Best way to test new builds and fresh OCs

[Zka17]

I'm opening here a new thread to discuss a subject which may be really helpful for many of us. In this forum (different sections of it) we can meet some very often/familiar situations: somebody just finished building a new system and/or just experimenting new overclocking (OC) settings. In both cases, natural questions raise: Will it work properly? Is it stable enough?

So, the question to discuss is: How to test new builds and fresh OCs?

To start the discussion, I would mention two possibilities raised over the last few days in other threads:

1. Based on my - I recognize, limited - experience, I would use synthetic stress-testing softwares for assessing system's stability after a new build and/or OC'ing. Primarily I'm thinking here about prime95 (for mostly CPU and memories) and FurMark for the graphic cards. I admit that both of these softwares will max out the load on the components in a manner that practically no other applications could do. My idea to use stress-testing is based on the assumption that if a system can withstand this, then it would be no problem for everyday use.

2. I learned recently, from chrsplmr (a long time, very experienced member on this forum), that there is an other method. Loading many applications (net browsers, image-handling softwares, games, videos) to fill up the memory practically, put the system to sleep - then wake it up! If it wakes up with no problems, then testing is considered successful - hope, I got it right, Chris (please, join and share your thoughts)!

HiVizMan (one of our respected moderators, also with extensive experience on the field) already shared his doubts using stress-testings in other threads - I hope, he will give us some explanations and hints too...

So, my fellow ROGers, let's share some thoughts, ideas, experiences about this topic! Thanks in advance for all of those who will participate, "enjoy!" for those who will read.

[xeromist]

With overclocking synthetics are a decent way to provide reproduceable and focused conditions every time. They may not align with real world usage but if you are overclocking a specific component it is important to stress that part in a similar manner each time in order to find its limits, then you move to the next component.

I agree on Prime and Furmark, although there are other perfectly valid choices. LinX is pretty good for CPU and the FutureMark tools are certainly a more fun benchmark while providing a decent amount of stress.

2) Yes that will work as a general S3 test but honestly I prefer Memtest86(+) for purely memory testing(if that's what you meant). Once again, consistency and targeted stress are the key. You could load up a ton of apps in Windows to produce a crash but what is the cause? You'll still have to do more testing to determine what went wrong. But if you've already tested the CPU, memory, etc and you still have sleep/boot problems the test you mentioned might be a good way to test minor voltage tweaks.

[chrsplmr]

.. I said that is what 'I' did. I think the whole point
of ROG is to climb to whatever level you seek.
I want only to run what I need.
For myself all that software ect .. is software.
Not my software .. I need it to do all that I do..or whim to.
If the V had an Extreme version I'd be ROG'n an E.
So I asked -- for this 'rig' What would be the most
Extreme demands that 'I' would put on 'This' Machine ?
Thats what I did .. put it to sleep .. came
back pressed the space bar .. back to 'whatever' ... and now..
I have done that @least 2000 times .. minus the extreme of course.
(but close..hehe)>>>see for yourself>>>"Joe ROG Consumer Review" CrossHair V Formula
[[Then I did this Review for a T-Shirt ..hahhahaa..to prove V was All ROG without the FX..still is.]]
For those that go to other Extremes .. oooh thar is aPlenty ..
with their boards and set ups .. pinpoint accuracy is a must.
Some seek and know the sweet spots through other means.
I asked advice, got it running and havent had to look back.
What works for you and you are comfortable with. Works.
Good Question.

:::EDIT TO BELOW:::
HiViz++

[HiVizMan]

Not sure if I have extensive experience in stress testing at all. As those that know me are aware I am not a big fan of the belief that running Prime for 24 hours is an indicator of anything other than your system can run Prime for 24 hours.

The starting point for these kinds of discussions I believe needs to be working out what it means to be stable. And there lies the rub. Stable for a gamer is very different than stable for a technician responsible for a server room at an financial house say. Entertainment versus enterprise. The one wants stable for fun and the other has to have stable for work. For me stable is a very subjective set of behaviours. Simply put, my system is stable if it is able to complete those tasks and activities I use my computer for. Nothing more and nothing less. I do not care if my system fails Prime after 5 minutes or 5 days all I am interested in is being able to do what I do with my system and not have it crash.

When I build systems for people I do certain things in the process that cuts out much of the uncertainty that others will use prime to resolve. Like xeromist suggested memtest is a key part of the process for me. Before I do anything with the core frequency of my processor I establish my memory stability. I use memtest to achieve that. I will run multiple sessions to ensure that my memory is rock solid before I have ever booted into OS with that CPU. Memory is key to so many of the functions that we take for granted in modern systems. The different sleep states for example are quite picky and will fall over if their is a weakness in the memory. Memory with modern systems is the key element to stability. Prime does test that. No doubt about it. So why am I not a fan of Prime then.

Well it is simple, the extended 100% stress that is placed on the CPU itself. The cores of the CPU are all run at 100% for the duration of the test. That I do not like. I have run my SandyBridge at 5.7GHz since February last year. It needs no more voltage to run today as it did when I first bought it. I have never run Prime on that CPU and never will.

I struggle to take on board what good has been achieved by running Prime for these extended periods of time, the memory I can test with memtest and that will not cause any degradation of damage to my ram (and certainly not my CPU) the CPU I can test with real world things like the applications I am planning on using. Or doing something similar to that which CP shared, flood the system with tasks and do the sleep test. Now with Prime I actually feel sorry for the CPU's being subjected to 24 hours and sometimes even longer periods of 100% usage. That is never going to happen in the real world, heck I challenge folks to show me what application will max out all the cores and threads of those processors that are HT enabled in a sustained way. For no other reason so that I can avoid them.

Look I respect the work and effort that folks put into getting their systems to a point where they can run 24 or 48 hours of Prime and not fail. Outstanding guys really I mean that with no irony at all. As I know how much effort needs to go into setting up a system.

For me it is not something I would aspire to do or ever recommend others to do. For me all that 24 hours of running Prime (as an example) demonstrates is that the system has completed successfully a sustained run of Prime. Maybe it falls over in some other applications. I have seen that happen more than once myself. Will I advocate hours of Prime to ROG members to see if there system is 'stable' no, what I will suggest is that they play the most taxing game they enjoy. Or do the most taxing task with their system and see how it goes.

Just for the record you can substitute LinX, Intel Burn Test and any of the other tests that are similar. I do have nothing against Prime and simply used Prime in my reply for continuity sake.

[xeromist]

That is a good point. I don't think I've ever run Prime for 24 hours. I'm more concerned with a duration of a few hours or less. If it doesn't fail within that time then it's good enough for gaming. The most extended stress test I do is 24/7 Folding@Home.

[Anicra]

Always review Event Viewer for non OC related problem look for ! (critical under system) and LinX I always run for 30 runs or Prime 95 for 24 to 48 hours.

[Dimsum]

HiVizMan,

I really like your post. I have always thought people expected a system to have to run on these stress testers before being considered "stable" but you are absolutely right, I never have ever stressed my cpu or system as hard as these tests push them. I have overclocked some prior systems that never could handle prime or linX but never crashed under daily use. I just got use to the idea from reading forums that a system had to pass these tests before people would consider you "worthy" You brought up a lot of great points. Thanks for sharing!

[Zka17]

Wow, a lot of precious info... Thank you, guys!

Now, let's see if I got it right...

I agree with HiVizMan's statement: "The starting point for these kinds of discussions I believe needs to be working out what it means to be stable." As he said, "stability" may be a very subjective thing... I guess, from now on when we are talking about stability we also have to define "for what". I mean, the system has to be stable for gaming or for everyday's use (simple documents, web browsing, music, videos) or 24/7 working or just simple OC'ing testing, and so on - and of course, any combination of these.

Stability practically means, also as HiVizMan said, that the system can perform the specified tasks.

Based on these, there are endless ways of stabilities... and, perhaps, none of the testing method could comprehend all the testing needs alone.

So, practically there are different levels of testing. For new builds it seems very logical what xeromist and HiVizMan said, that to check the memories - use Memtest even before loading the operating system. If passes, the next step could be chrsplmr's way of testing: the sleep method (start all the applications intended to be used and put the computer to sleep, then wake it up). For overclocking and/or troubleshooting, I think, xeromist made a good point by stress-testing "may not align with real world usage but if you are overclocking a specific component it is important to stress that part in a similar manner each time in order to find its limits".

Now, some questions:
1. What is/would be the correct way to test with each method? How long has memtest, prime, furmark, lynx, etc. has to go until we can consider the system/specified part stable? What are the best settings of these softwares for the best/most reliable results?

2. What is the problem with running stress-tests for extended period of time? Can it hurt the components? In special, I'm thinking here about prime95 - loading it 12 or 24 hours can hurt the CPU or the mobo?

[HiVizMan]

Best practices for stability testing - who would have thought we would get to this point.

1. There is no universal best method, the method has to fit the need. Memtest for example can be run just once for a clunky idea of what is possible, or two three times depending on the time available or desired out come. I will let my final setting run for 3 cycles as that has not let me down as yet. Memetest is good in that it will not degrade or potentially damage the CPU even when running over night say.

Which is a nice lead in to question 2.

Short answer from my perspective is that yes it can. Do I have metrics at hand that can support that thesis sadly not. Does that absence of empirical data invalidate my thesis no it just means I can not validate my position, therefore it remains simply a thesis. And no I have no intention of taking a suitable large sample of CPU's and potentially damaging them to find out definitively. I just do not care enough one way or another to do so. I make the subjective call not to take that risk based on anecdotal accounts and personal observations. I have touched on the 100% loading for the entire duration of the extended test and my doubts of the value that provides as no real world PC usage will mimic that in my initial reply so there is no need to revisit that I think.

[Zka17]

I've got your point about not using stress-testing, I trust your experience - 3 cycles of memtest are included in my testing-ptocol from now on!

But, as I said, seemingly there are several ways/different levels of assessing "stability". I was using and suggesting stress-testing based on my experience - which obviously is way less with the computers than yours. Since we can see that there are others who are using stress-testing too (like Anicra above using prime for extended periods too), I thought that would be helpful to know if there are bad sides of it.

By asking if stress-testing could affect different components, I didn't expect a large scale "testing-by-distroying"... I was just curious what it could cause? To what we can expect if using these stress-test over longer periods? What would be the signs of damages (excluding the raising smoke )?