2

Every now and then questions come up that ask for good ways for an end user to "stress test" a particular type of hardware. This is not one of them.

Rather, I am asking whether end user stress testing of hardware serves any actual purpose in practice?

Beyond things like running Memtest86+ for one or two passes to check that the RAM sticks are still OK after installation (no static discharge during installation, for example), one SMART conveyance test on the hard disk to verify that nothing slammed against it during shipping causing damage, and so on, with new hardware as shipped by a reasonable manufacturer, is there some likely, early-life failure mode which is likely to be detected by stress testing but not be detectable by normal use?

In order to keep this from being "opinion-based", please limit answers to what you can provide specific references for, and provide references for claims made.

user
  • 29,910
  • "In practice" of what? Product development or engineering? Or as an end-user? QA and DVT would perform stress tests to validate either the design and/or production processes. – sawdust Oct 21 '14 at 07:42
  • @sawdust I felt that was covered by my example mentions, but have edited to make it clear as well that I meant stress testing by the end user. – user Oct 21 '14 at 07:44
  • "is there some likely, early-life failure mode which is likely to be detected by stress testing but not be detectable by normal use?" You are missing a key point. There is some small percentage of systems that die soon out of the box. You could spend days getting stuff installed, configured, customized, tweaked, etc., and if it was one of those systems, you would lose all of that work. If you run tests for that period and it fails, you don't lose your time. The failure would be detectable by normal use, which is the reason to test it instead. – fixer1234 Oct 21 '14 at 09:04
  • @fixer1234 How does deliberate "stress testing" do more to detect those errors than booting to the Windows desktop (or even BIOS/UEFI setup) and then just letting the system sit there for several days? Or running Memtest86+, SMART conveyance tests, etc.? – user Oct 21 '14 at 09:09
  • My point was that if the system happens to be the one that will fail early with use, doing some unattended activity that exercises the system to trigger that failure is preferable to an investment of time (and potentially loss of files), to detect it through normal use. Running some/any activity will be better at triggering failure than sitting on but idle. Running something that stresses the system will be that much more likely to trigger failure. Different components can fail. The more of them you stress, the more likely of finding (triggering) the premature failure. It's statistics. – fixer1234 Oct 21 '14 at 09:28
  • @fixer1234 That doesn't really answer my question though. Memtest86+ stresses RAM and to a minor extent the CPU. The computer just sitting at full throttle stresses the CPU cooling (try booting into firmware setup and watch the CPU temperature rise). SMART conveyance tests stresses the hard disk. And so on. I'm asking what stress testing beyond that adds, if anything. Also as pointed out by sawdust QA testing should already take care of most failures not related to shipping. – user Oct 21 '14 at 09:39
  • Stress tests are useful when you're running high performance parts and intend to push parts to their limits. Something like furmark or a CPU stress test will let you know how your system will behave at the extreme end of its performance envelope. You'd have an idea if your system cooling is sufficient (By getting an idea of how hot your system gets), and if your cooling and power systems hold up to the strain. At the end of the day, its really personal preference. My system is pretty much hitting 99% gpu usage on some newer AAA games, and a quick 2-3 hour burn-in replicates that decently. – Journeyman Geek Oct 21 '14 at 10:58

2 Answers2

1

Some miniscule percentage of computers will fail prematurely. Your question isn't whether it's worth testing at all but if you test, is stress testing better. The starting point is that your machine is the one in 10,000 that will fail prematurely and your goal is to cause that failure. What is the best way to make that happen? (When it doesn't fail, it has passed the test.)

There are a bunch of different parts that could be what fails and the trigger could be simply energizing the system for a short time, it could be heat in general when the system warms up, it could be a specific component getting hot, it could be mechanical (vibration or thermal expansion and contraction), etc. There is no way to know which part or which trigger will be the cause of the premature failure.

If you just turn the system on and let it idle, that might be enough to trigger a problem sensitive to energizing the system or general temperature. If you run Memtest86+, that will test the RAM. If you run a comprehensive stress test, that is likely to push many more triggers than not running it. The more things you test and the longer you test, the greater the likelihood of finding (causing) the failure.

I would venture to say that no statistics exist (at least available to the public), on the actual benefit of testing in any specific way. The premature failures are too rare to generate meaningful numbers to tweeze apart the benefits of one kind of testing over another, or the benefit of testing for one time period or another.

If you happened to be the unlucky soul who got that ill-fated computer and you spent a lot of time setting it up and accumulating critical files and then it died, you would probably be pretty miffed. So is it worth doing some type of out-of-the-box test before you start using it? What could it hurt? If you decide to test it first, the most useful test would be the most comprehensive stress test you can find because that is the purpose of the exercise.

Absent any solid data on relative benefits, you have to make a "management decision" on how much testing will satisfy your comfort level. Something to ponder: suppose you test the heck out of it with no problems and then spend time setting it up and accumulating critical files and then it dies anyway. Would you be more miffed than if you hadn't spent the time to test it first? That's another decision factor.

fixer1234
  • 27,486
0

No. A simple cursory test will indicate if the hardware is doomed to fail early, and no stress testing is necessary. Stress testing is done on prototypes to ensure that the majority of market-bound products will not fail in due time. This is the purpose of a "manufacturer's warranty." They've already tested a sample of their products, so you can be sure that the warranty is the minimum amount of time they're confident that the device will last, barring abuse (e.g. running at 100% constantly) or manufacturer's defect. In fact, if you do burn out a piece of hardware after a week of stress-testing, they might claim your warranty is void.

Hardware is meant to run comfortably around 80% capacity under normal circumstances. The remaining 20% should be considered a "burst" tolerance. The hardware is not expected to fail under short periods of extreme usage, so long as the average usage is within tolerance. You can compare this to a car. A car might have a maximum RPM of 8000 on its engine, but the "red line" will be lower, perhaps 6500 RPM. Running the car consistently above 6500 RPM will shorten the engine's life. Similarly, stress-testing a piece of hardware will only cause earlier failure somewhere down the road.

File systems that are above 80% capacity will reduce the OS's ability to write new files efficiently. Power supplies above 80% capacity will wear down the internal hardware quicker, failing faster. CPUs that constantly run above 80% capacity will likely overheat sooner, or operate at diminished capacity while the hardware cools down. Each part of a computer is built with a maximum tolerance, but it is not expected to be the daily norm.

Generally, hardware run at a normal rate will give indication that it will fail far sooner (with more time to repair/replace) than hardware that is run at maximum capacity all the time. CPUs will start to become slower, power supplies will have periodic "brown outs", and the time to save files to a hard drive will increase. If you run the same hardware at capacity all the time, odds are, you won't notice a problem until its too late to do anything more than replace the hardware, even if its a short stress-test.

It's better to buy hardware that has more capacity than you need, than to buy only the capacity you expect to use. In the long run, such purchases will result in a longer life for the hardware. Hardware you do buy shouldn't be abused, as even a simple week-long stress test can reduce the life of the hardware by several months or years down the road.

phyrfox
  • 3,053
  • Great answer and personally I'd agree with your claims, but it lacks those references the OP was talking about. Do you have any "proof" or references to go with your answer. – agtoever Oct 21 '14 at 10:03
  • It's everywhere, and nowhere. Know that you can't defrag a Windows drive past 90% full. PSU manufacturer's recommend you buy a unit to run at 80% capacity under normal loads. Most large database vendors suggest you keep at least 20% physical memory free. New laptops come with "Battery extender" software that caps charge to 80%. This number appears practically everywhere in literature, so much it's hard to ignore. Engineers have decided that 80% is the magic number, ignore at your own peril. – phyrfox Oct 21 '14 at 10:33