as throttling due to heat transfer degradation mentioned above though):

Maybe the longest path is increased due to dielectric charge build-up, causing the processor to scale down in order to function. That is, when a vector of inputs is given to a logic circuit, a finite time passes while the physical logic system rattles into place (which sets an upper bond for clock frequency). Dielectric degradation happens to every transistor, making a transistor require higher voltage for the same rise time, or equivalently, lower rise time (less speed) at the same voltage. If a sufficient amount of transistors degrade (unevenly), the longest path might very well change, which may degrade performance in a processor that operates near its logical speed limit.

1
  • 1
    I think if your rise time varies much, the transistor will no longer clock properly (it won't assert its signal long enough for the next part of the circuit to latch before the trailing edge of the clock). This will lead to hard faults, not a slowdown. Your CPU will operate just as fast, it'll just give wrong answers (or flat out reset itself, or wedge solid).
    – TMN
    Aug 2, 2012 at 15:42
3

CPU is synonymous (for most) with Multi-core Processor, which I suspect you are more likely to be asking about.

It's possible for some multi core processors to disable cores that develop faults, either intermittent over-temperature faults, or permanent failures. See the 80-core Intel research chip's self correction functionality. A bad core is effectively marked unusable, and its responsibilities are distributed to other cores, Less cores means your processor has fewer total CPU cycles available and therefore, it will be slower to perform work.

I imagine this will become more common as manufacturers try and keep up with Moore's law, and cram ever more cores onto processor dies.

Edit:

left in so James's Comment makes sense.

According to How-Stuff-Works, the PS3's Cell processor has similar redundancy, it is made with 8 SPEs, uses 7 of them, keeping 1 in reserve in case of failure. I doubt the processor would work if 2 SPEs failed, but I can't find any more information.

4
  • This sounds like a catastrophic oxide failure within a single core. If the core is disabled as part of a self-correction function, this would reduce the overall operations per second as seen in a benchmark. Would the remaining cores though would operate operate at the same level of performance, provided no catastrophic failure though? Aug 2, 2012 at 0:22
  • True, but today's 2, 3 and 4 core systems don't have this kind of self-correction capability.
    – vy32
    Aug 2, 2012 at 11:36
  • @Jon: I don't think you're right about the Cell Processor. Everything I have seen suggests that the this technique is about improving manufacturing yield. So some of the chips come out of the factory with one faulty SPE and would otherwise have been unusable. I have not seen any indication that the processor could still work if an SPE were to fail when the processor was in use. However, if you do find an article then feel free to prove me wrong.
    – James P
    Aug 2, 2012 at 16:11
  • I agree, I was fishing for confirmation at risk of downvoters. Thanks for not being a twit :)
    – jon
    Aug 2, 2012 at 18:20
3

how a CPU operates when looking at fundamental operation of CMOS requires and understanding that CMOS slew rates cause heat dissipation and rising temperatures reduce the slew rates thus increases slew rate even more and propagation time increases as well. If there is a set margin in timing before a race condition then it can be said with constant clock speed that the MPU may run slower rise times and increased clock delays so the margin before lockup due to a race condition in the chip or external memory may cause failure. This explains why MPU's that run hot will work after a cool down period.

Apparent aging of CMOS gates can occur if moist dust accumulates on the exposed bus soldered lands. This can add many pF of loading which can reduce the rise time of bus signals and increase the internal heat dissipation causing further reduction in slew rates.

Another cause of apparent aging is the increased number of background tasks installed by the user startups and resulting in excess heat during so called idle activity. trimming the startups can reduce the overall CPU load and thus restore normal temperature rise due to excess processes running. For example XP on clean install of a retail version might have 25 processes running and an OEM version with many user auto-installed services and startup processes in the registry, might increase this number of processes as shown in the TaskManager process Tab to say 50, and even up to 100 from my experience of inexperienced users. DIsabling these processes using simple programs such as MSConfig can help, but WinPatrol is even better and free and restore cool operation as a new.

As pointed out by others, there are internal failure mechanisms which also slow down the slew rates of gates called time-dependent dielectric breakdown from ElectroMigration growth on the semi-conductor material. This is dependant on stress levels of heat and voltage and also exposure to gamma radiation in space.

All of these factors contribute to why the temperature rise and loss of time margin occurs in laptops from aging, even after a fresh install of OEM image. So 5yr old latops will run hotter which means they must have longer slew rates and thus elevated temperature rise above ambient and that means it must be running slower rise times. But the clock rate is fixed so the performance if working will be the same until the margin drops to zero without warning. So monitor your temperature rise and do not exceed 70'C for reliable operation is my best advice. 60'C is preferred maximum where most CPU fans start to run at full speed.


There are many reasons why CPU's get hotter with aging. One reason requires and understanding of complementary switching. Simply put, it is a synchronous pull up switch that turns on while pull-down shuts off. During the interim there is a momentary short circuit if there is a crossover from unequal slew rates or switching times. New technology of CMOS may compensate for this characteristic that is temperature & voltage dependant to introduce faster switching times but with a controlled dead time to eliminate transient power loss during crossover. Although ElectroMigration is one reason of additional delays, it is not obvious if this is symmetrical.

Never-the-less CPU temperature rise is a widespread phenomena with aging {with laptops sensed by users lap of gradually getting hotter over the years} and this helps to explain the reasons. i.e. aging cause gradual slew rate increase which affects dynamic power consumption of a steady clock frequency or a repetition rate of cross-over transitions. Since we know the steady state leakage power is negligible, it is this effective driving force of complementary outputs with momentary current surge that drives CPU temperatures up. So CPU idle temperature is a strong indicator of aging or slowing down of slew rates if everything else is constant.. (CPU load, V+, ambient temp, cooling efficiency, dust elimination) YOur CPU will still execute instructions at the same speed but run hotter and thus with less timing margin before a race condition occurs. ( read data when not ready due to propagation delay)

The same phenomena exists in desktop CPU's but users may not be aware of gradual increase in fan speed over the years that compensates for increased heat dissipation from gradual aging. There is no empirical study to my knowledge, but it is my personal observations of CPU's over the last 20 years that this happens in many cases, but not all.

1
  • This was a very insightful answer! Your comment: "But the clock rate is fixed so the performance if working will be the same until the margin drops to zero without warning" supports my understanding that the CMOS gates do get slower with more usage, however this is masked by the clock rate. As long as the gate performs within the timing margin, the CPU will operate as normal. This margin decreases over time however, as the gates age. Aug 2, 2012 at 20:29
2

A few extra bits and pieces about some of the other answers.

  1. crystals can/do slowly drift over time, but they're much more affected by temperature than time. For example, right when you turn on the machine, it's probably running a bit different speed than when it's been running for hours. These differences are, however, much too small to be perceptible.

  2. It is entirely possible to have an intermittent failure in connections on a chip. When fabricating a chip, they (obviously) do their best to prevent this, but it's still possible and does still happen. As chips have started to run hotter, this has become more common. When/if this happens, however, it's a lot more likely to cause the machine to shut down completely than run normally, but slower than it did. That's not to say that a slowdown is impossible, just very unlikely.

  3. While self-correction can detect errors and shut down parts of a CPU, the CPUs in (at least most) current PCs don't include such capabilities. For this, you're looking at either a high-end mainframe, or a PC of the future (though, admittedly, not all that distant of a future any more).

1

Though this has very little to do with everyday life, there is a concern about electronics component aging. In a nutshell, and this is true for any electronic component or system :

  • If your CPU has worked a few hours (which founders have them do as part of factory testing, a process known as burn-in) without fault, it will last the same for years. Probability that it will fail during this time are close to 0
  • After several years, failure probability starts increasing, it's time to change your CPU. In consumer products, this typically happens after the component has long been obsolete so you don't really worry about that
  • If you like maths, have a look at http://en.wikipedia.org/wiki/Failure_rate

So: yes, if your CPU is very old you can guess that some component of the CPU (some cache tat does not respond and always make page faults; or a CPU core that is lost) can slow it down. But you'll most probably will have better success looking elsewhere.

Also, keep in mind that a computer has many big or small components that age much faster than the CPU. Including :

  • hard disks with mechanical parts that wear off
  • connectors that corrode
  • heatsinks that move and get dusty
  • chemical condensers
  • weldings that corrode or move through vibrations
1
  • 1
    sorry, but I downvoted this because failure != slowing down, unless taken in the context of jon's answer.
    – Sirex
    Aug 1, 2012 at 23:51
1

If you are not cleaning the heatsink and the fans you cpu will get hotter and system performance will be slower. Since dust particle take some time to settle in those areas, we feel like in time cpu speed and performance are being reduced.

2
  • This should be the correct answer! Pragmatically, if you own a laptop for two years, it will be slower. Why? Because you never cleaned the gunk out, the ventilation is worse, and the CPU gets hotter. The circuitry is still the same, but the CPU is now receiving more frequent messages that say "woah buddy, you're too hot. Slow down!" Aug 2, 2012 at 19:02
  • 1
    This answer introduces a variable for accumulated dust on the cooling device - one which I didn't want to factor in. I was asking more in the technical sense about the degradation of the internal hardware of a CPU. A hotter CPU will most likely have a shorter lifespan, but I was looking for a more informative answer as to the internal forces at work. Aug 2, 2012 at 20:33
0

Yes it does — it depends upon the usage of the user, the harddrive is the one thats get's aged soon as it gets infected with bad sectors as it ages.

Then, when higher end programs run on the old configuration it sucks the maximum visuals, so it gets slower, and as ages pass by, your technology increases where your system could not meet the software requirements... so it's said that your/our system gets slower when it ages.

1
  • Welcome to Super User! Please try to keep your cookie or register your account, so you can edit your posts in the future.
    – slhck
    Aug 2, 2012 at 12:01
0

Heat is the single most important factor in CPU speed. This said, depending on what CPU is in your machine it might dynamically reduce speed to stay within a "safe" temperature range. Most CPUs can do this. You might not know it's happening. The temperature is however not something that should go up with age, if you clean the heatsink regularly, and the thermal paste is not improperly applied.

-1

This is debateable. It depends. Generally as per the theory, Simple NO. But depending on your hours of usage, load on CPU power supply and external power status, if working without UPS, the motherboards degrades and so the load on CPU might rises. But working under ideal conditions it will be as same as new. Since the CPU's contains billions of transistors across inside and so over the time, if their performance degardes by any means it will degrade the CPU performance. So in general sometimes we face system slowdowns even after fresh installations.

But in general its no.

Not the answer you're looking for? Browse other questions tagged .