Can this delidding catastrophe be salvaged?

[crashtech]

@DigDog: These kinds of questions are hard to answer because you have to deal in averages using very incomplete data. From my very unscientific observation, delidded Haswells still have a frequency deficit of approximately 400MHz vs Sandy. Even at the slower average clockspeed, Haswell maintains the IPC advantage, though not by a whole lot.

Maybe the fact that Sandy Bridge can't be delidded makes your question hard to answer, or I don't understand what you are asking.

 

[aigomorla]

incidentally, how does a delidded 4770k compare to a delidded 2500k? ermmm .. let me rephrase that.

the difference in OC room between lid-on / lid-off, is it much greater on haswell than it is on SB?
Did Intel make a chip (Haswell) which is a real improvement over SB, but then fucked up with the TIM / lid ?

no..

as others said the reduction in temp was great.
however once u hit a certain voltage wall, unless ur on water or phase, u cant bring the cpu temp down by conventional means.

@DigDog: These kinds of questions are hard to answer because you have to deal in averages using very incomplete data. From my very unscientific observation, delidded Haswells still have a frequency deficit of approximately 400MHz vs Sandy. Even at the slower average clockspeed, Haswell maintains the IPC advantage, though not by a whole lot.

Maybe the fact that Sandy Bridge can't be delidded makes your question hard to answer, or I don't understand what you are asking.

that 400 is highly dependant on sink.
If you had a cold plate with water, im sure u could go up even more.
Add phase, and u'll be really milking the efficiency at the extreme end.

Im starting to think the vregs built onto the cpu is whats causing this runaway heat ramp after u hit a certain voltage.

 

[Kenmitch]

No, but maybe next week if I can summon the courage

What about the individual core temps? You can check the maximum recorded temps in the Statistics page in Aida64. Cores will be at least 10-15C higher than package.

I believe Tcase (maximum temp allowed for the package) is 72C.

The highest core was 78c coolest upper 60's.

Thinking my gap under heatspreader isn't too bad/uneven. Maybe just some Costa Rican humidity trapped in there.

If I decide to delid Im gonna use a digital microscope to view the edges of the heatsink 1st. Think I can do 200x with it and take photos.....If I cand find the software still.

 

[blackened23]

These temperature reductions are extreme. Intel should have spotted this because the benefits definitely outweigh the cost reductions and negative PR.

That isn't even the main issue with Haswell. The main issue is that there is a lot of variance from chip to chip in terms of voltage required to reach a desired clockspeed - let's say the best 4770k chips can reach 4.6GHz at 1.15V. There are also 4770k chips that can't reach 4.6ghz unless they are 1.35V. That's the thing with Haswell, there is more variance with voltage. By contrast, IVB and SB had less variance in terms of voltage required for desired clockspeeds. So delidding will lower the temperatures, but it doesn't remove the voltage variance which is a side effect (presumably) of the new iVR of Haswell. Once you reach a certain voltage threshold, nothing will make temperatures tenable whether you have a delidded processor or not.

All this being said, Haswell isn't a bad overclocker. Most chips will reach 4.4ghz easily, and that 4.4ghz 4770k will beat a 5.2ghz SB and will also beat a 4.8ghz IVB. Also, asus has tested thousands of chips and something like 70% of them will reach 4.4ghz with nominal voltage - that really isn't bad. The overall clockspeed will be lower but it is still the best purchase for a new system; it just becomes questionable as an upgrade for existing SB or IVB users.
Obviously most users don't consider it a worthwhile upgrade.

Another complication to consider is that stress testing really doesn't work well with Haswell; since Haswell has an iVR it tends to "over" correct the voltage whenever AVX stress testing is used. That means it will add another .1V on top of what your upper most LLC setting is, and that obviously will wreak havoc on temperatures during stress testing. But you won't get anywhere near those temps during real world application use.

 

[JimmiG]

The highest core was 78c coolest upper 60's.

Thinking my gap under heatspreader isn't too bad/uneven. Maybe just some Costa Rican humidity trapped in there.

If I decide to delid Im gonna use a digital microscope to view the edges of the heatsink 1st. Think I can do 200x with it and take photos.....If I cand find the software still.

That's pretty sweet. Mine top out at 87C running the FPU-only test with HT on, and about 5C lower with HT off. I think I would definitely benefit from a de-lid since it's the temps that are holding me back.

You can get around the voltage boost by using a fixed voltage, BTW. I've never seen the voltage go about the 1.218V I've set in BIOS, even when running AVX tests. If it did add another 0.1V, my temps would be over 100C for sure.

 

[Chipfiref]

To go back to when I still had the stock cooler before delidding, I found a picture of when I removed the Intel stock cooler - it had been installed for a couple of weeks at this point. Can anyone judge from the picture whether I had a good even contact?

 

[crashtech]

Did you say one pin may not have been functioning? That seems plausible from the pic, yet there was still full contact, just less pressure in the upper right corner.

 

[Kenmitch]

That's pretty sweet. Mine top out at 87C running the FPU-only test with HT on, and about 5C lower with HT off. I think I would definitely benefit from a de-lid since it's the temps that are holding me back.

You can get around the voltage boost by using a fixed voltage, BTW. I've never seen the voltage go about the 1.218V I've set in BIOS, even when running AVX tests. If it did add another 0.1V, my temps would be over 100C for sure.

I've been using a fixed vcore pretty much from the getgo. Pretty much does not fluctuate at all like you said, maybe 1.205-1.207v for example.

Offset vcore control is a pain in the arse with Haswell.

I'm currently experimenting with offset vcore. The xtra .1v makes it a much harder overclock than fixed vcore value. Not sure if I can get a overclock this way that will satisfy the Nerd inside me tho.

Well I'm just about ready to give up on offset vcore mode. Seems like anything above 4ghz requires higher voltage than needed.

Example: What it requires for 4.2ghz with offset vcore = What is required for 4.5ghz with a fixed vcore.

Guess I can try another multiplier or two to see. Doesn't look promising tho.

To go back to when I still had the stock cooler before delidding, I found a picture of when I removed the Intel stock cooler - it had been installed for a couple of weeks at this point. Can anyone judge from the picture whether I had a good even contact?

Kind of hard to tell. The original Intel TIM is kind of thick and somewhat plastic like. I'd imagine some stays on the heatspreader and some on the heatsink when removed. Guess would need both images to decide.

We can all agree on the OE cooler sucks big time with Haswell.

 

[JimmiG]

To go back to when I still had the stock cooler before delidding, I found a picture of when I removed the Intel stock cooler - it had been installed for a couple of weeks at this point. Can anyone judge from the picture whether I had a good even contact?

Hard to tell without also seeing the surface of the CPU heatsink. However I believe the Intel stock cooler comes with thermal paste pre-applied so it should be fine. Even if you had an even layer to begin with, removing the cooler can mess it up.

The most common problem is applying too much. As long as you apply the right amount, roughly in the middle, and mount the CPU cooler properly, you'll get a nice, even layer. Remember, it's only for filling in *microscopic* gaps between the heatsink and IHS.

Offset vcore control is a pain in the arse with Haswell.

No kidding. I couldn't sit down for days after trying to use offset Vcore

The surprising thing is that idle temps are great even without the benefit of lower Vcore @ idle. I think it simply power gates idle cores, so adaptive frequency/clock speed are not as important as it used to be. I get about 27C idle with the CPU still running full blast @ 4.3 GHz/1.218V.

 

[Chipfiref]

Hard to tell without also seeing the surface of the CPU heatsink. However I believe the Intel stock cooler comes with thermal paste pre-applied so it should be fine. Even if you had an even layer to begin with, removing the cooler can mess it up.

The most common problem is applying too much. As long as you apply the right amount, roughly in the middle, and mount the CPU cooler properly, you'll get a nice, even layer. Remember, it's only for filling in *microscopic* gaps between the heatsink and IHS.

The last few Intel stock coolers I have installed have the paste in a semi solid form, shaped on the heatsink like this:

But those pins don't like to stay evenly inserted...

 

[spat55]

Glad to see it working!