4790k - getting 30-40c less in Aida64 than P95

[b33k3rz]

In Prime95 by the second wave of tests I'll hit 100c and it'll star throttling, while in Aida64 the max load is 67c.

What gives?

I'm on stock settings except I enabled XMP for my ram to go to 2400mhz. Mobo is Asus z97-a and cooler is Hyper 212 evo.

 

[richierich1212]

Which version of prime95 are you using?

 

[(sic)Klown12]

Use the FPU test in AIDA to push the CPU and it should be much closer to what Prime95 shows. The normal CPU test isn't all that intensive, as you've found out.

 

[BSim500]

In Prime95 by the second wave of tests I'll hit 100c and it'll star throttling, while in Aida64 the max load is 67c.

What gives?

What gives is the "feature" Haswell has of arbitrarily adding +0.1v under AVX loads (like Prime). In no way, shape or form does it reflect normal everyday constant loads (eg, X264 encoding). If you ran the FPU test in AIDA, you see the same unnecessarily high temps, and if you used an older non-AVX version of Prime (pre version 27, eg, v26.6), you should see much lower Prime temps.

 

[ViRGE]

What gives is the "feature" Haswell has of arbitrarily adding +0.1v under AVX loads (like Prime). In no way, shape or form does it reflect normal everyday constant loads (eg, X264 encoding).

I thought X264 was AVX optimized a couple years back? So wouldn't it be an AVX load that triggers this exact situation?

 

[cytg111]

Welcome to Haswell .. should really be Hashell, cause the chip runs insanly hot.
If enough ppl were to come together I bet you could have a class action lawsuit potential vs Intel for this foobar.

 

[Techhog]

Welcome to Haswell .. should really be Hashell, cause the chip runs insanly hot.
If enough ppl were to come together I bet you could have a class action lawsuit potential vs Intel for this foobar.

Over what, exactly?

 

[GLeeM]

I thought X264 was AVX optimized a couple years back? So wouldn't it be an AVX load that triggers this exact situation?

I don't know about X264.

I do know AVX and AVX2/FMA3 are different. Only Haswell has AVX2/FMA3 so far.
AVX does not get as hot as AVX2/FMA3.
As you could guess from the heat produced, these instruction sets give tremendous speed boost - twice as fast!!
In BOINC, project Asteroids@home uses AVX and PrimeGrid uses AVX2/FMA3. So I have three different OC settings depending on what I am running.

Years ago we would rejoice in the speed boost from new instruction sets, today people are more concerned with ... well I'll be kind and not say it ()

If I am not AVX2/FMA3 stable I would need an OC that was twice as fast to keep up.

So then, I could brag that I am AVX2/FMA3 stable at 4.5 GHz or I could brag that I don't need AVX2/FMA3 because my overclock is 9.0 GHz

 

[james1701]

Did all this not start with the Ivy? Asus put out recommendations not to stress with P95 and to use Aida when they launched the 77 series boards.

 

[(sic)Klown12]

Did all this not start with the Ivy? Asus put out recommendations not to stress with P95 and to use Aida when they launched the 77 series boards.

I believe it's only with Haswell that Intel started doing the automatic 100mV jump for AVX/2 loads. My IB-E or my IB based laptop don't have the automatic jump so I'm assuming the desktop IB also doesn't overvolt.

 

[Kenmitch]

I believe it's only with Haswell that Intel started doing the automatic 100mV jump for AVX/2 loads. My IB-E or my IB based laptop don't have the automatic jump so I'm assuming the desktop IB also doesn't overvolt.

What gives is the "feature" Haswell has of arbitrarily adding +0.1v under AVX loads (like Prime). In no way, shape or form does it reflect normal everyday constant loads (eg, X264 encoding). If you ran the FPU test in AIDA, you see the same unnecessarily high temps, and if you used an older non-AVX version of Prime (pre version 27, eg, v26.6), you should see much lower Prime temps.

My 4690k running on a GA-Z97MX-Gaming 5 does not get the .1 avx vcore boost. Is this not normal behavior for the combo? I figured it was a z97 thing. I'm running with vcore = normal using a offset for overclocking. I couldn't find any info on the subject when I first noticed it. No boost with Intel Burn Test, no boost with Prime95 ver 28.5 so far. Guess I could try Linpack11 also. In the past IBT triggered the boost tho. Fresh install of Win8.1 vs the regurgitated install from the combo switch both have the same results as far as the .1v boost goes.

Edit: Running the latest Linpack11 also gets no vcore boosting....Only a little more vcore drooping.

The droop was too much and caused a bsod. I'm only overclocked to 4.2Ghz which requires around 1.12v's to be stable. I had to bump up my offset by .009 to run Linpack11 as it was drooping down to the 1.11's....Was fluctuating between 1.125-1.13v's. Sure does crank up the heat during it's what seems to be never ending load....Hottest core hit 81c using a H100i with the fans locked at 1500rpms.

 

[james1701]

http://youtu.be/4mkGQhE1o2w?t=3m57s

There is some answers to Aida.

 

[hapkiman]

I wouldnt run Prime95 with Devils Canyon or any Haswell. The Aida64 stability test is much better for it (or even OCCT), and wont set your room on fire. Haswell just cant handle Prime without getting close to 100c.

 

[crashtech]

I know Prime95 was considered the ultimate test of stability, but I have found Linpack (Intel Burn Test, OCCT, LinX) to be a great predictor of stability even without utilizing AVX.