What do you use to test stability?

What do you use to test stability?

  • Prime 95 Smallest FFT with AVX2

    Votes: 1 4.3%
  • Prime 95 Small FFT with AVX2

    Votes: 1 4.3%
  • Prime 95 Blend with AVX2

    Votes: 4 17.4%
  • Prime 95 Smallest FFT no AVX2

    Votes: 1 4.3%
  • Prime 95 Small FFT no AVX2

    Votes: 2 8.7%
  • Prime 95 Blend no AVX2

    Votes: 0 0.0%
  • Cinebench 20

    Votes: 2 8.7%
  • CPU-Z stress test

    Votes: 0 0.0%
  • Other

    Votes: 12 52.2%

  • Total voters
    23

M76

[H]F Junkie
Joined
Jun 12, 2012
Messages
14,006
prime95-1.png


People always gloat that it is not stable unless it can run prime95 But it has multiple settings and I rarely see anyone clarify.
So I wish to know what is your choice.

How many hours is a question for another day.
 
I always used Blend, because I thought the idea was to push not just the CPU, but also the RAM, since its stability would play a part in overall stability of the system and would be affected by the overclocks
 
Usually the Blend with AVX and at least one other prime95. Then a ridiculous scenario of me attempting to run two different heavily multi-threaded games and a 4k video simultaneously all on different monitors.
 
I voted "Other" because to stest stability I use Tequila. Just kidding. Scotch.
 
View attachment 201639

People always gloat that it is not stable unless it can run prime95 But it has multiple settings and I rarely see anyone clarify.
So I wish to know what is your choice.

How many hours is a question for another day.

Small FFT with AVX is more irrelevant as a stress test than before, because the chips become completely impossible to cool unless you are using non conventional cooling methods (direct die, multiple 360 rads, etc). You can't say this is a valid stress test if your chip needs enough voltage so it runs at 115C just trying to pass it without a BSOD, CPU L0 error, or a dropped thread. It's a thermal solution test, not a stress test now.

FMA3 testing (AVX2) is even more worthless because NO game uses FMA3 instructions. Or AVX 512 instructions. I am not even sure if any game uses AVX2 instructions. I know AVX2 is part of directmath, but still. People MUST remember that if your prime95 FMA3 (subset of AVX2) reaches 105C and crashes and burns, that does **NOT** mean that a game that uses FMA3 (if it exists) will crash! Not even close. Why does Prime95 small FFT FMA3 reach 105C on 8 core processors? Because the instructions REMAIN in the CPU caches and do not ever access main memory, so the CPU never has a chance to cool down or run other instructions. That is NOT how real applications work (except applications DESIGNED to crush prime numbers, which games do NOT do).

Even SSE2 4.2 is a bit of a stretch. Phenom doesn't support it, for instance.

If you're trying to be Battlefield 5 game stable, Realbench 2.56, Cinebench R20 and HWbot x.265 or x.264 stress tests are a lot more realistic.
 
Last edited:
Small FFT with AVX is more irrelevant as a stress test than before, because the chips become completely impossible to cool unless you are using non conventional cooling methods (direct die, multiple 360 rads, etc). You can't say this is a valid stress test if your chip needs enough voltage so it runs at 115C just trying to pass it without a BSOD, CPU L0 error, or a dropped thread. It's a thermal solution test, not a stress test now.

FMA3 testing (AVX2) is even more worthless because NO game uses FMA3 instructions. Or AVX 512 instructions. I am not even sure if any game uses AVX2 instructions. I know AVX2 is part of directmath, but still. People MUST remember that if your prime95 FMA3 (subset of AVX2) reaches 105C and crashes and burns, that does **NOT** mean that a game that uses FMA3 (if it exists) will crash! Not even close. Why does Prime95 small FFT FMA3 reach 105C on 8 core processors? Because the instructions REMAIN in the CPU caches and do not ever access main memory, so the CPU never has a chance to cool down or run other instructions. That is NOT how real applications work (except applications DESIGNED to crush prime numbers, which games do NOT do).

Even SSE2 4.2 is a bit of a stretch. Phenom doesn't support it, for instance.

If you're trying to be Battlefield 5 game stable, Realbench 2.56, Cinebench R20 and HWbot x.265 or x.264 stress tests are a lot more realistic.
Who said anything about games? I want my system to be stable in every possible real world scenario it might encounter, not just games. And especially not select games.
 
I’ve seen prime fail after 24 hours, so to me it’s a waste of time. Also, if you can’t run an app because it makes your cpu too hot then you shouldn’t be running at that clock speed.
 
depends on the needs of the box. Typically I’ll just play a few games and keep an eye on it over a few days. If there’s issues, I’ll re assess, otherwise I’ll keep it rolling until that day comes. My 2500k rig is starting to dislike its 1ghz oc so I’ve had to tweak a lot more lately. And that’s all I’ve done. A little instability does not bother me.

For something I or a family member where dependability matters, I’d either not OC or run occt and furmark for a bit and watch temps. No bench will prove a completely stable system, but as long as temps are well within spec, I know I’m not throttling at least.
 
I’ve seen prime fail after 24 hours, so to me it’s a waste of time. Also, if you can’t run an app because it makes your cpu too hot then you shouldn’t be running at that clock speed.

Prime95 is NOT an "App". Neither is LinX 0.9.6
They are power viruses. Prime95 can be used as an app if you know what you're doing with the test options.

With that logic you wrote in your post, no one should be overclocking their video cards if they burn up the VRM's running furmark.
Come on, man.
 
Well, not sure what to tell you, I can run them. Yeah things can get toasty if I am at the limit of my hardware, but I can run them. There's no difference between now, and the good old days. If it makes it too hot, you gotta turn it down. Sure you can use offsets now a days when certain instructions appear, but to me, that's cheating.. like just to be able to say me too. Fake it to make it, know what I mean? Sorry I'm not trying to be a jerk, that's just how I see it. If what you do works for you, all the power man, I'm not going to tell you how to run your pc. I set mine up so that I can run anything I throw at it. And if it makes it too hot, then I gotta turn it down, because I just use straight voltage, no offsets. My GPU is kind of out of my hands, it does its own thing, usually 1 bin higher than I set.. but its just stock voltage. I am not worried at all about my VRM's in the slightest, as I have many, many CFM's at my disposal.
 
With that logic you wrote in your post, no one should be overclocking their video cards if they burn up the VRM's running furmark.
Come on, man.
If your car spins the tires you need bigger tires, you can't blame the asphalt. So if your CPU gets too hot you need better cooling, and not "let's avoid doing that test then". It's a similar situation as VW cheating on emission tests.

A stress test is supposed to be more demanding than real world situations, otherwise it wouldn't be much of a stress test.

I'd rather have 200Mhz smaller epeen, than have my CPU crash during a 24 hour render.
 
I'd rather have 200Mhz smaller epeen, than have my CPU crash during a 24 hour render.
This right here is why it's almost impossible to say what the "best" stress test is. Use case is so fundamental in the choice.

If your overclock isn't stable, you may lose a day of render time. Of course you're going to want to most demanding test and ensure that your system will not crash under any circumstances.

If my overclock isn't stable, my leisure time gets one reboot's worth of interruption. Maybe I lose a few minutes worth of work on a 3D model. I can be much less stringent with my testing methodology because the potential impact is so small for me.

This hardline "there is only one best way to stress test" mentality has got to stop. It's not reasonable.
 
  • Like
Reactions: noko
like this
This hardline "there is only one best way to stress test" mentality has got to stop. It's not reasonable.
When I say the best way I mean the way that ensures stability in every real world scenario, while not testing for loads that are entirely impossible in real world application.
If a stability test is only good for gaming application then it is not good in my book. Then again there can be wild differences from game to game too. So I wouldn't trust any game as a test. The only way to test for all around stability is synthetic stress tests.
 
When I say the best way I mean the way that ensures stability in every real world scenario, while not testing for loads that are entirely impossible in real world application.
If a stability test is only good for gaming application then it is not good in my book. Then again there can be wild differences from game to game too. So I wouldn't trust any game as a test. The only way to test for all around stability is synthetic stress tests.
That still doesn't make sense though. There's no such thing as "loads that are entirely impossible in real world application." Computers are general purpose machines. Sure, there are applications that are much less common, (calculating primes as fast as possible comes to mind), but they are still real world applications for someone. Most of the tests I know of are by-products of someone's effort to do real work. And claiming a system is 100% stable is essentially attempting to prove a negative. You can't say "this system will never crash." You can only say "this system has not crashed yet."

I know my argument is very pedantic, but I reiterate it to preface this point: it makes no sense to use stress tests that aren't good for your use case, and the total number of real use cases is essentially the sum total of PC users, plus one: Kyle Bennett counts twice.
 
Last edited:
I really should just use MH World as my gpu stability test, it crashes under such a weirdly light stress load.
 
The all of twice I've had to, I run multiple tests. P95 is a good start, but I mainly use OCCT as it has a more broad-spectrum tests, with a lot of options (heat generating, error-generating, Linpack, memory hammer, etc).
 
just tried Prime 95 (Blend) ... It's my new go-to along with MSI Kombuster [furry donut] for stress testing the GPU (can run both programs at the same time)
 
I don't think any 1 or 2 programs will cover all use cases for what would be considered a "stable" overclock. I think it requires a combination of stress test, games, productivity applications, and even general usage.
 
Back
Top