Overclock.net banner

1 - 20 of 73 Posts

·
Aussie Overclocker
Joined
·
2,791 Posts
Discussion Starter #1
Disclaimer: This thread is not an attempt to convince people that one is better than the other. These are my results and conclusions which I wanted to share so we could discuss and hopefully learn something

Test bench:
Intel 3770K @ 4.7GHz (each test required different settings for stability)
ASRock OC Formula
G-Skill Trident X F3-2400C10D-16GTX 16GB (2x8GB) 2400MHz
Corsair H100i
DMM is a DigiTech QM1571

Aida64
I ran Aida64 for 2 hours (my normal test time although Aida64 recommends 12 hours) with only Stress FPU selected (FPU stressing means AIDA64 System Stability Test will use a floating-point calculation task that stresses the FPU part of your processor. Modern processors all have an integrated FPU, and from all the components that are integrated, the FPU is the most complex one. Hence stressing only the FPU actually stresses most of your processor, and usually drives the processor to its maximum temperature)



Prime95
I ran Prime95 for 5 hours with the settings outlined in the Complete Overclocking Guide: Sandy Bridge & Ivy Bridge | *ASRock Edition* thread. People might complain that these are not "optimal" settings but they are what most people on this site who have this board would use.



Voltage (recorded with a DMM)
Aida64 requires slightly less voltage than Prime95.



Temperature
Aida64 causes higher temperatures



Conclusion
I have been overclocking CPUs for a number of years and I have never used Prime95 to test the stability of an overclock. I have never had a system crash that has passed a few hours stressed with Aida64 (and before that when it was Everest) I have, however, helped many people who have run Prime95 for many hours only to have their systems crash when they fire up a game (for example)
 

·
Registered
Joined
·
3,467 Posts
Fanboy flames incoming
 

·
Registered
Joined
·
602 Posts
I've been using Intel Burn In Test, and the voltages in CPUZ look low 1.18 @ 4.2 for a 3570k. However, it seems similar to Aida64 in results. IBT will find a flaw very quickly.

The other thing I noticed is that IBT, it will throttle the CPU up and down as it ramps each test. I think that this is key for catching instabilities under transient loads. EG, firing up a game. Using Openhardwaremonitor.org I can actually watch what load vs TDP vs temp looks like real time. 100% processor doesn't equal 100% load, you can watch the temp load up with each test as they queue up. The tools from the motherboard mfgs suck and don't seem to post real time data. I was using Overclockix and Mprime and that just simply couldn't catch anything. Once stable results in Mprime=prime95 in Linux wouldn't even boot into Windows.
 

·
Aussie Overclocker
Joined
·
2,791 Posts
Discussion Starter #6
Quote:
Originally Posted by SynchronicBoost View Post

I've been using Intel Burn In Test, and the voltages in CPUZ look low 1.18 @ 4.2 for a 3570k. However, it seems similar to Aida64 in results. IBT will find a flaw very quickly.

The other thing I noticed is that IBT, it will throttle the CPU up and down as it ramps each test. I think that this is key for catching instabilities under transient loads. EG, firing up a game. Using Openhardwaremonitor.org I can actually watch what load vs TDP vs temp looks like real time. 100% processor doesn't equal 100% load, you can watch the temp load up with each test as they queue up. The tools from the motherboard mfgs suck and don't seem to post real time data. I was using Overclockix and Mprime and that just simply couldn't catch anything. Once stable results in Mprime=prime95 in Linux wouldn't even boot into Windows.
Interesting. I haven't used IBT a whole lot myself but know people who prefer it
 

·
Registered
Joined
·
1,739 Posts
Interestingly my conclusion is opposite of yours. I've had more overclocks fail on me after passing AIDA64 than with Prime95. Both are not fail-safe
 

·
Registered
Joined
·
602 Posts
I was running stress testing after that last post and I can confirm that Prime 95 can't load up and put temp into the CPU like IBT can, nearly 16c differential. With IBT I've also found that you can pass with the smaller RAM tests at 1024 and fail by the time you get to the highest stress level.

A side benefit I've found with IBT, don't know about AIDA64, but IBT will catch over voltage. So if you are running too high CPU core V, it will BSOD 124.
 

·
Aussie Overclocker
Joined
·
2,791 Posts
Discussion Starter #9
I think the lesson to be learned is that there is not a single solution that will work for everyone. The best thing to do is try a variety of stress tests and see what works for your system
 

·
Registered
Joined
·
88 Posts
Well I had to try it to see if you're right OP!!!

Well here's what I got...

Volts were the same....1.160V 100% load but I see a 2-3C decrease in temp...using prime...

But the temps did not seem to jump about as much on prime as it dose on Aida....
rolleyes.gif


Edit: this was with a 4.2 OC
 

·
OC...D
Joined
·
28,109 Posts
Quote:
Originally Posted by tw33k View Post

I think the lesson to be learned is that there is not a single solution that will work for everyone. The best thing to do is try a variety of stress tests and see what works for your system
Nice work tw33k. I think this post captures it correctly. all of these are limited in the instruction set they call upon in the cpu and what cache load they produce so each brings a unique hit to the architecture. my preference has been to use at least 3 (IBT/linpak, p95, superPi 32M, and AID64), but not for more than a few hours on any. Maxing out the ram is critical. In addition to gaming as a stressor, I'd add one more: download a copy of Chimera 1.8rc and run it's benchmark - it will use 100% RAM and hang many systems.

With iGPU chips, none of these stressors actually light-up the entire die... but how many overclockers are using the iGPU anyway.
 

·
Premium Member
Joined
·
2,378 Posts
You definitely need to run more than just one stability test, and even if you run them all you still need to use your rig for a week testing all your program/games to see if the oc holds.

Also I would have to say your Prime95 version is way out of date, and most people do not use those settings. Default torture test settings so 8-4096, 15min, and use whatever number in memory used so at least 90% of your available ram is stressed as shown in task manager.

If you have time I might suggest redoing the Prime95 test with 27.9 and the settings above, and you can also add IBT at maximum, and Intel's XTU to the roundup.
 

·
Registered
Joined
·
180 Posts
Is the voltage measurement tool from a DMM attached to the voltage reading points and connected to your PC via a serial port?
Wish I had that one.

// Edit
You might also be interested in this sheet:


I didn't let the tests run for 2 hours though, only for about 10-20 minutes, but there's not much change beyond that anyway. Also these are Open Hardware Monitor readings, so not the real voltages measured with a DMM.
Vcore in Prime95 v26.6 with a 960k FFT size settings seems to produce the lowest Vcore, it seemed to stay longer on 1.336v compared to LinX or v27.9 with 60k FFT (really wish I had that DMM of yours).
 

·
Aussie Overclocker
Joined
·
2,791 Posts
Discussion Starter #15
The DMM connects to my PC by wireless USB and the software records the readings in real time
 

·
Premium Member
Joined
·
1,983 Posts
I've also used Aida64 for years, since it was Everest and I'm now finding out it has a stress test.
tongue.gif


I pretty much use it to get sensor readings on the screen of my Logitech keyboard (did before with the G15 and now with G19), and I used to run some benchmarks with it but haven't done in a long time.
 

·
Aussie Overclocker
Joined
·
2,791 Posts
Discussion Starter #18
It was cheap too the (DMM). $120 I think I paid.

@justanoldman...you have totally missed point of this thread. These tests were run on a test bench, the sole purpose of which is to run tests. It's not about testing the stability of the system. Running Prime95 again with different settings would do what? Lower the voltage required? I doubt it. Just because a specific test of stability works for someone does not mean that is the "right way" or the "only way" either. Would the test bench crash if I played a game? Watched a video? I don't know and never will because it's not used for those things...it is a test bench.
 

·
Premium Member
Joined
·
2,378 Posts
^Sorry, wasn’t trying to bring in a stability argument. I don’t know exactly how you ran it, but it could definitely change the temps. Since temps was a part of your argument and also stressing ram, I though the testing was incomplete. To compare Prime95 to something else is fine, but to use a very outdated version, run it with less than optimal settings, and to not set the ram level high enough renders the conclusion incomplete. Just my opinion.
 

·
Registered
Joined
·
180 Posts
Different FFT sizes will produce different load scenarios, which is why you see that rather large fluctuation in your Prime data.
If you choose only one FFT size (see my sheet for examples), you will receive much more stable results. I assume the AIDA test does basically the same, testing the FPU only with specific calculations and not wildly variate as the default setting for Prime does.

Also, in v27 support for the AVX instruction set was added to Prime, so you will see higher temps and more watt usage than in previous versions. Vcore doesn't seem have changed though compared to 26.6 (again, see my sheet). I don't know the differences to 25.11 though.

And LinX does basically all of it. The lowest Vcore, the highest temps and the highest watt usage. It will fluctuate as well though when finishing a round and starting a new one. But only then.
 
1 - 20 of 73 Posts
Top