P370EM heatsink fits P370SM

I wanted to run a stable 4 GHz OC on my 4900MQ for gaming without an atmospheric re-entry event. The P370SM CPU heatsink only has 2 heatpipes, as opposed to the triple pipe used in P370EM (which also uses the slave GPU fan for cooling as well). I figured the triple pipe + slave fan sharing would allow the CPU to run much cooler, so ordered a set of P370EM heatsinks for both the CPU and the slave 680M and gave it a whirl.

Before swapping out the heatsinks I made careful measurements with a digital caliper, and I can say with confidence that the dimensions of the P370EM and P370SM CPU heatsinks are identical to within 5%. Obviously the 370EM has the third heatpipe whereas the 370SM doesn't. Then used some pressure indicating sensor film to see how well the heatsink would contact the CPU die. For some reason the 370EM heatsink I had made good peripheral contact but was lacking sufficient pressure along the center of the die. Not an issue though since TIM is applied along the center anyway.

On the GPU side, aside from the fact that the 680M heatsink comes in 2 pieces vs the 1 piece construction for 780M, it is virtually identical to the 780M minus the fact that there is a washer/bracket thing at positions 3 and 4 on the GPU core heatsink (see attached pictures) Once again, dimensions as measured by the caliper showed the heatsinks were virtually identical, and the washer/bracket did not seem to interfere when installing the heatsink.

Once the swap was complete, I fired up Crysis 2 on max settings and gamed for 2 straight hours (still on stock clocks since I wanted to test the waters before going nuts). The results were somewhat of a mixed bag: CPU was 8-10 C cooler (peaked at about 67 C), but the slave GPU now consistently ran 10 C higher than the master GPU (peaked at 87 C). I'm not too surprised, considering the I'm technically using the wrong heatsink for a hot card, and part of the GPU cooling power is lost to the CPU. I was also pretty sloppy when repasting the slave GPU, so that would've boosted the temps as well.

However, the benefit to the CPU was noticeable even at idle, since I can now reach a chilly idle temp of 33 C on the 4900MQ. I'll try repasting and see if the temps improve, but if I can get the slave GPU to within 5 C of the master GPU, I may just opt to keep this setup. That being said, I'm still uncertain what the long term effects would be if minute differences resulted in uneven/excessive pressure on the CPU/GPU die.

EDIT: Pics!

 

The dimensions are the same yes, you could try using a thin copper plate to increase pressure.

 

^yes if I do decide to keep this setup I may look into getting some copper shims, although I think the slave 780M will benefit far more than the CPU, see below.

Part 2: After checking the die contact carefully with pressure sensing film, I think I may have an explanation for the high temps seen on the slave GPU. I present to you the following pics:



The more red you see, the higher the contact pressure ie the better the contact between the heatsink and the die. In the very first pic on the left is the pressure map of the P370SM heatsink for Haswell CPUs. As you can see the contact is pretty good, with significant gaps only present at the corners. The 2nd pic is the pressure map for the P370EM heatsink. Keep in mind that this was designed for Ivy Bridge CPUs (even though the external dimensions are the same). The situation is basically a reversal of the Haswell heatsink -- peripheral contact is good, but contact along the center of the die is sorely lacking. However since I apply the TIM as a line along the center of the die, this is not a sufficiently serious problem that concerns me.

For the slave GPU however it's a different story.



On the left is the pressure map with the stock 680M heatsink. Pretty much only the top half of the plate is contacting the die, with the bottom half either not having any contact, or not having sufficient pressure to register on the film (the detection limit is 28 PSI I believe). So I did two things:

1. Sheared off the washer/brackets, and



bent the heatpipes ever so slightly so that the plate now sits squarely on top of the die instead of "floating" above it with the completely stock heatsink.

After the 2 quick and easy mods, the pressure map now looks like this:


Contact between the plate and the bottom half of the die still isn't great, but at least there's sufficient pressure along the perimeter.

After screwing everything back together, I fired up Crysis 2 and gamed on the last level (A Walk in the Park) for 30 minutes, these are the results:


The temps on the CPU remain amazing -- only one core registered above 70C, but the CPU package topped out at 68 C. Disappointingly, the slave GPU still runs hot, peaking at 89C and a good 9C hotter than the master GPU. At this point I'm honestly not sure if this is due to bad die contact, or thermal loading of the slave GPU heatsink/fan due to sharing the cooling with the Haswell chip. I'm guessing it's probably a bit of both.

Bottom line: with the P370EM heatsink, the CPU runs 10C cooler thanks to the extra heatpipe and shared cooling with the slave GPU fan. However, this comes at the expense of the slave GPU, which now runs about 10C hotter than the master GPU. In a way, it's almost as if the CPU completely unloaded its extra heat onto the slave GPU. Whether this is worth it in the long term remains to be seen, but considering a 780M costs more than the 4900MQ, for me the answer is probably not, unless I can solve the thermal issue with some copper shims and/or better paste (right now using MX-4). Not to mention that OCing the 4900MQ will only lead to even worse thermals on the slave GPU.

Btw should probably mention these temps were obtained with an ambient temp of 18C (64F). So during the summer months I expect the temps to crawl by at least a couple degrees.

 

Pressure may be down to needing to tweak the thermal pads too without breaking out the toolbox on the cpu. As for the gpu that is interesting, there may be some variance in production.

 

I"m happy with the results on the CPU, so I think I'll leave it alone for now (though I'm sure tweaking the heatsink will bring even more benefits). As for the slave GPU, in all fairness I am using the wrong heatsink, even though external dimensions appear to be the same, if the die/plate thickness is off by even 0.5mm that could throw off the surface contacts.

 

Would you by chance know the thermal pad size(s)? I'm looking to repaste my laptop and figured I would go ahead and replace the thermal pads...I just am unsure of the needed thickness is all.

Thanks Ahead of Time.

 

1.5mm (little vrm chips), 1mm inductors and 0.5mm vram.

 

If you mean the strip of thermal pad on the CPU heatsink it's 1.5mm.

 

Is it straightforward when removing and cleaning both the GPU and the CPU heat sinks, or would it be better to find a diagram to follow?

 

If it helps here's a repository for Clevo service manuals for basically every model that Clevo has released.

 

It's a fairly simple operation, just keep a pot for each type of screw and make a note on paper under each one to say where they go.