17

I have a Ryzen 1700X (got it on launch), running on my main Linux machine, with other components that are on par with it. During the day when the computer is on and I'm not using it, I like to run BOINC client on it, which maximizes all 16 threads of the Ryzen to 100% without a single one falling back, and it can maintain that for hours. My question is, if I were to run the computer like this, ignoring the electrical bill and the failiure of the PSU (which isn't yet to come, my 1050w 80 gold is running very smoothly), how long can I expect the CPU to run for before significant damage is done to the silicone. I haven't overlocked it, not even to the "AMD guaranteed +400M Hz (OC on Ryzen seems unstable), and its cooled by Corsair Hydro H60, which is acceptable, but I have many fans setup which never let the temperature rise above 56 C when the CPU is at 100%.

9
  • 8
    Much older generation so I'm making this a comment rather than an answer, but my Uncle has a 286 in his lab which was bought brand new in the mid 80's to run a long term experiment. It has been controlling said experiment ever since (he's a geologist, it's barely got going) and shows no signs of failure. Nov 29 '17 at 10:27
  • 1
    @JosephRogers What kind of experiment needs constant monitoring for 30 years?
    – user137
    Nov 29 '17 at 10:55
  • 8
    @user137 The Pitch drop experiment, for example ;-)
    – flolilo
    Nov 29 '17 at 11:14
  • 2
    significant damage is done to the silicone Common mistake ;-) Silicone is what is used in certain implants, kits etc. Silicon (no e) is the metal used for making chips. Nov 29 '17 at 13:12
  • 2
    @WannabeCoder that’s my guess based on the fact that the POST produces PEI error code 0x54, “Unspecified memory initialization error”. The memory is fine (tested in another system), as is the motherboard (tested with another CPU), and the CPU produces the same error code in another system. Nov 29 '17 at 14:33
18

A CPU itself will probably last tens of years.

The problem with system reliability is actually all the other components in the system. The CPU will undergo thermal heating and cooling cycles as the load changes and as you turn the system on and off. The same is true for pretty much every component in the machine. The CPU silicon die is connected to pads or pins on the CPU by tiny bits of wire which will be affected.

Hard drives will spin up and spin down, resulting in mechanical stress on the motors and thermal stress as the drive power controller warms up and cools down.

Damage to the silicon in the processor is unlikely to be any kind of failure mode you will see, the voltages are low and the paths are well designed. It is only in NAND Flash memory devices that you are likely to see a failure due to silicon insulation breakdown, but then that is because they are intentionally driven to cause a non-catastrophic breakdown (but it eventually will be catastrophic).

You are more likely to see problems due to thermal or mechanical stress, as temperatures change and components move very slightly, than you are to see any failure within the actual CPU silicon.

2
  • 2
    Since he is running at 100% load, thermal expansion is less of an issue for him.
    – PlasmaHH
    Nov 29 '17 at 10:15
  • @PlasmaHH While the CPU might be running at 100%, other components may not be. There may be only rare or intermittent load on hard disks or other devices. I do mention that the CPU will likely last a long time, but was trying to get across the fact that it is actually the rest of the system that it likely to die sooner.
    – Mokubai
    Nov 29 '17 at 10:59
24

CPU failures from ordinary use are very rare. Manufacturers typically provide the default "it isn't going to fail" figure of 100,000 hours, which is just over ten years. But most likely, it will continue to work until it's technologically obsolete.

13

Thermal failure is the only mode I've seen personally for a CPU (as opposed to the support equipment). The last time I saw (or heard) of this was with an old Pentium III laptop that went on fire (literally) one morning.

Thermal management is now a much more sophisticated operation in systems and so CPUs are not allowed to run too hot. Typically a CPU will throttle it's clock to reduce temperature if need be, and the whole system can shut down if it detects a temperature that's too high.

So I think with modern CPUs failure (before obsolescence) not a real issue.

Motherboard failures seem more common (but still comparatively infrequent) than CPU failures. Drive failure is common as dirt, particularly in heavy duty server settings. I'd not give a thought to the CPU any more.

2
  • The laptop caught on fire? How much dust was built up inside that thing for the CPU's heat to set it alight? Is that a common occurance with Intel CPUs? worries about 5+ year old Acer laptop Nov 29 '17 at 14:05
  • 1
    That particular system had a (very) hot spot all the time I had it - a design flaw, I think. I don't think it was dust that went on fire, but probably gas or vapor released by the overheated material. This was fifteen years ago or so. I wouldn't worry at all about a modern laptop. Intel and AMD introduced speed throttling tech in 2005 and 2003 respective and laptop thermal design has advanced quite a lot.
    – StephenG
    Nov 29 '17 at 14:28
3

CPUs usually fail because of overheating or some power(stroke) issue. The only problem i see with CPU running for years is, that heat sink will fill with dirt and clog. and won't be able to cool it properly. As long as you keep heat sink in proper working conditions CPU will be fine.

1
  • 2
    or some power(stroke) issue Indeed, as an Electronics Engineer I see the PSU as a more important point of failure (compared to the CPU) as the components in the PSU usually are under more stress than the CPU and motherboard. Also fully agree with the regular cleaning if not in a dust-free environment. Nov 29 '17 at 13:15
3

In the 14 years of assembling and selling industrial PC's, I've seen one or two CPU's dead (out of thousands sold). But, everything around the CPU can be a different story.

Heatsinking is generally the part where computer makers often cheat - and it's difficult to tell negligence from planned obsolescence. With sloppy thermocoupling of the CPU to the heatsink (especially in passive/fanless computers) it's not the CPU that dies - it's stuff around the CPU. VRM elyts used to be a classic, then there came a period when they were a non-issue (when each aluminum wet elyt was replaced by a comparable solid-poly), then the board/computer makers pressed on and reduced the number of solid-poly capacitors in the VRM, or made the VRM all-ceramic but running extremely hot, or some such... Modern ceramic capacitors (MLCC's) are no longer immortal, and the dependency of MLCC lifetime on operating voltage and temperature climbs along 2nd or 3rd power (some say even higher). Another rule of thumb is, that every 10*C down in temperature mean twice the lifetime.

BGA packages combined with lead-free soldering - those are a pain, especially in applications where the BGA packaged chip runs hot (the heatsink is small, or big enough but not properly thermocoupled to the chip) and where the temperature keeps changing up and down.

My ancient DIY/HAM sense still seems to be valid: if you can keep your finger on it, and it doesn't smell, it has some chance. I like designs where the heatsinks only ever gets lukewarm (and where I know that the heat source inside, typically the CPU, is well thermocoupled).

Not the answer you're looking for? Browse other questions tagged or ask your own question.