Delidding an i7 4790k?

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860
So where to start?

I'm aware of the vice method, as well as the razor method. I believe the vice is the safer method?

Also, would MX-2 thermal paste help me at all?

Basically the reason I want to delid my i7 4790k is this: it runs hot. I have an NHD-15 with a single fan running. Case airflow is exhaust out the top and back, and intake from the side and front. Ambient temperature is around 23-25C (varies depending on if some idiot turns the air off).

My i7 4790k with the NHD-15 installed with a single fan runs at about an average of 39C per core.

That is pretty high for me, and I suspect that the cheap thermal paste on the inside of the heat spreader is to blame

. I run it at stock, so 40C is a bit high for an idle temp with a cooler like the NHD-15. I don't expect the idle temp to go down more than 8C at idle, but load temperatures should improve tremendously from what I've read, and that is definitely going to determine how high of an overclock I can get. I do not plan to overclock right now, but I do not want temps past 65ish when stress testing if possible (at stock 4.4Ghz turbo speeds obviously)

Anyways here are my current settings:

i7 4790k @ 4.0Ghz with 4.4Ghz Turbo enabled

Core voltage (from BIOS) is around 1.00, Core voltage (from CPU-Z) is exactly 1.191 (why such variation?)

NHD-15 heatsink with single CPU fan running at about 700rpm idle

(1) Front 230mm fan at 1500rpm (intake), (1) Top 200mm fan at 700rpm (exhaust), (1) Side Panel 200mm fan at 700rpm (intake), (1) Rear 140mm fan at 2000RPM (exhaust)

Case is a Cooler Master HAF X, and it is missing only one top 200mm fan (didn't come with the case like the others).


 
Solution
You can realistically hit temps in the mid 80's on p95 or occt and still be fine. It's only when getting close to 100 that you should really start to worry. The Haswells run hot anyways due to their on-die VRMs, the memory voltage regulators, that were separate on the Ivy and Sandy Bridge cpus. Manually set dram voltage to 1.5, don't set memory any higher than XMP value.

Side note, prime95 will run cpu at @115% with small fft, Occt is 100% cpu and small % ram, Occt Linpack is @ 100% cpu and 90% ram (temp monster for haswells). intel Burn is @125% cpu. For overclocking purposes, these are run for 2 reasons.
1. Punish the cpu beyond normal heavy gaming usage. Simply, if you don't cook your cpu with one of these tests, you'll never have...

Darthutos

Reputable
Sep 15, 2014
757
2
5,160
reseat your D15. Use the thermal paste included with the packaging.
Use only a drop, pea size maximum. Do not spread. The d15 spread it for you with the pressure at installation.
Do not use the low power draw cable. Plug the HSF plug directly into the Mobo. Run fan at maximum.
Good luck
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


Is the paste that comes with it any better than Arctic MX-2? Also I did the pea size method and let the pressure spread it.
 

Darthutos

Reputable
Sep 15, 2014
757
2
5,160
I don't know if it's better than Arctic MX-2. The NH-1(exact name) that came with my D14 was exactly like the one I bought for 10 dollars performance wise.

FYI I overclocked my 4770k for 4.5ghz and my temperature according to realtemp gt is about 36 on all cores at 3 percent load.
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


Did you delid it at all?
 

Darthutos

Reputable
Sep 15, 2014
757
2
5,160
no.

I just thought of something.
Some guy on youtube mentioned that his mobo automatically put 1.5 volt (gasp) on his chip 4790k same as you (some gigabyte mobo) and he had to lower it manually.

Check your vcore. Higher vcore still makes your system stable, but won't help your temps.
 

clueless77

Reputable
Jun 5, 2014
146
0
4,710
If you're going to delid, use Cool Laboratory Liquid Pro. Out of user experiments that compared the different brands of TIM, apparently Liquid Pro is the most effective for delidding. If you're not going to wear an anti-static wristband (twice and I never have), make sure you're delidding in a low static environment and try to discharge yourself to your case beforehand. If you're going to use the hammer/vice method rather than the vice/inwards pressure method, apply force to your hits but not so much that the PCB will fly off the IHS and land somewhere. 4-6 good hits should be about enough to loosen the PCB from the IHS so as to separate them from each other by pulling it off.

The variation could be higher because of adaptive load voltages if you have turbo boost enabled.
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


Well my Vcore is automatically set to around 1.0, so it's probably not that. It can one of two things. Either I need to reseat my heatsink (which I will try first), or I need to delid. Also, the Noctua installation guide said to take some kind of protective cover off of the bottom of the heatsink. However, all there is is the silver plate. Is that the protective plate? It would explain my issue a lot, but I'm pretty sure it is not.

Nevermind. I saw what the protective plate looks like (it's plastic) and mine didn't even come with one attached, so its not that)
 

clueless77

Reputable
Jun 5, 2014
146
0
4,710
You mean manually set? Are you sure that you have your voltages set to fixed and not adaptive or auto? Because that voltage you described is my stock turbo voltage at load for my 90k, but that voltage isn't your problem. The protective cover on your heatsink should have just been some type of plastic casing, while mine is copper plated that silver plate is the thermally conductive plate on the heatsink.

I was also getting 40 c at idle at stock with this chip and a h80i, it's down to ambient temps now. It's not bad if you can easily keep your load temps in check though.
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


room temp is usually around 23 celsius. Load temps go insane with prime (hit 90) and past 70 with AIDA64.

I'm not worried about my idle temp as long as I can get my load temps down to a reasonable level. As much as I'd love it, I don't expect 4.4ghz to run under 40C (as that is the same temp as my FX 6300 @ 4.4ghz), even though people have reported near 33c at idle. I just want to keep my load temps as far under 70 as possible.

I just don't see how in the world an NHD-15 (even with a single fan) isn't keeping my processor under 70 at stock speeds. First I'll try reseating the heatsink, then I may have to delid if it doesn't help. I'm not sure what to do :/
 

Darthutos

Reputable
Sep 15, 2014
757
2
5,160
It doesn't make any sense. i7 4790k was supposed to have better internal TIM than 4770k, and people on youtube have reported much better temps.
and the new nh d15 can't be worse than a nh d14...
How old is your nh d15? maybe you could get a refund and buy a coolermaster nepton 280L or something.
Before you delid, try everything possible. Once you delid, your warranty is gone. Maybe you could tell Intel you got a defective 4790k?
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860
It's brand new. 2 days old. I'm thinking it's because I don't have the second fan installed, but my rear fan runs at 2000rpm, so it should easily pull whatever the Noctua is pushing. I can't get a solid temp reading (they go from 30 to 50 in a second, etc), but I do know that it skyrockets when I use prime95 to torture test the turbo stock speed. I was going to try an H110, but it doesn't fit my case. I think if noone can help me resolve this situation in a few days, I'm going to order a corsair h100i (which does fit my case) and maybe that will cool this son of a gun. I know Haswell runs hot, so I don't expect idle temps to be crazy low, but I do expect load temps not to exceed 70 celsius when my AMD build with a Evo keeps temps to about 65 on a 4.4ghz torture test with much higher voltage (turbo voltage 4.4ghz on Intel is about 1.191, where as my AMD FX 6300 runs at 4.4ghz with 1.35 volts, and does not exceed 65 on prime95 torture test). If I use the evo I might be able to throw the Noctua fans on it (would that help?) and see how that goes, but I'd rather keep that for my AMD build. I want to exhaust every possible option, because as you said, it voids my warranty to delid my processor, and I want that warranty. However, I'd rather not have a warranty and be able to overclock it (which voids warranty too) than have a warranty and get insane load temps.

The reason I'm thinking about using the Evo's heatsink is because it is smaller and I would probably be able to fit both of the Noctua's on it, even with my insanely high RAM heat spreaders (they are installed in A2 and B2 so that also gives me a tiny bit of clearance on the side, but not enough to install a second fan on this huge NHD-15 heatsink. Also, I'm not 100 percent correct, but does the heatsink matter that much as long as you have good fans? I mean, its just a huge tower of fins.

So if anyone has any suggestions, PLEASE help me :/

If not, I will try the h100i and see if it helps.
 

Darthutos

Reputable
Sep 15, 2014
757
2
5,160
The reason I'm not using the h80i right now is because the backplate problems.
I had trouble making the heatsink stick to the cpu ihs, resulting in insane temps. like 80 degrees celsius in bios.
Some people told me that h100i backplate is the same as the h80i backplate, if so, I suggest you invest in a couple of washers.
In fact, more than a couple to make sure that the ihs and the h100i have good contact.

edit: something tells me that the evo won't help. I mean the d15 has much better surface area than the evo. If the evo give you better temps, it would be because you give the cpu better seating.

I myself is considering getting a coolermaster seidon 120 or 240 (the nepton 280 won't fit into my case.), but what am I supposed to do with a d14 that's left over?

 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


You could sell it, or save it for a new build. Are the seidons better than corsair?

Here is what I plan to do:

1. Try reseating heatsink and applying Noctua thermal paste

2. If 1 doesn't work, I'll try my Evo.

3. If 2 doesn't work, I may look into water cooling.

4. If nothing works, I will delid my CPU and see if it helps.

I'm thinking the Evo MIGHT work better because it has less surface area. This way the heat is transferred to the fins quicker vs. through both towers of fins and thus the heat is taken away and out the back (or top) of the case. Of course, I would use one Noctua fan on one side, and the other on the other, so that I would have a push pull config (even though my 2000rpm rear exhaust fan is sucking air like crazy), plus the Noctua NHD-15 comes with two 140mm fans. I'm worried that the 140mm fans wont fit on the Evo though

I'm also thinking that the washers provided by Noctua may not allow enough contact for the heatsink, so if reseating it doesn't help, I may try that as well.
 

Karadjgne

Titan
Ambassador
The EVO is a budget cooler. It's usually good for somewhere around 4.3-4.4 GHz, on a 4 core Intel. You throw it on the i7 and enable HT and it'll act similar to the stock cooler when turbo engages. Max out the fan and get loud. Using 2 fans push/pull will get you @2° cooler, pretty useless and noisy for little return. The EVO uses 120mm fans, the 140mm Noctuas won't fit. Understand, its not just the fans but it's mainly the surface area of the fins combined that dissipate the heat that make for a good cooler, the d15 has almost 3x the surface area of the EVO, and more/larger heat-pipes, so will dissipate a lot more heat given adequate ventilation .

Put the other fan on the d15. If you have ram clearance issues, raise the front fan a little higher till you have no further issue.

Most brands of ram will allow you to remove the heat spreader and since it really doesn't do anything except in extreme overclocks, you can safely remove them if possible. They are usually more for looks than any actual purpose.

The Noctua paste is better quality and performance than the mx2.

Your exhaust fan does absolutely nothing to help any tower dissipate heat. The draw from the fan comes from every direction and is not concentrated enough to directly pull enough from inside the fins to count. The fins are far too much of an obstruction, so the exhaust fan will get its air supply from someplace easier.

Conclusion : repaste (correctly), and put the fan back on the d15! Forget about the EVO.

Btw, also forget about comparing voltages between AMD and Intel cpu's, they are totally different architectures and manufacturing processes.
 

Damo66676

Honorable
Jan 9, 2014
14
0
10,510
I had an i5 4670k that was running hot at load, I used the razor blade method for delidding as the vice way looked a little rough. Replacing the TIM I used Cool Laboratory Liquid Pro, as I heard it to be the best out there. My temps dropped by 25 degrees on load.
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


Thanks for the information :) I cleaned off the processor and NHD-15 with some isopropyl alcohol (91% is the best I found around the house). I applied the Noctua thermal paste, and screwed the heatsink as much as I could (basically until the screws wouldn't go anymore. I didn't force it after that). I also took a quick look at G.Skill's heatspreaders and there are two screws on the side of Trident X'es, and then they just slide off. Since I'm running them at 1600mhz right now anyway (vs. the 2400mhz factory rated speed), they wouldnt do much good if they even work for that matter. So now I'll throw the second fan on and see how much of a temp difference I'll get. I'll have to move both the fans slightly off to the side (since the RAM sticks themselves are still too close to the fan), but hopefully I'll get better temps.

So here's what is going on now. I have both fans installed (heat spreaders for ram were a piece of cake), and AI Suite III is reporting a temp of 60C while doing a Prime95 torture test. However, I'm not sure of the reliability of AI Suite III, as I've gotten mixed reviews about it. Is there a more reliable method for measuring temperature?

AIDA64 reports a maximum temp of 70C on Core 3 (highest out of the 4), so I'm not sure which temp to go with.

Also, there seems to be an odd issue with both AI Suite III and AIDA64, or rather maybe even my own logic? They are both reporting 59 celsius for the CPU itself (and I did these one at a time. I know having more than one temp application open at once will cause conflicting results) and yet all the cores are almost hitting 70 celsius. How in the world does 70C per core average out to 59C? >.>
 

Karadjgne

Titan
Ambassador
AiSuite is actually very good, until you talk temps on the monitor tool. Those temps come from a predetermined sensor and are not always accurate. Fan Xpert2 uses the bios temps (much more accurate). AiSuite would be reading my cpu temp as 40° during prime95 when I knew for sure it was @70°. Best bet for Intel cpu's is speccy, realtemp or coretemp, which read the actual core temps not the socket or package temps. Also, some software will read just a single core and claim that as cpu temp, so if reading 59° then at least 1 core (usually core #0) was 59° while other cores (for me core #1 is the hottest, then core #3, core #0 is the coolest) may well have seen 70°. It also depends somewhat on thread usage. It's possible Windows is shunting more usage to core #1, with spillover to core #3, so those cores will run hotter than cores #0,#2. If you use realtemp, you can set it to read 'max' which reads all 4 cores then shows which one is running the highest temp. OCCT rums similar in its stress tests and has a setting alarm to shutdown the test upon broaching that temp, something I find better than prime95

Either way, temp readings from software should be taken with a grain of salt, used as a guideline, not as gospel.
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


Still getting about 38C avg per core on idle with both fans installed.

Load temps are around 70C avg.

On the topic of coolers, how would a Swiftech H220 or H320 fair against an NHD-15? I have room for a 360mm radiator which is why I even looked at the 320, but I've heard good things about the 220. I know it is water cooling versus air, but I feel that until I get lower load temps than 70 I'm not going to be comfortable. The 320 is about 25 US dollars more than the 220 so price isn't much of an issue. I just want the best performance with the lowest noise possible... I mean I can deal with SOME noise, but I don't want it to sound like a jet engine when its running. I have a box fan I run to keep my room cool and that drowns out almost all the other noise in the room, including the noise from my PC, so as long as it isn't irritatingly loud, I can deal with it. I'm not sure whether I should try delidding first or if I should just go with the water cooler.
 

Karadjgne

Titan
Ambassador
Ok, its a start. Now comes fan curves. In bios, set all your fans to performance, with duty cycles at 100. In AiSuite, find the correct fans and set them up according to tastes. The Noctuas have a very low activation threshold, in the neighborhood of 18% speed, or about 400 rpm. I'd raise that some, to @ 550-600rpm for idle temps upto @40° then start curving up till 100% @65°. Your other fans you want to be sure to provide adequate airflow so when the cpu fan ramps up, so does intake and exhaust, but those fans will read case temps so set those to reach 100% by 50°. They will not go as low starting threshold as the Noctuas so expect @50-60% to start.

Btw, the d15 is king of the aircoolers, it'll put idle temps of @3° above case ambient temps at idle, easily comparable if not better than a Corsair h-100i. If your case temps are hot, any cooler, liquid or air, is going to see higher temps. If your case ambient temp is 40°, the best you'll see is a cpu temp at 43° idle. Liquid, clc or full loop, will not change that unless you use the cooler as intake, thereby using room ambient temps. This has the drawback of putting all the cpu exhaust heat back into the case, making it that much hotter, which will directly affect gpu temps, as the gpu will now be using hot air to cool itself. Get case temps down, cpu/gpu goes down accordingly.
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


Well all of my case fans are running off of Molex adapters, so I'm guessing they are near 100% or is that incorrect? The only fans under BIOS control are the two Noctua fans for the CPU.

Noctua's are running at 30 percent from 30c, 50 percent at 45c, and 100 percent starting at 60c. Still getting average temps of around 70 celsius under full load Prime 95 with maximum heat.

Also, here is my case airflow setup. I have one front 230mm intake fan, as well as one 200mm side panel intake fan. I have one rear 140mm exhaust fan, and one top 200mm exhaust fan. Ambient temps are about 23 celsius avg.
 

Darthutos

Reputable
Sep 15, 2014
757
2
5,160
Noctua's are running at 30 percent from 30c, 50 percent at 45c, and 100 percent starting at 60c. Still getting average temps of around 70 celsius under full load Prime 95 with maximum heat.

That's a pretty good temperature for the noctua.
 

swiftleeo

Distinguished
Jul 20, 2014
448
0
18,860


I suppose but I'd prefer 60C at full load on stock... May just have to delid.