Noctua NH-D14 - Assumption is a waste of time to lap heatsink...

[Caveman]

Yes?

Using it to cool a 4770k...

 

[jji7skyline]

Probably belongs in Cases & Cooling?

I think it's only worth lapping if you lap both sides. Do you have a good 4770K?

 

[Caveman]

Probably belongs in Cases & Cooling?

I think it's only worth lapping if you lap both sides. Do you have a good 4770K?

Thought this was as much a CPU question... Not sure what you mean by "good"...

 

[jji7skyline]

Thought this was as much a CPU question... Not sure what you mean by "good"...

A lot of 4770Ks apparently had issues with the TIM wasn't properly applied between the IHS and the die, resulting in bad overclocking performance regardless of what cooler was used.

 

[soccerballtux]

I didn't get anything by lapping mine.

I think the TIM is the most important part. in fact, with small enough TIM particles, more surface area, could get even better conduction, imho

 

[Caveman]

A lot of 4770Ks apparently had issues with the TIM wasn't properly applied between the IHS and the die, resulting in bad overclocking performance regardless of what cooler was used.

Well, that's a bummer... Watched some videos... The die is the CPU itself and the IHS is the copper "cover" that has the TIM sandwiched in between?

 

[Ferzerp]

in fact, with small enough TIM particles, more surface area, could get even better conduction, imho

It doesn't work that way... At all.

 

[MongGrel]

I used to lap em decades ago, my D14 I didn't bother a few years back.

Waste of time I'd think, most come pretty nicely flat these days, unless you received an odd ball.

 

[BonzaiDuck]

Somehow, I don't think lapping the heatsink base is going to make it harder to either re-sell or re-use, whatever the impact on "warranty." If you take the usual caution while pushing it around a piece of wet-or-dri 400-grit, there's no risk to it; just make sure you grasp it by the base without stressing the pipes or fins.

It will make more performance difference for you if you lap the IHS of the processor, though, but it would have an effect on any resale value. If done with a modicum of care, there shouldn't be any risk to the processor either. Use latex gloves; use only a couple drops of water on the 400-grit; clean off the IHS to remove copper particles at regular intervals in the sanding.

Bare copper surfaces will conduct heat better than nickel-plated surfaces. This -- in addition to perfectly flat surfaces.

Unless you plan to de-lid the 4770K, this is the best you can do in face of the TIM problem that was likely more troublesome with the Ivy Bridge cores. The only other thing I'd recommend is the choice of TIM between the heatsink base and IHS. I use a micronized diamond TIM -- "IC Diamond" by Innovative Cooling, which is troublesome to spread but can be re-used by refreshing it with any old TIM with a silicon grease.

Diamond is second-best, and Liquid Ultra or "CLU" is probably the best performer.

I just believe in all these measures because reducing temperatures a few degrees this way or that way adds up.

 

[Caveman]

Thanks for the knowledge folks...

So, from what I gather, since the problem was more of an IB than Haswell thing, it might pay for me to leave it well enough alone... That said, I can't stand the thought of not being "optimized". I have experience working on delicate space-flight hardware in a clean room environment, so I'm not concerned about breaking things per se... Just would feel bad if I made a mistake and blew $300.

One thing that surprises me is how most seem to use the same IHS instead of cooling the die directly... WHY???

 

[Ferzerp]

One thing that surprises me is how most seem to use the same IHS instead of cooling the die directly... WHY???

Because most of us found that the difficulty in getting good contact with the tiny die made us end up with better cooling by leaving it on and just repairing the issue.

We also found that certain replacement materials were an extremely poor option because they would work for a month or two and then not anymore (AS5 for one).

 

[MongGrel]

Last one I ever lapped was a solid Copper ThermalRight I still use on my HTPC that used to be on a Q6600 at 4 Ghz.

It's still hanging around, with a couple Noctua fans in the bedroom.

It is running a slightly overclocked X5650, which is of course a lot faster, and just hanging out these days.

 

[crashtech]

The difference after lapping is usually so tiny. Once in a while you will get one of those direct pipe contact HSFs (the D14 is not one) where one or more pipes have shifted their position, causing some of the heatpipes to have poor contact. In these cases some remedial surfacing is called for to bring the surface back into true.

 

[PhIlLy ChEeSe]

Wouldn't he be better off De-lidding the CPU?

 

[Yuriman]

Thanks for the knowledge folks...

So, from what I gather, since the problem was more of an IB than Haswell thing, it might pay for me to leave it well enough alone... That said, I can't stand the thought of not being "optimized". I have experience working on delicate space-flight hardware in a clean room environment, so I'm not concerned about breaking things per se... Just would feel bad if I made a mistake and blew $300.

One thing that surprises me is how most seem to use the same IHS instead of cooling the die directly... WHY???

Wouldn't he be better off De-lidding the CPU?

Most mounting hardware won't work if you take the IHS off, because the CPU will not be nearly as tall, so you'll have to fabricate something.

For years I ran a Swiftech waterblock mounted direct-die with custom mounting hardware, and it was worth 5-10c at the extreme, but ultimately it did nothing for day-to-day use. The voltage needed for a given clock remained the same, and I wasn't any more comfortable putting over 1.3v in it for more than a few hours of benchmarking. If I could do it again, I'd probably have left the lid on and kept the resale value.

 

[DrMrLordX]

Diamond is second-best, and Liquid Ultra or "CLU" is probably the best performer.

Actually, IX is the best.

One thing that surprises me is how most seem to use the same IHS instead of cooling the die directly... WHY???

If that is your question, then . . .

Most mounting hardware won't work if you take the IHS off, because the CPU will not be nearly as tall, so you'll have to fabricate something.

. . . this is the most common answer. Furthermore, sometimes you will not see an appreciable gain from removing a properly lapped IHS. Think about it: it's just a slug of copper. So long as the contact surfaces are mated properly with stuff like CLU, CLP, or IX, adding another millimeter or so of copper between the die and the dissipative surface is not going to significantly harm thermal conduction. What the direct-die guys are going is eliminating non-copper materials (nickel and TIM or nickel and solder) between the emitting surface (the die) and the dissipative surface (the HSF/rad fins). They are also eliminating the potential for less-than-ideal mating between the IHS and the die. When you do a delid + relid, you can mitigate that problem somewhat.

Then there are HDT HSFs, which are difficult to mount directly to the die. Having an IHS often makes them work better.

 

[crashtech]

Direct die also eliminates one layer of TIM.

 

[Idontcare]

Benefits of lapping NH-D14

Unless your CPU doesn't have solder under the IHS

In which case delidding gives you benefits

 

[BonzaiDuck]

Actually, IX is the best.

If you mean Indigo Xtreme, I would agree, but there's the drawback of (a) first getting it to melt by heating up the CPU, and (b) removing it when you need to.

I forgot about that option. It did test out marginally better than anything else. And the reasons I forgot about it were the extra hassles.

On the delidding. There are certainly the risks. But then, you have to weigh the improved "polymer TIM" against using CLU -- which as I recall, was the consensual choice of delidders. I dunno -- ask IDontCare -- he knows.

I've seen this or that member get some decent temperatures with the 4790K -- overclocked -- without any delidding and only extra care and measures in cooling. So personally, I'd just grind off the nickel-plate and otherwise leave it alone.

 

[MrTeal]

Benefits of lapping NH-D14

Unless your CPU doesn't have solder under the IHS

In which case delidding gives you benefits

Interesting posts on the lapping of your AOI coolers, I missed that post when it was active. I'd add that I've had some really bad experiences with the surfaces of Asetek AOI coolers. I'm not sure what the manufacturing process in on them and when they machine the cooling engine vs surfacing the mounting plate, but I'd almost speculate that they might cut the channels on the cooling engine after they create the flat for the bottom. Trying to do direct die attachment I've seen some AOIs like the H55 that were so convex that they actually cut in to the cooling channels before the blocks got flat.

Mirror finish really isn't as huge a deal as some make it, but the terrible flatness of some coolers is an issue for getting good temps, especially if you're attaching to a die where you can't just rely on the mounting pressure deforming the IHS to get good contact.

 

[Joepublic2]

Probably a waste of time if you're not going to delid/operate on the chip and lap the heat spreader. Wish Intel would get some good x86 competition cause they're really being cheapskates these days with their TIM (no engineering reason not to use solder, big haswells get it) and concave heat spreaders.

I got a 20C reduction in load temps from delidding/relidding my haswell (hammer + vice method, CLU) using a hyper 212+.

So personally, I'd just grind off the nickel-plate and otherwise leave it alone.

It's not the nickel plate that's messing things up it's:

1) Top of IHS is not flat (most that I've measured have been concave although a few were convex/wavy)
2) Bottom of the IHS isn't making good even contact with the cpu slug. Cores 1/4 should always be coolest (they are after I rebuild them, and within about 5C of each other) because they're surrounded by darker silicon. Before working on them I've seen all kinds of different temperature deltas/huge spreads (up to 15C) between loaded core temps.

 

[Caveman]

So, I gather from more research that the root cause of the overheating issues on IB and Haswell is poor contact of the IHS to the Die, not the TIM material on the Die... Poor contact due to poor tolerance control of the Si based epoxy "lifting" the IHS too high off the surface of the Die giving poor contact.

So... Assumption is everyone removes the gray/black Si Based epoxy when they do the delidding... Do they adhere it back down somehow? What holds the IHS to the Die at that point?

BTW... After research a year or so ago, the undecidedly best TIM was PK-3 because it offered the best bang for the buck so to speak as far as dropping temps the most but being easy to use... Since it is insulating... It was only 1-2 degrees out from the other more liquid style thermal compounds yet 10X easier to use according to several sources that had done comparisons of multiple TIMs.

 

[Yuriman]

When you clamp the socket retention mechanism down, it presses the headspreader flat against the die, so you don't need to glue it down. In fact, if you remove the headspreader, you need to remove the metal retention mechanism because it's taller than the CPU die (if I remember correctly), and the CPU will actually need to be held in by pressure from the base of the heatsink.

I use Noctua NT-H1 between the IHS and heatsink, and CLU between the die and heatspreader. Don't use liquid metal thermal compounds with anything non-copper.

 

[DrMrLordX]

Direct die also eliminates one layer of TIM.

Well yes, though the actual TIM itself (contact problems aside) is no problem if you're using IX or CLU or something like that.

Mirror finish really isn't as huge a deal as some make it, but the terrible flatness of some coolers is an issue for getting good temps, especially if you're attaching to a die where you can't just rely on the mounting pressure deforming the IHS to get good contact.

Correct. Even before I started using CLU, I only lapped to 800 grit and saw gains from that. Now I stop at 400, which is the minimum to flatten out irregularities on the IHS and get rid of the nickel. CLU alters the surface texture, making lapping irrelevant except from a concavity/convexity standpoint.

So... Assumption is everyone removes the gray/black Si Based epoxy when they do the delidding... Do they adhere it back down somehow? What holds the IHS to the Die at that point?

I can't speak for the Intel delidders, but my strategy on Kaveri was to remove the epoxy from the underside of the IHS, but to leave the epoxy on the PCB. That "lowered" the IHS a bit, but it still gave me a guide to follow when replacing the IHS prior to installing the HSF. The epoxy remnant was thick enough that it held on to the IHS, preventing it from moving anywhere, and installing the IHS produced a mild rebonding effect so that the IHS was partly glued on there. It was easy to pull it off by hand, but it did require minimal effort.

BTW... After research a year or so ago, the undecidedly best TIM was PK-3 because it offered the best bang for the buck so to speak as far as dropping temps the most but being easy to use... Since it is insulating... It was only 1-2 degrees out from the other more liquid style thermal compounds yet 10X easier to use according to several sources that had done comparisons of multiple TIMs.

If I am going to go through all the trouble of delidding my CPU and lapping the IHS, then I'll go through the trouble of spreading around CLU with a clipped brush. That's just me, though.

 

[Joepublic2]

So, I gather from more research that the root cause of the overheating issues on IB and Haswell is poor contact of the IHS to the Die, not the TIM material on the Die... Poor contact due to poor tolerance control of the Si based epoxy "lifting" the IHS too high off the surface of the Die giving poor contact.

So... Assumption is everyone removes the gray/black Si Based epoxy when they do the delidding... Do they adhere it back down somehow? What holds the IHS to the Die at that point?

Yeah pretty much. Luckily I've had access to a dead chip to play around with (burned out during a LN2 overclocking/overvolting session) before experimenting with live ones. I've been getting good results cleaning off all the silicone epoxy from the package/IHS and applying 4 small drops of sensor safe RVT, two to parallel sides, against the outside edge of the IHS, not underneath the edge, and clamping down the reassembled CPU (using my thumb and a fair amount of pressure against the center of the IHS to keep it from sliding around too much while clamping it down; it's going to want to scoot around a little bit) in the socket before it cures and using the mounting pressure of the closed socket to ensure the IHS is sitting flat on the die as the RVT cures overnight. That seems to be enough to keep it together if gently handled when out of the socket after the RVT drops cure. I thought about insulating the pads/components under the IHS with something due to the CLU being conductive but was worried about throwing something out of electrical tolerance (very easy to do with high frequency circuits) so I'm just very careful applying it with a brush (you don't need much, just enough for a shiny layer covering the die) and repositioning the IHS. Haven't had any problems.

All that being said, I wouldn't recommend delidding and voiding your warranty unless you're really going to push your chip hard and need the extra temperature headroom.

I can't speak for the Intel delidders, but my strategy on Kaveri was to remove the epoxy from the underside of the IHS, but to leave the epoxy on the PCB. That "lowered" the IHS a bit, but it still gave me a guide to follow when replacing the IHS prior to installing the HSF. The epoxy remnant was thick enough that it held on to the IHS, preventing it from moving anywhere, and installing the IHS produced a mild rebonding effect so that the IHS was partly glued on there. It was easy to pull it off by hand, but it did require minimal effort.

Didn't know AMD was pulling this too, geeze.