What settings are you using for IBT?

What settings are you using for Custom Prime95? When you select Custom, are you making sure that Blend is selected first?

To get right to the answer while I wait for your answers to my questions here, Prime95's Custom Blend test for about 24 hours using about 80-85% of your memory is ideal.

 

Quote:

Originally Posted by TwoCables

What settings are you using for IBT?

What settings are you using for Custom Prime95? When you select Custom, are you making sure that Blend is selected first?

To get right to the answer while I wait for your answers to my questions here, Prime95's Custom Blend test for about 24 hours using about 80-85% of your memory is ideal.

For IBT I'm running Maximum amount of RAM.

For Prime I select Blend first, custom, then set 5500MB and 5 Minutes per FFT.

 

Quote:

Originally Posted by Speedster159

Quote:

Originally Posted by TwoCables

What settings are you using for IBT?

What settings are you using for Custom Prime95? When you select Custom, are you making sure that Blend is selected first?

To get right to the answer while I wait for your answers to my questions here, Prime95's Custom Blend test for about 24 hours using about 80-85% of your memory is ideal.

For IBT I'm running Maximum amount of RAM.

For Prime I select Blend first, custom, then set 5500MB and 5 Minutes per FFT.

The only thing that I can think of is your copy of IBT doesn't use the AVX instruction set and your copy of Prime95 probably does (which would mean that you should just stick with using Prime95). Which version of Prime95 do you have?

 

Quote:

Originally Posted by TwoCables

The only thing that I can think of is your copy of IBT doesn't use the AVX instruction set and your copy of Prime95 probably does (which would mean that you should just stick with using Prime95). Which version of Prime95 do you have?

I have 28.5 Build 2 and I always make sure it reports that it has detected the Core architecture and is optimizing itself for it.
\

 

Latest prime95 blend with 90% ram ran for 24 hours if you want to pretty much never worry about stability problems again. This has always been the most reliable test of stability for me even back to the days of conroe.

I have never seen a situation where I could pass prime95 but not pass any other test or regular program. However, many times I was able to pass alternatives such as OCCT/IBT/Aida and have those fail when testing with prime95. So if you're limiting yourself to a single test then I think prime is your best bet. I would also suggest using a secondary test just to make sure. If you use IBT, make sure you're using the updated linpack libraries because the ones included with IBT are very outdated.

 

Quote:

Originally Posted by Speedster159

Quote:

Originally Posted by TwoCables

The only thing that I can think of is your copy of IBT doesn't use the AVX instruction set and your copy of Prime95 probably does (which would mean that you should just stick with using Prime95). Which version of Prime95 do you have?

I have 28.5 Build 2 and I always make sure it reports that it has detected the Core architecture and is optimizing itself for it.
\

Then I would just stick to Prime95, especially considering what Derp just said. Thank you, Derp!

 

One more thing that I would suggest after passing your stability tests is to go back and bump the Vcore a few times just in case you're on the edge of stability. Notice how Intel uses wayyyy more Vcore than what's needed at stock? This is the same idea just to a lesser degree.

 

Prime95 + 90% available ram + 12 hours success = stability in my past experience.

 

This might sound silly but it's sound logic: Why not test it with what you are doing? Encoding.

 

Quote:

Originally Posted by $ilent

Prime95 + 90% available ram + 12 hours success = stability in my past experience.

That's the method I use. 12-16 hrs with no WHEA loggers in event viewer is stable for me.

 

Quote:

Originally Posted by fateswarm

This might sound silly but it's sound logic: Why not test it with what you are doing? Encoding.

That would be fine if he weren't overclocking, but lots and lots and lots of things change when you overclock. It's like you are manufacturing a new CPU: you have to put it through its paces to the best of your ability; try to bring it to its knees, so to speak. If it can handle that at the new settings (the overclock), then you should be fine. If you don't properly test the stability, then you will never know if it's truly stable and you could run into data corruption.

Stress testing is a part of overclocking; it's not an option.

 

Why? At the Haswell thread their favorite test is x264. For all uses.

 

Alright going to do the Prime custom blend and see how that works. (+REP) Would you suggest test stability with fixed voltage then try to replicate the Voltage on Offest when stability is established?

My CPU has some attitude to it huh? I just did a straight 3 hour Prime Custom Blend and it passed it, I leave it for an hour unattended and it crashed to restart giving a 124. I retry the test without changing settings and I can't get past an hour without a worker stopping, crashing, or getting a BSOD with 124.

I'm now at 1.315 VCore and 1.085 on VTT. No BSOD's from 1.305 and above with the same VTT but I get a Worker that stops within an hour.

Also is there an information on "VRM Current Threshold"?

 

Quote:

Originally Posted by fateswarm

Why? At the Haswell thread their favorite test is x264. For all uses.

Haswell is very different from Sandy and Ivy Bridge.

 

Quote:

Originally Posted by Speedster159

Alright going to do the Prime custom blend and see how that works. (+REP) Would you suggest test stability with fixed voltage then try to replicate the Voltage on Offest when stability is established?

My CPU has some attitude to it huh? I just did a straight 3 hour Prime Custom Blend and it passed it, I leave it for an hour unattended and it crashed to restart giving a 124. I retry the test without changing settings and I can't get past an hour without a worker stopping, crashing, or getting a BSOD with 124.

I'm now at 1.315 VCore and 1.085 on VTT. No BSOD's from 1.305 and above with the same VTT but I get a Worker that stops within an hour.

Also is there an information on "VRM Current Threshold"?

Wait a sec: are you using an Offset voltage?

 

I still don't get why an encoding stress test is worse for ..encoding he uses.

 

Quote:

Originally Posted by fateswarm

I still don't get why an encoding stress test is worse for ..encoding he uses.

I already explained it. I'm sorry, but it's the only way that I personally know how to explain it. If you really want a better explanation, then ask in a new thread why you can't just test the stability of your overclocked CPU by just using your computer the way you normally use it.

If you don't want to do that, then you're stuck with whatever answers anyone might care to give you in this thread (that is, if anyone decides to try to give you an answer even though this isn't your thread).

 

Quote:

Originally Posted by TwoCables

I already explained it. I'm sorry, but it's the only way that I personally know how to explain it.

The problem is you have no arguments and you sound convinced. "Things change when you overclock" means nothing. An encoding stress test is also "bringing it to its knees" and it's also similar to the usage in need.

Anyway, there is no point discussing it further. The arguments have been laid out.

/Over and Out.

 

Prime 95 Custom Blend, 90% RAM, 48H

 

Quote:

Originally Posted by fateswarm

Quote:

Originally Posted by TwoCables

I already explained it. I'm sorry, but it's the only way that I personally know how to explain it.

The problem is you have no arguments and you sound convinced. "Things change when you overclock" means nothing. An encoding stress test is also "bringing it to its knees" and it's also similar to the usage in need.

Anyway, there is no point discussing it further. The arguments have been laid out.

/Over and Out.

If you really want a better explanation, then ask in a new thread why you can't just test the stability of your overclocked CPU by just using your computer the way you normally use it.

If you don't want to do that, then you're stuck with whatever answers anyone might care to give you in this thread (that is, if anyone decides to try to give you an answer even though this isn't your thread). I'm not going to sit here and try to find different ways to explain it when this is the best way that I know how to explain it.

Edit: I'll make the thread because I know that you won't.

Edit #2: Done. http://www.overclock.net/t/1507632/why-cant-i-just-test-the-stability-of-my-overclock-by-using-my-computer-normally Subscribe and wait for the answers.

 

Quote:

Originally Posted by TwoCables

If you really want a better explanation, then ask in a new thread why you can't just test the stability of your overclocked CPU by just using your computer the way you normally use it.

That's a strawman argument. I didn't say to normally encode. I said to use the x264 stress test (there's a special stress testing tool, used regularly over at the haswell thread).

 

Quote:

Originally Posted by TwoCables

That would be fine if he weren't overclocking

Wouldn't even be fine even if there was no overclocking.
Quote:

Originally Posted by fateswarm

"Things change when you overclock" means nothing.

Correct.
Quote:

Originally Posted by fateswarm

An encoding stress test is also "bringing it to its knees"

Not really.
Quote:

Originally Posted by fateswarm

it's also similar to the usage in need

Exactly why it's a terrible stress test.

If the settings he's using have a 20% chance per hour of failing while encoding, he could run a six hour encoding stress test and still have more than a 26% chance of never seeing a problem, then over the next few months, crash or produce unusable encodes several times.

Basically, in order to convincingly demonstrate that a system is stable for a task, by performing that task, you'd need to run that task as a test much longer than the amount of time you'd anticipate actually needing to do that task for anything else.

This is why stress testing is typically done with much more stressful programs than will be used day to day, because doing anything less would take prohibitively long.
Quote:

Originally Posted by fateswarm

I said to use the x264 stress test (there's a special stress testing tool, used regularly over at the haswell thread).

This test is just a CLI transcode that has a batch file allowing you to automatically set a number of loops. It's no more demanding than x264 normally is, and is probably less demanding than future versions of x264 that people will actually use for transcoding.

There is nothing abnormal about the encoding the x264 Stability Test is doing, other than looping the encode.

 

Quote:

Originally Posted by fateswarm

Quote:

Originally Posted by TwoCables

If you really want a better explanation, then ask in a new thread why you can't just test the stability of your overclocked CPU by just using your computer the way you normally use it.

That's a strawman argument. I didn't say to normally encode. I said to use the x264 stress test (there's a special stress testing tool, used regularly over at the haswell thread).

Again, Haswell is very different from Ivy Bridge. Look at his sig rig.

Now stop wasting my time!

 

Quote:

Originally Posted by Blameless

This is why stress testing is typically done with much more stressful programs

The x264 or x265 stress tests are not the same with regular encoding. They are stress testing. His video encoding is always a lighter task.

 

Quote:

Originally Posted by TwoCables

Wait a sec: are you using an Offset voltage?

I'm currently using a fixed voltage and monitor my voltages periodically from the back of the board from the capacitors surrounding the socket since the Software and V-Probe is just not accurate enough.

I'm at 1.315VCore, 1.085VTT and still doing a stability test right now for an hour and counting. Temperatures Max at 82C on the hottest core.