Haswell Stress testing.. so much mixed info... help!

C

Copyright

Supreme [H]ardness
Joined
Nov 17, 2000
Messages
6,004
Been working on my OC for a little while now and have made some mistakes along the way but the one thing I am having a hard time understanding is how to properly stress test this chip and which method is safe.

Prime 95 was always my go to stress test program. Then I read about Intel burn in and found my so called "stable tune" wasn't stable at all. Then again do you have to pass every stability test to be considered stable? Also what version of Prime 95 is correct? I see mixed info there.. some say use the new one with AVX.. some say do not it's not needed. Should Prime 95 really be used at all? I heard Intel XTU uses prime for stress testing?

Intel XTU.. now what I have read about this one is that it's safe to use when voltage is in adaptive mode. I was failing this test after 3 hours.. small bump in voltage and now I am passing 8 hours no problem. I hear passing 8 hours of XTU is good enough to call stable and then I read it's not good at all.. so which is it??? My system peaked at 80c with this test during the 8 hour run.

AIDA64... The one thing AIDA64 did I didn't notice with anything else is it pushed my voltage VID higher than any other test.. no idea why.. probably becuase I had adaptive voltage on and I should not have.. will start all over again with manual and work from there. How long do you run AIDA64 before saying OK its good?

LinX - This test kills my temps.. I hit during my 20 passes 100C and throttled yet still passed with no errors. Now my voltage is 1.316V @ 4.8ghz.. so because of these temps do I abandon my OC? Gaming my peak temps hit 68C.. normal use it never gets very high. What would I do that could hit 100C temps and throttle? Handbrake? I know Audio conversion can hit some high temps but most of the time those are short runs.


Max Voltage VID.. 1.316V @ 4.8ghz on air to much? Other than LinX and AIDA64 my temps are pretty low.. AIDA64 only seems to push my temps high because it calls in extra voltage. Here again I was on adaptive and should have been manual voltage.

OCCT I have not used and thought about it.. at what point do these tests become redundant is what I ask my self. Plus I feel like I am just beating my CPU to pieces with all this work.

4.8ghz so far has given me zero stability issues.. even when pushed into thermal throttling the system remained error free and stable. So my question is do I let it eat like it is? Do I start over and find my manual voltage stability and then try to get Adaptive to work correctly? Is it just to much heat and I should try a 4.6ghz with some low voltage for a more silent PC setup.. I am losing sleep over this... lol.. overclocking in my head.

I am 1.290V with offset .025V ... I heard to find manual voltage stability.. put that in and then use an offset of .0001V so it will throttle down as needed. That correct?

Thanks for any help.. I know I wrote a book. I have yet to find a write up that says use this test or that test.. or multiple stability tests... and for how long you should be able to pass each one and move on. Test or don't test AVX? I keep seeing mixed opinions. The fact that LinX pushes me into 100C is that truly an issue when no other test I use does?
 
Asus recommends against using AVX I'd think they are in a position to see what happens. Found that advice somewhere in the Rampage V forums.
 
Prime95 IMHO with haswell it work just as a cooling test, its unreliable completely, you can pass 24 hours of prime blend, 24 hours of small FFTs and 24 hours of large FFTs and then other random day fail withing the first 10minutes of small FFT.. but also shouldn't be used with adaptive voltages with anything newer than 27.9 because it will skyrocket the voltage causing overheat and throttling same as the newer Linpack and LinX..

With Aida64 all depend in how you are using it... aida can be used with FPU test, CPU test, Cache test, depend on the test used Aida can push more or less voltage.. the fastest way to detect a instability would be to use Only FPU stress Test, a more real world but it also take more time, would be CPU+FPU stress test.. the Cache + CPU + FPU isn't really useful at all as it can become a bottleneck so it will be kinda unrealistic lower load..

Actually in my opinion as I mentioned in other thread the best stability test actually I think its Asus RealBench... it will present any kind of instability immediately in the Heavy MultiTasking test and Video Encoding..

and yes the best way to find a stable overclock its with manual voltage and Regular or 0% LLC. then once you find your stable speed + Voltage proceed to make some Automatic Voltage test to see how much voltage its using the processor and then start to apply positive or negative offset voltage to reach the desired voltage..
 
bob616 said:
Asus recommends against using AVX I'd think they are in a position to see what happens. Found that advice somewhere in the Rampage V forums.
Click to expand...

Good to know.. that source seems good enough to me! lol. So many mixed opinions on Prime I am thinking of just going with my Intel XTU results and letting it eat.. My other concern is I am 1.290V and .025V offset... that offset voltage might be a little high.. I know we are told to OC and stress with manual voltage but XTU is supposed to be safe with Adaptive and might help me lock in my final adaptive setting. I just don't understand how much offset I really need see as one site recommends starting with .0001V offset to get the voltage to fluctuate.
 
Araxie said:
Prime95 IMHO with haswell it work just as a cooling test, its unreliable completely, you can pass 24 hours of prime blend, 24 hours of small FFTs and 24 hours of large FFTs and then other random day fail withing the first 10minutes of small FFT.. but also shouldn't be used with adaptive voltages with anything newer than 27.9 because it will skyrocket the voltage causing overheat and throttling same as the newer Linpack and LinX..

With Aida64 all depend in how you are using it... aida can be used with FPU test, CPU test, Cache test, depend on the test used Aida can push more or less voltage.. the fastest way to detect a instability would be to use Only FPU stress Test, a more real world but it also take more time, would be CPU+FPU stress test.. the Cache + CPU + FPU isn't really useful at all as it can become a bottleneck so it will be kinda unrealistic lower load..

Actually in my opinion as I mentioned in other thread the best stability test actually I think its Asus RealBench... it will present any kind of instability immediately in the Heavy MultiTasking test and Video Encoding..

and yes the best way to find a stable overclock its with manual voltage and Regular or 0% LLC. then once you find your stable speed + Voltage proceed to make some Automatic Voltage test to see how much voltage its using the processor and then start to apply positive or negative offset voltage to reach the desired voltage..
Click to expand...

What about the amount of time these test's should be run to say good enough? I know there is no garuntee but how long should we be slamming our CPU's with 100% load to verify stability? Is LinX really worth using? I heard if you can pass 20 passes that is all you need.. then I hear some are running it for an hour!
 
yes, Asus does not recommend to use Prime95 with haswell, and specially with haswell-E Due to excessive amount of wattage used and can literally fry the CPU.. they even have a guide about that..

0.025V offset its actually very very Low that mean you have the LLC in automatic mode or very high-ultra settings (+50% -75%).. (just guessing)....

I've never found LinX worth, Always fan of IBT in Maximum mode 20 Passes.. IXTU 30Passes are also good to go... but those as synthetic are never reliable as real world usage, because with adaptive the maximum load voltage isn't applied and truly instabilities are present when the voltage are increasing and decreasing across the load in the cores, and the best exponent of this could be simply Heavy Gaming.. Crysis 3 can detect a unstable overclock amazingly fast... so if you pass 1 or 2 hours of continue Crysis 3 without crash or BSOD, I can assure you are stable because the voltage are constantly up and down according the Core Load which its also varying across the entire game.. if you want to stick to benchmark you can try FryBench, or Real world usage with Handbrake.. those are very good to detect unstable overclock..

about a Prime95 Time, well.. as mentioned above its unreliable some purist like to use it more than 8 hours in small FFT.. but I think 1 hour of small FFT are good enough...
 
Araxie said:
yes, Asus does not recommend to use Prime95 with haswell, and specially with haswell-E Due to excessive amount of wattage used and can literally fry the CPU.. they even have a guide about that..

0.025V offset its actually very very Low that mean you have the LLC in automatic mode or very high-ultra settings (+50% -75%).. (just guessing)....

I've never found LinX worth, Always fan of IBT in Maximum mode 20 Passes.. IXTU 30Passes are also good to go... but those as synthetic are never reliable as real world usage, because with adaptive the maximum load voltage isn't applied and truly instabilities are present when the voltage are increasing and decreasing across the load in the cores, and the best exponent of this could be simply Heavy Gaming.. Crysis 3 can detect a unstable overclock amazingly fast... so if you pass 1 or 2 hours of continue Crysis 3 without crash or BSOD, I can assure you are stable because the voltage are constantly up and down according the Core Load which its also varying across the entire game.. if you want to stick to benchmark you can try FryBench, or Real world usage with Handbrake.. those are very good to detect unstable overclock..

about a Prime95 Time, well.. as mentioned above its unreliable some purist like to use it more than 8 hours in small FFT.. but I think 1 hour of small FFT are good enough...
Click to expand...

Ok, So I have decided 4.8ghz is just to much voltage and to much noise etc.. I dropped to 4.6ghz.. @ 1.2V I failed LinX... at 1.21V in bios 1.2123 VID.. I passed 20 runs of LinX. Peak temps 88C.. MUCH better than the 100C adn throttle I got before. Going to run Intel XTU tonight.. It doesn't do passes but a timed run.. its very different than Intel burn in which I will give it a shot now as well. Thx for the help.. please stay tuned! Yes LLC was left as Auto.. I basically set XMP profile.. Set Manual Voltage to 1.21V and Multi at sync all cores and 46. Getting the manual voltage stability figured out and then I will move on to Adaptive. .025V for adaptive offset is higher than I see most run.. Now if it was JUST OFFSET and not Adaptive.. then yeah that would not be much at all.
 
So far.. 4.6ghz @ 1.21V manual V

Passed...

20 passes LinX Peak temp 88C
10 Passes IBT on High Peak Temp 83C
5 minutes of Intel XTU Peak Temps 64C

I will probably do an 8 hour run of XTU overnight.

Installing Asus Real Bench now.
 
good temps.. try 5 passes of realbench, that should be more than enough...
 
Araxie said:
good temps.. try 5 passes of realbench, that should be more than enough...
Click to expand...

It gave me a timer.. I did 15 minutes.. 67C peak temps.. Is that enough or run it longer?

Also, I will probably go in and set Adaptive to 1.210 and offset to .0001V and use that with Intel XTU.. It's supposed to be safe wiht adaptive voltage and can give me an idea if my adaptive setting is still stable.


EDIT:... so 1.21V and .001 offset is giving me VID of 1.2531 with intel XTU.. and 1.2874 in AIDA64.. I don't need that much voltage.. why is it spiking so high?
 
Last edited:
VID its the voltage the chip its asking to feel comfortable with the selected speed.. VID doesn't represent a real voltage requirement, for example in my case my VID running Aida64 FPU only its 1.20v but the real voltage needed its 1.256V, so don't pay too much attention to VID, the Voltage spikes are due LLC, because you have it Auto it will vary according load then adjust in real time so voltage can make some spikes.. if you select the LLC to 0% (or regular) the spikes should be little to nonexistent...

About realbench it have stress test and Benchmark.. select benchmark and select 5 runs...
 
OK, going to go play with LLC to bring those peaks down some and ill run teh bench 5 times.. after that I will do an overnight run of Intel XTU and call it DONE! Thx for the help!
 
My LLC is done in numbers.. and from what I am reading those spikes are because of Adaptive mode and not LLC? If it was LLC I would have gotten those in manual mode as well correct?

EDIT: Verified LLC does not effect this.. apparently its a normal function of adaptive mode and the CPU but what a stupid function.. it pushes extra heat voltage i do not NEED.. I passed everything just fine at 1.21V and now I am getting higher than that and about a 4C raise in temps for no reason.. Almost better off running manual
 
Last edited:
in my test of stress program linpack with avx and prime95 blend mode came out on top.
occt's own cpu test and prime95 small mode where good runners up. (Prime96 large where not tested)

AIDA64 and all the older favorites like superpi, hypepi, wprime,fft-z, hot cpu, core dmg, burnmmx, burnp6, burnk7, distrubutednet, etc etc etc
Where all incredible bad at showing instability in comparison. on all systems.

my test where done over both amd and intel cpu both avx and pre avx systems.

i dont know why asu would recommende against avx testing it seems to show instabilities alot faster but i guess is the same reason corsair once recommende against using memtest86+. they just dont want their product to be tested into the extremes.

the system where both stock and overclocked. ab instability where insured by lowering the voltage slowly in multiple steps and see which programs would have the shortest time to detect the error.
 
I'm in the same boat with my AMD 8350. I'm using Prime95 because it has what I've always used but I do see some conflicting information. I also just ran it for an hour and it failed then I immediately re-ran it and it's been going fine for 3 hours.
 
I found stability 1.21V for 4.6ghz.. 1.275V for 4.7ghz and 1.31V for 4.8ghz.. Air cooling is basically my limitation and for some reason my chip in Adaptive mode responds kinda odd. I chose to run 4.6ghz because its so much cooler ~13C cooler and my fans rarely throttle up even under full load which makes it nice.
 
You must log in or register to reply here.
Back
Top