20 runs Intel Burn Test/LinX v.s. 24 hours Prime95

[andy5174]

Please also state the reason of your choice.

 

[OvrClkr]

I voted p95 but I no longer think that you actually need to do a 24hr test. As long as the CPU can do at least 8Hrs of large FFT's it should be good to go...

 

[Rustyy117]

I voted Prime95, I tested an overclock on a E8400 with IBT and it passed 10 runs, then I ran 3dmark06 which crashed within a minute.

 

[andy5174]

According to your opinions, it seems that the 20 runs LinX test just done by me is meaningless and unnecessary.

 

[OvrClkr]

I think both apps are equally reliable (give or take). Im just used to p95 and do not see the need for any other program...

 

[bob5568]

Andy, I had been using 50 minutes of LinX as a preliminary indication. Once I thought I was there, I upped the time for LinX to overnight (500minutes). My first run produced an error after 95 minutes. My second test succeeded to run without error for 500minutes.

Clearly the second is more stable than the first. How much more stable is undefined.

What we're missing is the study that provides a probability distribution for future stability based on length of error free testing per test tool.

Until we see that study, we'll always be at some undefined risk that an error will occur in the very next hour after we stop the test, no matter how long we test, and no matter what tool we use. Thus, what tool we use and how long becomes a gut reaction choice, with no rigorous logic to support it.

Bob

 

[OvrClkr]

Andy, I had been using 50 minutes of LinX as a preliminary indication. Once I thought I was there, I upped the time for LinX to overnight (500minutes). My first run produced an error after 95 minutes. My second test succeeded to run without error for 500minutes.

Clearly the second is more stable than the first. How much more stable is undefined.

What we're missing is the study that provides a probability distribution for future stability based on length of error free testing per test tool.

Until we see that study, we'll always be at some undefined risk that an error will occur in the very next hour after we stop the test, no matter how long we test, and no matter what tool we use. Thus, what tool we use and how long becomes a gut reaction choice, with no rigorous logic to support it.

Bob

Bob, im sorry but I do not see the point or logic of your post, what was the exact issue that you were having and what clock did you have issues with?

When a test goes bad it is usually due to wrong values (bios), CPU overheat or too high of a CPU clock.. Not sure what went wrong with your test since you stated that the first run produced an error but the second one didnt.

Until we see that study, we'll always be at some undefined risk that an error will occur in the very next hour after we stop the test, no matter how long we test, and no matter what tool we use.

errors happen because of user error, not because of the program/programs that we use..

please correct me If I have misunderstood...

 

[Rustyy117]

I think he's saying that its doesnt really matter how long you run a stability program, even if you run it for 24Hours, and then stop the test, the CPU may have failed within the 25th hour.

@bob, you may be right...but its not all gut instinct for example, if you know you run your PC for a maximum of 6 hours a day then you can safely assume that 7 hours stability testing will be enough for you, as your PC will only ever be on for 6 hours.

 

[OvrClkr]

I think he's saying that its doesnt really matter how long you run a stability program, even if you run it for 24Hours, and then stop the test, the CPU may have failed within the 25th hour.

@bob, you may be right...but its not all gut instinct for example, if you know you run your PC for a maximum of 6 hours a day then you can safely assume that 7 hours stability testing will be enough for you, as your PC will only ever be on for 6 hours.

Correct but it would not matter if indeed it failed at the 25th hour because nobody is going to stress thier CPU for that length of time or shall I say nobody needs to stress thier CPU's more than 24hrs at a time.

Once a CPU can achieve 8-10 hours of large FFT's with 0 errors you are done. I dont see the point of 24hrs unless you are running a "blend" test with a new build. Of course I may be wrong here, im just pointing out what has worked for me 😉

 

[bob5568]

Correct but it would not matter if indeed it failed at the 25th hour because nobody is going to stress thier CPU for that length of time or shall I say nobody needs to stress thier CPU's more than 24hrs at a time.

Once a CPU can achieve 8-10 hours of large FFT's with 0 errors you are done. I dont see the point of 24hrs unless you are running a "blend" test with a new build. Of course I may be wrong here, im just pointing out what has worked for me 😉

OMG read me correctly. All I'm saying is that you choose your length of time for testing, and you do so without any definitive science behind it. You think 8-10 hrs of large FFTs is best, but others may think 12 hrs is best....no one can pin their "feelings" on any solid basis.

The only solid basis I've heard is from the folks that consider the length of time you use a computer should compare to the length of time you stress test. I suppose that's something, but I suspect that errors caused by overclocking are a bit more random than that. If you have an unstable overclock, does the error in prime95 show up at the same time point test after test? I don't know myself, but I'll bet it doesn't.

My points all along are simple: There is no basis to determine the "proper" test regimen before we call our situation "stable". We all just evaluate according to our own version of "common sense".

 

[bob5568]

Bob, im sorry but I do not see the point or logic of your post, what was the exact issue that you were having and what clock did you have issues with?

When a test goes bad it is usually due to wrong values (bios), CPU overheat or too high of a CPU clock.. Not sure what went wrong with your test since you stated that the first run produced an error but the second one didnt.

Until we see that study, we'll always be at some undefined risk that an error will occur in the very next hour after we stop the test, no matter how long we test, and no matter what tool we use.

errors happen because of user error, not because of the program/programs that we use..

please correct me If I have misunderstood...

As I mentioned in my last post, my writing was not about a problem I had, just examples to make my point that all testing periods chosen are a persons own common sense, and that any test interval can be found to be insufficient.

Having said that, I'm happy to describe my own special circumstance between failing LinX at 90 min, and passing at 500min. It may be interesting to someone. At 174mhz bclk x 21 I was able to pass 50 min of LinX, but failed 90minutes into a 500min test. Reducing bclk to 171x21 allowed me to pass a 500min test. Both tests used a vcore of 1.3125v [bios](LLC Disabled), vtt of 1.21v, vpll of 1.86v, vdimm of 1.6v. Previous testing suggests that vcore increases beyond 1.3125v must be ever higher to produce higher clock stabilities and I pay the price in power consumption and heat dissipation. Both 174x21 and 171x21 conditions presented here, produce core temps of between 65 and 70c. I don't wish to see higher temps than this.

Thus, I suspect that 1.3125v vcore (LLC Disabled) is near a point of inflection on a curve that relates required vcore vs error free performance at a given bclk. I suspect this is partly caused by the unfortunate limitation I accept for max Vtt of 1.21. If I could raise Vtt beyond 1.21 as those of you with x58 machines can, then I'd bet there would be a different (higher) levels at which vcore increases would remain somewhat linear with newly enabled performance.

 

[OvrClkr]

OMG read me correctly. All I'm saying is that you choose your length of time for testing, and you do so without any definitive science behind it. You think 8-10 hrs of large FFTs is best, but others may think 12 hrs is best....no one can pin their "feelings" on any solid basis.

The only solid basis I've heard is from the folks that consider the length of time you use a computer should compare to the length of time you stress test. I suppose that's something, but I suspect that errors caused by overclocking are a bit more random than that. If you have an unstable overclock, does the error in prime95 show up at the same time point test after test? I don't know myself, but I'll bet it doesn't.

My points all along are simple: There is no basis to determine the "proper" test regimen before we call our situation "stable". We all just evaluate according to our own version of "common sense".

Ok, now I am going to give you some p95 101 😉

The only solid basis I've heard is from the folks that consider the length of time you use a computer should compare to the length of time you stress test.

That right there^^ is a negative, when you torture your CPU you are actually trying to find out if your settings / (CPU) / clock are stable (has nothing to do with the length of time you use your PC). The longer the torture the more stress you put on the CPU (to a certain level). If you use your PC 24/7 you do not need to stress test your CPU for multiple days. WHY? Because the PC will not be running all its cores at 100% ALL of the time. Lets say that you run your PC only 2 hours a day, does that mean that if it passes a 3 hour stress test it is good to go? Of course NOT..

When you are done with a build and lets say you want to be 100% positive that it is indeed stable for 24/7 operation all you need to do is "blend" test it for at least 16-24hrs (no need to run the test past 24hrs. Then you give it 8-10 Hrs of Large FFT's, if it passes w/o any errors you are done, no need for anymore stress testing until it is time to up the clocks again 😉

 

[OvrClkr]

As I mentioned in my last post, my writing was not about a problem I had, just examples to make my point that all testing periods chosen are a persons own common sense, and that any test interval can be found to be insufficient.

Having said that, I'm happy to describe my own special circumstance between failing LinX at 90 min, and passing at 500min. It may be interesting to someone. At 174mhz bclk x 21 I was able to pass 50 min of LinX, but failed 90minutes into a 500min test. Reducing bclk to 171x21 allowed me to pass a 500min test. Both tests used a vcore of 1.3125v [bios](LLC Disabled), vtt of 1.21v, vpll of 1.86v, vdimm of 1.6v. Previous testing suggests that vcore increases beyond 1.3125v must be ever higher to produce higher clock stabilities and I pay the price in power consumption and heat dissipation. Both 174x21 and 171x21 conditions presented here, produce core temps of between 65 and 70c. I don't wish to see higher temps than this.

Thus, I suspect that 1.3125v vcore (LLC Disabled) is near a point of inflection on a curve that relates required vcore vs error free performance at a given bclk. I suspect this is partly caused by the unfortunate limitation I accept for max Vtt of 1.21. If I could raise Vtt beyond 1.21 as those of you with x58 machines can, then I'd bet there would be a different (higher) levels at which vcore increases would remain somewhat linear with newly enabled performance.

+1

everyone's outcome is different 😉

and just for the record, I am not in dissagrement, just wanted to point out a few differences

 

[Rustyy117]

what I do to really test my overclock is to run OCCT for 8 hours and also run 3dmark06 at the same time.... I figured if its stable it should be able to run OCCT and 3dm06

ovrclkr you have a good technique for stress testing, using blend and large is a good idea. I've found some overclocks to be stable under small FFT's and fail under blend as it uses more memory.

 

[OvrClkr]

OMG read me correctly. All I'm saying is that you choose your length of time for testing, and you do so without any definitive science behind it. You think 8-10 hrs of large FFTs is best, but others may think 12 hrs is best....no one can pin their "feelings" on any solid basis.

you are absolutely correct, I was just giving you an example of what has worked for me for many years 😉

im sorry if you misunderstood me, im not a US native (its hard to get my point across sometimes)...

 

[overshocked]

Hey andy,
Not sure exactly what you mean by "reliable". Are you asking wich i like better for stress testing?

 

[bob5568]

OMG read me correctly. All I'm saying is that you choose your length of time for testing, and you do so without any definitive science behind it. You think 8-10 hrs of large FFTs is best, but others may think 12 hrs is best....no one can pin their "feelings" on any solid basis.

you are absolutely correct, I was just giving you an example of what has worked for me for many years 😉

im sorry if you misunderstood me, im not a US native (its hard to get my point across sometimes)...

I enjoyed reading your contributions. Communication can be difficult!

 

[bob5568]

Hey andy,
Not sure exactly what you mean by "reliable". Are you asking wich i like better for stress testing?

+1 precisely...no one can define "reliable".

 

[OvrClkr]

I enjoyed reading your contributions. Communication can be difficult!

hehe, yes it can

 

[Deders]

After lowering the voltage on my ram to the advertised 1.5v, I ran the Intel Tests and had no problems with it. Prime 95 on the other hand clocked out within seconds.