Why is there a gap between the GPU and the heat sink?

Lately, I have been wondering about the following:

Instead of using a thermal pad why don't manufacturers design their heatsinks in such a why that there is no gap between the heat sink and the GPU core?

There are several posts over the net about nVidia's GO6150, GO7150, 8400, and the 8600 chipsets. Lot of people think that these chipsets failed because heat sinks were not design properly. I'm not sure that this was the case as several manufacturer (HP, Dell...) had a gap between the heat sink and the GPU core. I just find it hard to believe that HP for instance made a design/ engineering mistake when they designed the heat sink for the Pavilion dv2000, dv6000, and dv9000 series. They had all the necessary data sheets from nVidia so if they put a thermal pad that means that according to their calculation the pad was enough to transmit the heat from the GPU core to the heat sink.

 

could've been a production or design mistake, much cheaper to add thermal pads in there than to re manufacture the correct copper heatsink I guess

without the general public knowing the difference

 

It's not only the IGPs either - HP uses thermal pads on actual discrete GPUs, even though the clearance is small enough to use thermal paste. My 8530p had a thermal pad, and a 6830s (also business series) I opened up also had a thermal pad on the discrete Radeon 3450 - burnt to a crisp because the owner overclocked it to play games...

That's why I always recommend buying some good TIM and repasting the laptop even if it's new...

 

Because they can. It is a lot easier assembling a laptop with a thermal pad than thermal paste. It is also "good enough."

 

This what I don't get. How can they make a mistake when they know the size of the chip?
I mean here we are talking about gaps as wide as 2mm. In my opinion that's too wide to be a mistake

I have to disagree with you and I'm talking from experience, there is no additional effort when assembling or disassembling a laptop with or without a thermal pad . The effort is the same and the time required to do the job is about the same.

 

Is squeezing out and spreading a dab of thermal paste correctly as easy as simply placing a pre-made pad on the chip? No. Also you can't mess up with the thermal pad, while with the thermal paste you can.

 

Agreed, slapping a thermal pad on is easier than pasting. My HP 8460P had a thermal pad on the GPU and I replaced it with paste. It ran cooler with paste, so the obviously the gap was pretty small.

 

The "about the same" isn't same when factory pushes out 100k+ laptops every month. I don't have any exact figures but HP alone sold about 16 million computers in 2010 and that makes 43k computers / day. Of course that includes desktops too.

 

They are already doing it for the CPU, so they can do it for the GPU too. Also, They do not use a hard paste like AS5, MX4 and so on. I think they use a liquid paste that hardens over time. So the only thing they do is to drop a couple of drops on top of the CPU and another few drops on top of the GPU. My theory regarding the liquid paste that hardens explains why the paste looks like this in the below photo.



Qing Dao, Mr.Mischief, and KLF, I think you guys are drifting away from my main question. There's no doubt that HP and the other manufacturers have the right people and the right procedures to ensure that they don't make such insane mistake as to leave an unintentional gap between the GPU and the heat sink. If the gap is there then there must be a reason for it. I mean it simply not logical to make such a mistake considering the cost for HP, Dell, and the others to fix it afterwards. I mean a lot of people had there motherboards replaced for free under the warranty and that must have cost a lot of money to Dell, HP,...

I've read that they might have opted for the thermal pad because it is a material that can expend and contract. This ensures that the GPU doesn't get stressed by the heat sink. What are you thoughts on this?

 

from whatever angle I look at it, I still think it's plain stupid to use thermal pads nowadays.

 

There are three advantages to thermal pads:

1) It is faster and easier to apply than thermal paste. When you are manufacturing 16,000,000 systems per year like HP did in 2010, reducing the manufacturing error rate by even 0.1% means 16,000 fewer defective systems.

2) It allows you to manufacture a cooling system (heatpipe) in a single flat plane with higher tolerance for manufacturing variance.. Direct-contact thermal paste would mean that you need a precision-manufactured heatpipe that contacts three separate chips (CPU, GPU, motherboard chipset) that sit at three different heights. If any one of the three contact points on the heatpipe is the incorrect height due to simple manufacturing variance, then the entire heatpipe is useless. Thermal pads avoid this costly problem.

3) Thermal pads are "good enough" for cooling. As an end-user enthusiast, it may make all the difference to you in the world whether you can overclock your GPU by 5% vs 8%. Or whether your CPU idles at 31C vs 35C. To a system manufacturer like HP or Dell, they don't care... their only obligation is to ship you a system that works at factory settings, as inexpensively as possible.

The problem with those nVidia chips is that some of the manufacturing materials were improperly selected (i.e. the underfill, a type of glue in the silicon package manufacturing process). This means that the GPUs would physically move and shift if the temperature was high enough, causing failure in the chip soldering joints. The end story is that the nVidia GPUs did not perform within their documented specifications.


Yes, this problem may not have been as bad if the cooling systems of the affected laptops were over-engineered, and used higher-performing thermal paste design instead of thermal pads. But it is not a failure of the laptop cooling system design. Hypothetically, if the problem was caused by laptop cooling system design, then the error would only occur with certain models in certain manufacturers. The fact that it happened in every laptop manufacturer that used a certain nVidia GPU is a clue that it is an nVidia problem, not an HP / Dell / Apple problem. This is definitely a failure on nVidia's part due to manufacturing defect, which is why they paid out settlements to HP, Dell, and Apple.

The gap is there because it allows you to manufacture a cooling system (heatpipe) in a single flat plane with higher tolerance for manufacturing variance.

Direct-contact thermal paste would mean that you need a precision-manufactured heatpipe that contacts three separate chips (CPU, GPU, motherboard chipset) that sit at three different heights. If any one of the three contact points on the heatpipe is the incorrect height due to simple manufacturing variance, then the entire heatpipe is useless. Thermal pads avoid this costly engineering & manufacturing challenge.

This is why you only see thermal paste being used in single-chip contact coolers (CPU cooler, GPU cooler). Multi-chip contact coolers tend to use thermal pads, for this reason.

 

Aren't Thermal Pads more forgiving for Shock Resistance too?

 

The gap is there for the GPU to relieve stress. Right.

 

No. But i know thermal paste, once it hardens, becomes brittle. A good jolt could put a crack between the CPU/GPU and the Heatsink. A Thermal Pad is a little more forgiving.

 

Thermal paste shouldn't harden, only the cheapest stuff (that OEM's use) will do that. A thermal pad is only a good option for a gap, although not a great solution because it's not as thermally conductive as a very thin layer of paste. Many times you'll see pads on a heatsink to GPU vRAM only because the focus is on a tight gap between heatsink and GPU, where the heatsink might extend outward to also cool the vRAM. But vRAM doesn't typically heat up as drastically as the GPU and is less sensitive to heat.

 

that's a problem with the choice of thermal paste. I never use thermal paste that hardens with time, nor I use electrically conductive paste. Never had single problem, even though my paste may not be from the very best thermal conductive compounds out there. But the way I see it - tiny compromise for the benefit of the more important stuff.

 

I'd say it's mainly due to easier assembly and as mentioned above the possibility that they're more shock tolerant.

Heatsinks don't have to be precision manufacturerd to use paste on more than one chip. My heatsink does exactly that. If the heatsink were rigid like a desktop CPU cooler then tolerances do come into play, but heatpipes used in laptops are flexible enough that a 1mm tolerance is acceptable given that each chip usually has its own retaining screws to put pressure directly on it. I can provide evidence for this too, given that I accidently bent one of the heatpipes on my heatsink and it still worked more or less the same as before.

 

That was sarcasm, sir. LOL.

 

Sorry.. my sarcasm meter was broken yesterday.. Seems they used cheap thermal paste on it and it developed a fracture. Terrible thing really, having an overheated sarcasm meter..

But.. they put a thermal pad on it and im all up and running at normal. Snark Away!

 

kent
While I still disagree with you regarding the time that could be saved when using the thermal pad over thermal paste, I completely agree with you when you say that they had to have a gap because they wanted to save money by making the heat sink from one flat sheet of material.

Think about it, it takes 10 seconds for a human to remove the thermal pad from it's plastic sheet and put it on top of the GPU core, it also take 10 seconds to squeeze a couple of thermal pastes drops on top of the GPU core. The paste is probably preheated so it is easy to apply. So there is not any time saving either way. A machine in a conveyor belt will do the above process in about a second.

And by the way, thanks for confirming that thermal pad are adequate to transfer the heart from the GPU core to the heat sink.

 

I bet fish paste performs better than thermal paste!!!

 

Who knows!!!! There is a link in a post here in the forum regarding thermal past performance. The link shows the performance of several thermal pastes. Beside the usual brands they tested with butter and chocolate. The performance with chocolate was bad but the performance with butter was rather good.

Edit:
This is the link for the performance of butter. It's near the end of the table

 

Fresh toothpaste good for stop-gap cooling while you purchase some real paste LOL.

 

Yeah, one site actually used it in their comparison of pastes, pretty funny. Will work for a very short time because it will harden quickly and not become effective, plus temps will be higher, but still sufficient, as a stop-gap, as you say.

 

Assuming you are correct, but you need to consider the time for warranties/repair/maintenance.

Repaste every time after lift the heatsink up takes time. (No re paste after lifting heatsink up = cooking chips)
Thermal pad = snap it back in = done.

 

Regardless of what everyone thinks the OEM's should or could do, the main issue is mechanical stackup tolerances. I have done a number of designs and it is not possible to mass-manufacture heatsinks with sufficiently low tolerances that paste would ALWAYS work. We had to laser measure each casting and board stackup (3 CCA's) to ensure that we would not need a gap pad on a design where the CPU was interfaced directly to the case (fanless). We had to be certain it would make contact and at the same time not overstress the CPU die or underlying CCA's.

It is not really feasible to put a gap pad on a CPU (and some GPU's). The power doesn't get dissipated quick enough. Most designs are spring loaded such that the designed can control the mechanical loading on the CPU/GPU die (99% are flip-chip). Most CPU/GPU's have a pre-cut piece of material that is really a form of thick paste, not technically a gap pad, which start at about 0.5mm.....

- Tim

 

Water cooled laptop. Makes you feel cool on a hot sunny day with the refreshing sound of water going back and forth inside your laptop while carrying it

 

Ohh.. LN2!!

Actually, there is fluid in the heat pipes of most coolers.

 

I'd prefer butter. Smells nice as you cook the heatsink.

 

I`ve also heard about people smearing butter on all the components inside the laptop. According to them it works as a lubricant, making the bits and bytes go much faster between the components. Reducing the friction is very important.

It makes perfect sense

 

After smearing a layer of butter, spread a layer of cooking oil. Hear the sizzle!

 

It doesn't take time since 99.99% of the people that buy computers don't know how to upgrade them and thus will never upgrade them

Also, considering the number of motherboards that were replaced under warranty because of failing GPUs, using thermal paste is likely to have been more cost effective as it would have prevented the GPUs from failing

Very instructive...thanks

 

In particular when fried at about 70 to 80c.

 

Adding a 4" in diameter muffler to the cooling exhaust will also make it go faster.