i7-4770k high temp help

[Ascension6]

Hi guys,

I just finished my new build (i7-4770k @ stock speeds), with HYPER 212 EVO in a fractal R4 case. I added an extra noctua side intake fan to the R4 also.

I have been seeing really high temps during prime95 8-threaded blend tests, wondering if I could get some help.

I used some ancient AS5 thermal paste (maybe 10 years old?), and I was seeing idle temps around low 40s in realtemp. This seemed too high, so I went out and got some antec formula 7 paste. Now the idle temps are lower, around 29-34. I spread the paste rather than doing a dollop in the middle.

however, I still get crazy high temps during p95 @ stock speeds. If I run for 10 minutes or so, it is only around 80. After 15 minutes, I saw it go up to around 100, throttling cut in and I killed the program. I had the case fans set to low at this point, so I reran p95 with them set to high. Again, after around 15 minutes, I saw temps going into 90s so I killed the test. I could deal with 80s--this computer is probably fast enough for me without needing to OC it, but 90s? Too high...

Any advice or suggestions here?

Thanks.

 

[Zoeff]

These temps are normal for haswell, I'm afraid.

 

[Ascension6]

Heh, well it's not just me. Maybe I can pick up another fan for the CPU cooler to help some time...

 

[waldoh]

Chances are even during a long gaming session your CPU wont hit prime95 temps, so you could do a minor OC and be stable.


Sent from my iPhone using Tapatalk 2

 

[2timer]

Doesn't de-lidding solve this problem? Or am I just entertaining foolish hopes?

 

[Ascension6]

Delidding would help... but if these temps are within normal range I probably will just leave it...

 

[Phynaz]

How hot is heatsink the during the run? Maybe the heatsink fan isn't spinning up as temps increase?

 

[Rvenger]

Hi guys,

I just finished my new build (i7-4770k @ stock speeds), with HYPER 212 EVO in a fractal R4 case. I added an extra noctua side intake fan to the R4 also.

I have been seeing really high temps during prime95 8-threaded blend tests, wondering if I could get some help.

I used some ancient AS5 thermal paste (maybe 10 years old?), and I was seeing idle temps around low 40s in realtemp. This seemed too high, so I went out and got some antec formula 7 paste. Now the idle temps are lower, around 29-34. I spread the paste rather than doing a dollop in the middle.

however, I still get crazy high temps during p95 @ stock speeds. If I run for 10 minutes or so, it is only around 80. After 15 minutes, I saw it go up to around 100, throttling cut in and I killed the program. I had the case fans set to low at this point, so I reran p95 with them set to high. Again, after around 15 minutes, I saw temps going into 90s so I killed the test. I could deal with 80s--this computer is probably fast enough for me without needing to OC it, but 90s? Too high...

Any advice or suggestions here?

Thanks.

OP, that Antec Formula 7 paste is very thick and I don't recommend the spread method either. In your case, it would be the line method or the rvenger method in the picture below.



This method has been tested by me many of times and proven to be effective.






2 thin lines and a small dot in the middle. You need a little more paste so it will fill the grooves in the hyper 212 evo heatsink.

When preparing the formula 7 paste, grab a tea cup and fill with hot water. I normally line the teacup with a sandwich bag so the syringe doesn't get wet. Drop the syringe in the cup for 5 minutes and then take it out and apply the paste and then mount the heatsink. A little wiggle on the heatsink when installing doesn't hurt before you start tightening the screws.


I have had great temps with that paste so if you do it the way I stated above, you should have optimal temps with that heatsink and paste.

 

[Ascension6]

Good idea about checking how hot the heatsink is. I'm not sure, but I can hear the fan spinning up, and I checked that direction of airflow is correct (pushing towards rear exhaust). Oddly the exhausted air from the case does not seem that hot during p95, so maybe there is a problem transferring heat between heatsink and CPU.

It also makes sense about having a little extra thermal paste to account for the heatpipe grooves in the heatsink... I will try with the 2-line and dot method and repost results...

Thanks

 

[Ascension6]

@Rvenger:

Wow, I just tried your method of filling in the lines on the heatsink first, and then applying with 2 lines and a dot in the center... That seems like a big oversight from CM not to prefill somehow or at least mention these gaps...

I'm running the same test p95 test (8-thread blend) right now (15 minutes in), and temps have definitely dropped a bit, maybe 7-10 degrees on average! Core temps are mostly staying under 80 now, when before they were averaging near 85. Now 85 is the max. Looks like I still got a hot chip...

Oh nevermind, spoke too soon, it just hit 94 and I killed p95 again. Still, definitely cooler than before.

Good idea though, thanks a lot!

 

[blackened23]

Haswell's integrated voltage regulator overshoots the voltage with AVX heavy loads such as prime95. I wouldn't read too much into prime95 temps if you're using a Haswell - i've stated this before but it's entirely possible to have a 100% stable system despite prime95 having insane temperatures during stress testing. This is because the iVR will add +.1V (over auto correcting) which of course will cause temperatures to be absolutely insane. Now, some will say manual voltage prevents this, but it doesn't always. The VR is integrated onto the die, which of course means it has a mind of its own.

Anyway, I wouldn't read TOO much into prime95 temps. If you're under 95C during IBT/prime95 stress testing, that means you'll always be under 70C during *REAL WORLD* usage. So you'll be good to go.

 

[2timer]

Haswell's integrated voltage regulator overshoots the voltage with AVX heavy loads such as prime95. I wouldn't read too much into prime95 temps if you're using a Haswell - i've stated this before but it's entirely possible to have a 100% stable system despite prime95 having insane temperatures during stress testing. This is because the iVR will add +.1V (over auto correcting) which of course will cause temperatures to be absolutely insane. Now, some will say manual voltage prevents this, but it doesn't always. The VR is integrated onto the die, which of course means it has a mind of its own.

Anyway, I wouldn't read TOO much into prime95 temps. If you're under 95C during IBT/prime95 stress testing, that means you'll always be under 70C during *REAL WORLD* usage. So you'll be good to go.

Blackened, are you saying that AVX is responsible for a 25C temperature increase?

 

[blackened23]

Sure. It's possible. Let's say his normal voltage is 1.2V during prime95. Prime95 will generally cause it to "over correct" to 1.3V - this is a well known issue with Haswell's iVR.

The difference between 1.2 and 1.3V, or 1.15 and 1.25V on ivy bridge and Haswell is rather dramatic. Even on my IVB chip which obviously doesn't have an iVR (as Haswell does), the temperature difference between 1.2V and 1.3V is definitely 20c at a minimum. Going past 1.275V with IVB is pretty much the point where temperatures become unsustainable unless you're using a heavy duty cooler (or have a delidded processor). Just to add to this, my temperatures would easily be north of 90C with my 3770k at 1.3V with an H100; heck, they get pretty close to 90C at 1.25-1.275V...

But it's not a HUGE deal. My temperatures during daily use are around 60-65C max, i've never seen it go higher during "real" use.

 

[2timer]

Sure. It's possible. Let's say his normal voltage is 1.2V during prime95. Prime95 will generally cause it to "over correct" to 1.3V - this is a well known issue with Haswell's iVR.

The difference between 1.2 and 1.3V, or 1.15 and 1.25V on ivy bridge and Haswell is rather dramatic. Even on my IVB chip which obviously doesn't have an iVR (as Haswell does), the temperature difference between 1.2V and 1.3V is definitely 20c at a minimum. Going past 1.275V with IVB is pretty much the point where temperatures become unsustainable unless you're using a heavy duty cooler (or have a delidded processor)

Ahhh. I see what you're saying now, the voltage regulator upping the voltage is causing the dramatic heat jumps. Wow, thanks. I'd rep you if I could. :thumbsup:

Is that something that might possibly be addressed with the Haswell refresh? I mean, it just seems like there should be some fix to the voltage, dial it in perhaps. I guess it must be very difficult, and they are probably shooting for the safe side and stability. Just my guess.

 

[Ascension6]

Hmm good to know. I don't expect these temps to ever be reached in real world case. I just tried intel burn test and killed it when it was going into the 90s again.

I am tempted to delid this (I'm confident in my ability to do it), but maybe not worth the bother...

So are these temps normal for haswell based chips during p95?

 

[BallaTheFeared]

What is your stock VID?

Your cpu should be capable of running AVX code paths such as Prime95 with a stock cooler @ stock without throttling.

Hyper 212 isn't the best cooler in the world, but it's better than stock so there must be something else going on.

Just my 2c.

 

[Ascension6]

I think VID is 1.209V. I checked that using CPU-Z with sensor=0 in the ini file.

I only used the Hyper 212 because a friend had an extra one. I would consider switching if there is a quiet / not too expensive cooler that is significantly better cooling.

Everything should be stock unless gigabyte mobos come with some default settings above stock...

 

[Rvenger]

I bet when your temps are shooting in the 90s your vcore is jumping to 1.25v +. Anything over 1.25v temps start getting crazy high on Haswell.

 

[JimmiG]

If I run for 10 minutes or so, it is only around 80. After 15 minutes, I saw it go up to around 100, throttling cut in and I killed the program.

The Blend test starts with 448K FFT size which is quite large, so it's bottlenecked by the RAM, L3 Cache etc. This test takes about 15 minutes to complete, then it switches to 8K which is much more demanding on the FPU since the CPU doesn't have to keep fetching data from L3/RAM. I get about 67C max on the hottest core during the first 448K test with 8 threads/HT enabled, but then it shoots off once the 8K test starts. I can easily get 86-87C on a hot day like today (room temp around 26C). It's the only time the crappy IHS on Haswell actually transfers enough heat to my cooler to make it slightly warm to the touch.

"Small FFT" will start with 12K FFT size right away, so you'll got those hotter temps without waiting 15 minutes.

I think VID is 1.209V. I checked that using CPU-Z with sensor=0 in the ini file.

I only used the Hyper 212 because a friend had an extra one. I would consider switching if there is a quiet / not too expensive cooler that is significantly better cooling.

Everything should be stock unless gigabyte mobos come with some default settings above stock...

That sounds too high. At stock, it should run around 1.1V.

Haswell's integrated voltage regulator overshoots the voltage with AVX heavy loads such as prime95. I wouldn't read too much into prime95 temps if you're using a Haswell - i've stated this before but it's entirely possible to have a 100% stable system despite prime95 having insane temperatures during stress testing. This is because the iVR will add +.1V (over auto correcting) which of course will cause temperatures to be absolutely insane.

True, but what about future applications that actually use AVX? It probably does this for a reason (the FPU might actually need a higher voltage to work properly with AVX). I prefer to have a system that's stable today, and doesn't need adjusting in the future. The whole point of stress tests is to make sure your system is ready for anything. If you're just going to ignore the most stressful tests, you might as well not stress test at all.

BTW is there any way to see what the true voltage is? If anything, my indicated core voltage drops by about 0.002V when I run Prime95 (checked with CPU-Z, Aida64). Maybe manual mode is indeed overriding that voltage boost with my system.

 

[DigitalWolf]

I bet when your temps are shooting in the 90s your vcore is jumping to 1.25v +. Anything over 1.25v temps start getting crazy high on Haswell.

^^ This pretty much. If you take P95, Linpack etc and leave everything in your bios at defaults (no overclock etc) When you run those at certain points where your temp spikes you will see vcore spikes. I've seen over 1.3v on a stock 4670 for example with vcore on "auto". If you use adaptive the same thing happens...


If you monitor your vcore while using your normal apps/games you can see what the average it sits at is.. just for a baseline if you use that as an override aka fixed vcore then run those same tests.. your temps will drop a lot. Linpack is a great example of how to get insane temps with anything but a fixed voltage..


Altho for normal gaming/applications auto/adaptive seem to work just fine in regards to temps.

 

[Ascension6]

Well, you are are right about the 12k FFTs causing the heat. Running that for about 5 seconds caused the temp to go to 95C before I killed it. Core voltage is set to auto/adaptive, but it was still only reading 1.209V. Maybe my chip has particularly bad internal TIM. I am tempted to delid it just for kicks and see what happens.

 

[Rvenger]

What is your frequency at that voltage?

 

[Ascension6]

3.9GHz

 

[ricardo.vix]

I don't want to be annoying, but didn't you apply too much thermal grease?

Try to apply a little less.

 

[szu]

Obvious suggestion, but have you updated your BIOS to the latest version? My haswell processor hit a vcore of 1.236V on the shipped BIOS (which was officially incompatible with my CPU) and temperatures in the 90s. After updating to the latest version, vcore peaks at something like 1.16V and temperatures don't go above 80C.