Is long stress testing necessary?

[Stochastic]

Many OCers run stress tests for several hours, sometimes upwards of half a day or more. Is it really necessary to put that much strain on your PC for so long? Don't you risk accelerating degradation?

I've heard others say that if you can run a test for around an hour without crashing or errors then you're stable. What's the truth?

 

[paperwastage]

as long as you are comfortable with your limits (thermal/temperature, voltage), then stress testing for several hours is fine

i'd say stress test an hour or so to find the upper limits of frequency given your voltage, then dial the frequency down a little (or voltage up a little), stress more (>3 hours) to make sure it is stable

 

[MPiland]

When running Prime95, which test is preferred to use? I was using the long IFFT or whatever it is (I'm at work so don't have it in front of me). But I saw other people saying do the short or use blend? Anyone know which one is best to test with?

 

[blckgrffn]

"It depends."

If it can take a heavy pounding for an hour or two and you won't be hitting it that hard normally, then I would think it would be good enough for you.

Now, if you are going to be re-encoding blurays, heading off to a LAN party for a weekend or running some distributed computing you'd probably want to test longer to make sure you aren't ultimately wasting time and opportunities later.

My $.02.

Edit: I have found this to be the #1 reason I only go for modest, stock voltage OC's (~10%), I just don't have the time and patience to fart around with this like I used to.

 

[dma0991]

It depends as you want to achieve the maximum possible stability for the given speed that you want. I would say it is necessary to run stability tests if you want to be sure that your system is rock solid. There is nothing more frustrating than doing some WU crunching and it fails after 50% of the work is done.

Most of my BSODs tend to happen when there is no load or minimal load, so stress testing isn't a 100% guarantee that your PC is stable. That being said, the only setting that is near 100% stable is stock.

 

[Martimus]

The longer you stress test, the less likely you will have random crashes down the road. I found that 12 hours of Prime 95 was enough to eliminate random crashes, but 8 hours was not on my 4200+ X2.

 

[Avalon]

Sure is. I've seen crashes as far down the road as 4-5 hours into stress testing. Typically I'll run small FFT or large FFT prime95, and usually start it before bed and end it after work the next day, which gives it around 16 hours.

If it can pass that, I know there's very little I can do on my PC to cause any problems.

 

[Binky]

No.

 

[Concillian]

i'd say stress test an hour or so to find the upper limits of frequency given your voltage, then dial the frequency down a little (or voltage up a little), stress more (>3 hours) to make sure it is stable

I agree with this.

If you are searching for the ABSOLUTE MAXIMUM your CPU will do 24/7, you need LONG periods of stress testing with very stressful programs. This is because not every program stresses in the same way. When I was less experienced in this game, I though Prime95 overnight was enough. But after successful P95 overnight runs, I'd still find some games or programs that would BSOD in a matter of minutes.

It's much easier to use Prime for an hour and LinX / IBT for an hour, find stable, then dial back clocks 5% while keeping the same voltage, which is what I do now. 5% at 4.5 GHz is ~200 MHz. Some probably do not go to this extreme... 100 MHz may be something you decide is okay, but I just got sick of chasing down the causes of BSODs / freezes when they'd occur, so now I leave a fair bit of margin.

 

[Concillian]

When running Prime95, which test is preferred to use? I was using the long IFFT or whatever it is (I'm at work so don't have it in front of me). But I saw other people saying do the short or use blend? Anyone know which one is best to test with?

Small FFT should be the most demanding on the CPU

Large FFT will put more stress on the communication busses and memory (for BCLK overclocks or memory overclocks).

They are stressing independent portions of an OC. If you are leaving your BCLK alone and not OCing your memory, then use Small FFT.

 

[MPiland]

Small FFT should be the most demanding on the CPU

Large FFT will put more stress on the communication busses and memory (for BCLK overclocks or memory overclocks).

They are stressing independent portions of an OC. If you are leaving your BCLK alone and not OCing your memory, then use Small FFT.

This is what I was wondering! Thank you! Now, what about blend? I see a people saying to run a blend? Does that do a little of everything?

 

[exar333]

I agree with this.

If you are searching for the ABSOLUTE MAXIMUM your CPU will do 24/7, you need LONG periods of stress testing with very stressful programs. This is because not every program stresses in the same way. When I was less experienced in this game, I though Prime95 overnight was enough. But after successful P95 overnight runs, I'd still find some games or programs that would BSOD in a matter of minutes.

It's much easier to use Prime for an hour and LinX / IBT for an hour, find stable, then dial back clocks 5% while keeping the same voltage, which is what I do now. 5% at 4.5 GHz is ~200 MHz. Some probably do not go to this extreme... 100 MHz may be something you decide is okay, but I just got sick of chasing down the causes of BSODs / freezes when they'd occur, so now I leave a fair bit of margin.

Good explanation.

If you plan to run F@H on your rig 24/7, you better be sure that your OC works at average (or higher) ambient temp for 48hours+ with ZERO issues whatsoever.

I personally like to run Prime (1 instance per thread, based on your CPU) for ~48 hours at a high ambient temp, like ~30C) to feel comfortable. That is AFTER I have thoroughly mem-tested the RAM specifically to ensure I am not wasting my time.

I can certainly see some folks using a few hours of prime/linX/Intel-burn because they will likely only using their CPU for gaming and other not '100%' loading uses, but I still like to err on the side of caution.

Anyone who does less than an hour of stress-testing is risking issues down the road and if they say that's all that is needed, its bad advice IMHO. Certainly if you strenuously test your CPU successfully at 4.5ghz and decide to dial-it back a little to 4.3, you probably don't need to validate as much.

Keep in mind some of us are more 'old school' with overclocking and have had experiences with VERY finicky boards over time that might do 'X' or 'X-200mhz' but could be unstable at 'X-100mhz' because of memory holes or other issues.

 

[Martimus]

This is what I was wondering! Thank you! Now, what about blend? I see a people saying to run a blend? Does that do a little of everything?

Yes, blend will run both tests. Although not at the same time, it will run one then the other, and repeat. I like this test as heat fluxuates during this test so you end up with different temperatures during the tests than you would normally while running one or the other.

 

[aaksheytalwar]

IMO just use a proper test like ibt or occt for 5 mins. Makes temps higher than a few hours of prime and is a better test. I barely test for 5 mins and don't get crashes, if ever you get one just up the voltage 2 notches and you re set

 

[MPiland]

IMO just use a proper test like ibt or occt for 5 mins. Makes temps higher than a few hours of prime and is a better test. I barely test for 5 mins and don't get crashes, if ever you get one just up the voltage 2 notches and you re set

5 minutes is definitely not enough. Sure, if you're testing temps (even then that's too short if you ask me), but I'm looking for parity errors, not just heat. Hell yesterday I was stable at 4.6 for over an hour on prime and then it finally BSOD'ed on me so I dialed it back to 4.5. Figured I'd get that stable and be happy with it. I was having too much trouble getting 4.6 stable at a decent voltage and I was too tired and annoyed to keep at it. 4.5 is plenty for any real world application.

 

[jacktesterson]

I usually stress test using OCCT for 3-4 hours on small packet size. In my experience, OCCT will error a lot faster than Prime95 if there is stability issues.

I find if it can pass that, I'm usually stable. Actually, I've never run into issues O/C'ing this way.

 

[exar333]

IMO just use a proper test like ibt or occt for 5 mins. Makes temps higher than a few hours of prime and is a better test. I barely test for 5 mins and don't get crashes, if ever you get one just up the voltage 2 notches and you re set

5 min is definitely not enough. Prime is definitely a LOT better now with AVX support. It runs awfully hot now, pretty close to IBT or OCCT.

 

[talonz]

When you are testing your OC on a SB or later chip and you use offset voltage adjustment, make sure you test with various core loads.

VID changes depending on the wattage drawn by the processor, and it's important that it's stable under the load of 1, 2, 3 or 4 cores.

Just because it's stable at 4 doesn't mean it's stable with fewer cores due to a different core voltage, which is your typical usage case.

 

[aaksheytalwar]

Prime isn't enough for several hours. But occt 5 mins is more than enough if you are liberal with slightly higher volts

 

[aaksheytalwar]

Been overclocking since a64 and never faced issues after 5 mins occt

 

[MPiland]

From the research I've been doing, Prime95 and LinX test different instruction sets, so it's best to use both. OCCT is more for GPU. Yes, it has CPU and PSU stress tests, but from what I've read, they're not compareable to LinX or Prime95. On Prime, you should run small at first for a couple hours and then blend for 12 hours. LinX you should do at least 20 passes. Again, this is what I've gathered from reading various threads.

 

[exar333]

Been overclocking since a64 and never faced issues after 5 mins occt

I still would say that's pretty light on stress-testing. It also depends on how aggressive you are being. A 10-15% OC on many modern CPUs is very easy to achieve, and 5 min probably would be fine in many cases. If you are shooting for a max OC, then you probably want more stress-testing to validate the absolute maximum clock.

It never hurts to start the test before you go to be and see what the results were when you get up. It's worth even the 1% chance of corruption happening if your 5min went fine. Sometimes letting it go longer allows more heat to collect and really test your other components as well, such as MB cooling and other parts.

 

[aaksheytalwar]

I went 2.54 on my a64 which was the max possible with acceptable voltage. Even these days I do moderate ocs. Read 4.3 or 4.5 Ish, not 4.6+ tho

 

[tigersty1e]

the reason we stress test for so long on prime95 is not to stress it for a long time. we do it for so long because it takes a long time to run through all the tests.

prime95 blend running through all the tests takes about 10 hours on my rig. it's not about how demanding your rig is going to be. it could just be idling in windows and fail because your cpu might not be stable running through a particular set of instructions.

imagine testing someone's brain in multiplication in double digit maths. they can do everything except 93 x 1. prime runs through all the possible combos. and just idling or flashing your mobo can possibly run through a calc of 93 x 1 and now you get screwed.

 

[Kenmitch]

When you are testing your OC on a SB or later chip and you use offset voltage adjustment, make sure you test with various core loads.

VID changes depending on the wattage drawn by the processor, and it's important that it's stable under the load of 1, 2, 3 or 4 cores.

Just because it's stable at 4 doesn't mean it's stable with fewer cores due to a different core voltage, which is your typical usage case.

Doesn't seem to effect all chips but is true. My 2550k needs a little more to stabilize single core load than 4 core load when going over 4.7 ghz.