Better Cooling for the 7800GTX - Page 14

Pics of my Modding Project

I've opened up my "shimmed up" 6800U to take out the 4 (sorry, it's 4 shims and not 5) shims that I had put in it 6 months ago. Yes, the cards been used ever since without any problems.

Below I'll take you through my experience in removing the AS5ed shims from the 6800U.

Here's what you'll need (ArctiClean 1 and 2 is not need, if you have another solvent). T8 Torque screw drivers and an star screw driver (unless the card is already out of the system).



Before I started unscrewing the 4 spring-loadeded screws (named A, B, C and D), I used a permanent marker to mark the screw and a corresponding mark on the metal die.
This way I could measure how many turns it takes to unscrew all 4 completely. I measured exactly 6 complete revolutions for each screw (to loosen completely, of cause, making sure I unscrewed in order from A to D each time).

You will need these marks when you insert the shims and tighten the screws as they will guide you to tighten all 4 screws so that they exert an equal pressure on the shims and thus the shims sit complete flat on the chips.

Eventually, I'll be reusing the 2 copper shims on top of the memory chips in the 7800GTX.

----------------------
Here is the pic with 2 copper shims (with blotches of AS5 where they once contacted the memory chip) still stuck with AS5 on the metal core (after unscrewing and removing the green part).

-----------------------
The Green part below shows the memory chips with corresponding AS5 blotches.


Now we see the underside to the 2 copper shims after I pulled them off the metal die.

When pulled off, notice that the AS5 is seen as blotches (as I used the "1/2 rice grain in center and squooch the shim on" technique instead of the "spread a thin layer" technique).



Now you see why my technique did not allow for the AS5 to spread all over to make a complete contact. So, I'll used the "spread a thin layer" technique when I re-apply the shims to the 7800GTX card.


Then, I removed the aluminum backing which still shows the two Sterling silver shims. Again, note that the AS5 was applied only as a grain in the center and then allowed to spread outwards (although in some cases not so evenly covering all the chip's surface).


-----------------------
Closeup of the backing with the silver shims still stuck.


-------------
Now, I pulled the silver shims off the backing


Note below is a top view of the setup of the 2 COPPER SHIMS (after I cleaned the old AS5 up with ArctiClean 1 solution).


Another view shows the how the Green part will eventually close like a book onto the other part with the copper shims (so that the memory sits exactly over the copper shims).



You must be wondering how thick the 2 copper shims must be so that the Nvidia chip sits on the copper (exactly) so that it does not break the chip or leave a gap.
Well, I had to use a strip of paper to measure the exact thickness of the copper shims. It took lots of patience to find the corrrect thickness for the 2 coppers shims (dimensions below).



I've replaced the 2 sterling silver shims with two slightly longer .999 silver shims (not that it will make an incredible difference, but I had the .999 silver handy and they're longer too). Yes, and that's just about as fancy as you can get....
Note the 2 .999 Silver shims are show sitting on the other side of the green part and after AS5ing (both sides of the shims) the black cover will fold over like a book.




And finally here are the dimensions of the 2 Silver shims. NOTE: You can use copper ones (just fine). I was just to lazy to find some thinner copper sheets to cut and used what I had.



I've tried to attach all the pics in the order that I found them (and i'm not sure how this will work, so bear with me). Let me know if you have any questions. I'm going to get some sleep now. Will install them in the 7800 tomorrow and test.

ahh clarity is wonderful, thanks a lot for the time and effort, great post

Ok, if you could tell us your idle temps, that'd be fine, thank you ^^.

Oh yeah, anyone beein looking at this? Spire pacificbreeze laptop cooler? Simple, looks cool, lets your laptop stand better. Ill simply go with that ^^.Spire pacifibreeze Laptop cooler thread

Yes, I did..

Help required

[quote=Mr. K6]Wow, awesomework nkhan! I like the copper block design. It think that when I get home for vacation I might try a hybrid model of my fin design and the copper design for a complete cooling solution on the 7800GTX. Excellent work on both the card and the cooler, I'm looking forward to your results on the 7800GTX

K6, I am at the part where I must decide to either rip off the thin aluminum foil (from dell) off the copper plate or leave it on. Should i just clean it and AS5 directly over it or rip it out and AS directly on the copper heatsink? Anyones got a suggestion. Physics of conductivity is not my high and maybe someone has a good reason not to/or to rip the aluminum off.

Thanks in advance

help applying AS5

Hi,

I see you're still online. Can you tell me if this looks about right way to apply the AS5. Should I put an extra rice grain in the center of the Nvidia chip (before sandwiching) or is that fine. I just spread it with a card.

Holy massive glare, Batman!

k6 alot of 6s there

Still haven't gotten to going out and getting the metal yet

Nkhan, Wouldn't mind you posting them temps though

I love the idea of removing those thermal pads on the underside, but I'm still under warrenty and am to chicken to take them off. Same goes for the aluminum sink underneath

Once I get that copper I'm doing the fin design from Mr. K6 and using the topside ram to heatsink idea from Nkhan.

Nkhan and Mr. K6 keep it up

OK, that's what I thought, but this comment is what threw me: But you weren't talking same clockspeeds when you posted that.

First Test after installing the shims in the 7800

First of all, I like to warn people not to jump to get the copper pieces cut using my current dimensions (until I've completed the project). Those dimensions were good for the 6800 when I had them running, but since I'm going to go the extra length to ascertain the exact dimensions required(by re-opening the video card many times if need be) for the best contact (with the Chips), please bear with me (and hold those orders for now).


So first the temps.

When I put the modded 7800GTX in the Gen2:

On Idle (with same conditions; the cooler running on high, and fangui on gaming) I got 32/49. I know it takes about 200 hours for the temps to decrease (once the AS5 settles in properly), but 49 for the GPU is too hight considering that I was getting 45-47 before modding. Anyway, I did repeated the 3Mark05 test that I had done before modding:

The test was to run 3DMark05 at 1920X1200 and 4 x AA for 100 cycles (about 20 minutes) but selecting only the PIXEL SHADER TEST. (This quite a STRESS TEST and can be used to check stablility of overclocks too).

Before modding: No artifacts, and MAX temps were 62/104 (i.e. CPU/GPU)

After Modding: No artifacts, and MAX at 55/107

Anyway, I did another test with 3DMark06 (full version, default settings):

Before modding: 62/88 (and 88 for GPU is the MAX ever got out of the 4-5 3DM06 tests).

After modding: 61/105

Assuming the 107 for the GPU might eventually come down to 104, the temps are still to high after modding.

Also, after 3DM06 (which is not no STRESSFUL as the 3DM05 PIXEL test for 20 min at highest settings), if heat was dissipating from the CORE well, those the temps should have come down (but they stayed at 105)

So what's going on... No way of knowing until you get in under the hood.

Before we go on to see pics of the disasembled GPU (to verify areas of contact), I want to elaborate on the technique I used:

Before modding the GPU, I place the (in original factory assembly) GPU on a piece of paper and traced around it. Then I measured the distance (in inches) at the 4 torque screw points and 3 other areas around it



(note: before means before I unscrewed the original factory set card and after means after I re-assembled with the shims)


Recall, that when I first unscrewed the 4 torque screws, it took exactly 6 turns on all 4 (A, B, C and D screw) to unlock them(i.e. lift all four corners of the die up).
That means that once I re-assemble GPU with the shims (if the shims are the right thickness), after 6 turns (to tighten) each of the 4 spring loaded screws, I should get the corresponding measurements (which I did get from looking at the before/after measurements in the above pic).

Anyway, if you've been following my previous post:
.... after assembly, I realized the two .999 silver shims were too thick (and thus caused the black backing to remain lifted on the edges). They were 0.050" and the older sterling silver shims were .040" (which is still a bit too thick and should have been 0.035", but that did'nt matter so much).

So I removed the 2 silver shims and just used the 2 copper ones. (i.e. I left the thin dell pads attached to the backing (and only removed the 2 thicker heat pads to make room for the two copper shims) Picture coming below, so don't worry...

Tour of the dis-assembly:

Here's the opened GPU showing the side with the 2 copper shims.



Notice that even after FULLY SCREWING (tightening 6 turns each) and getting the right measurements after assembly (see traced diagram above), the nvidia chip (center) did not completely touch dells copper piece (center); the AS5 on the CPU did not touch all parts of the copper. The minimal gap is due to removing Dells thermal Sticker with the black paste. I'm afraid I will have to find a very thin aluminum sheet to add on the dells copper because the spring loaded screws can't be screwed any further (to close the gap). In other words, that mold was made with the gap and even if sand down the 2 copper heat sinks (it won't make any difference). Infact, if I sand them down too much, I'll loose copper as it will become loose and not fit snuggly anymore.


TO TRY: I will try to reassemble the GPU without the 2 copper shims and see if the gap is closed down or not (i.e. see if I can screw beyond 6X on each screw). This way I will find out if the copper is obstructing the further tightening of the gap (between the nvidia chip and copper heatsink) or the mold itself.



As for the other 2 copper (see the AS5 on my added copper) shims they were smeared with quite a bit of AS5 and marks on the copper showed that the memory chips did sit flat on the copper shims (as I spread AS5 all over the memory evenly as possible) but also showed evidence of air pockets (particularly in the center, where most heat is emitted).

If we go back a little: In my first modding (using these same copper shims) on the 6800U card, when I pulled the shims off the modded card (after 6 months), we saw solid round blotches (as you see below) and not gap-ridden smears(as in the pic above)



My feeling is that since the heat emission from the memory chips is probably highest at the centers of the chips, it's probably best to apply a 1/2 grain of AS5 in the center and then sandwitch the thing up. This way the air bubbles will be minimized (as AS5 will be pushed out from the center) and ensure better heat transfer. The "spread with your card" technique (that i used in the 7800 mod) either left too much AS5 (which is not good) or too little (no good either), but does little to ensure minimal air bubbles (evident from gaps). I may be wrong in my thinking. K6, what do you say?

Here is the other pics of the shims futher removed.



The other side without any shims (for now atleast):



NOTE: After I removed the the 2 .999 silver obstructing shims, I did not remove the AS5 and have gotten AS5 on the pads (yes, I botched up there! and it's undoable... until I replace the pads with a thinner (0.035" - 0.040") shim)


QUESTION: I don't know if the AS5 impeded the heat transfer cause now the heat must transfer through the AS5 and the pad (2 mediums)?

TO TRY:
Since I've already got the right (thickness) sterling silver shim, i'll used them in the next experiment (for now). I'll sand the .999 silver shims down a bit more later today


Sorry for going on and on with this winding experiment. I know for those who are still using 56K Modem connections, it's very taxing to download so many big pics.

I'll take a break to think out my next test. Any ideas?

Hehe, you must really love modding, eh? Seems very elaborate and tedious. Now you have to find the exact right thickness for the center piece. I'd be interested in your end results, seeing how low your temps were considering almost none of the HS was touching (I had roughly 2/3 touching with as5, got +1°C from stock, completely touching -3-4°C from stock).

My Idea: You could look for the 2mm screws (the torx screws), and exchange them. That's what i did, seems to have worked like a charm (i broke my torxs ^^). NOTE: Call and ask if they have 2mm screws with that winding- very hard to find >-<.

€dit: What's your guys' call: Should I rather exchange the other two torx screws as well(i got enough screws laying around :P), or buy a shim for the center? I'm guessing having the extra copper piece in the center would slow down the heat-conduction to the HS? Then again, it can hold extra heat >.< Thx,

Franz

[quote=Mr. K6]nkhan, excellent documentation, I must remember to take more pictures when I do my modding. For reference, go to page one and look at my original design. I think that if you construct a hybrid of your design and mine, you should have an incredible heatsink. First, those silver shims are much too thick. I used .025" copper in my design, and even that I feel was too think. I think that 0.0135-0.02" is the best range for the underside of the heatsink. Also, try constructing a fin design as I did so that the heat is taken out from underneath the card. The silver blocks are great, but if they dont remove the heat from the card they do little than just make the card heat up slower. Also, try adding a square piece of copper or silver to the top of the core to give better contact and pressure between the core and the heatsink. While it might seem counter intuitive to place another piece of metal between the heatsink and the core, it actually works in this place. First, it allows a very thin layer of AS5 to be used and second, it allows the core to be more highly compressed and transfer heat better. But the shims on the top side of the card look good, definitely keep them .

K6, thanks for the encouragement. I'm not in any hurry to get the computer up and running so i'm onto the copper search. Found a thin flat piece of .999 silver (0.035") that is exactly the same size as the copper heatsink (center). I've not tried it, but do you think it's too thick. I tried to assemble the unit without the two copper shims and with a piece of putty btw the heatsink and core. Then I tightened all 4 torque screws and after reopening, I measured the thickness of the putty to be about 0.025" and 0.030(to estimate the gap left). Do you have a measure of the gap after removing Dells FOIL from the Copper?

Anyway, see how nicely the .999 silver shim sits on the heatsink. I can always sand it down to the right thickness (but I want to make sure lest I screw up).

Sanding away my hand!

K6 and the rest: Just an update. I got too impatient with the sanding and installed the 2 copper shims with a 0.030 .999 silver shim in center (too thick). Tried to screw it but the shims on the side did not sit on the memory. So just for the heck of it I add another .04 shim (too thick) to on top of each of the copper shims (on the sides) and tested. Video card got thick and heavy that I could see a bulge on the keyboard). I had my doubts that the core was actually touching the silver (very well) and I was right!. 32/58 was the idle. Anyway, I ran the 3DMark Stress test (100X Pixel shader at full settings) and got 55/109. Then I ran 3DM06 on full settings (once) and it crapped (couldn't see the max temps, what's the MAX temps the bios allows before shutting down the system?) out and reset. When I opened the card up I found the core untouched except in a wee corner of it. Hope I didn't burn anything .


Anyway, will remove the extra 2 shims (sitting on top of coppers) and sand the center silver down to a foily 0.020-0.015 (as you suggested. Thanks! ). I guess once you AS5 that foil on, it's very difficult to remove without damaging it (so one should do the AS5ing right the first time)?.
(after sanding the silver) If the side RAMS (Still) don't sit on the memory (cause they're too thin), I guess I'll have to put back dell's blue Maxis on the mems untill I get 2 thicker copper shims) Be a good test to see how much the core heatsink modding really plays a role in the cooling). Do you think it would help to stick dells' pads on top of the copper (after AS5ing the side that touches the copper only), to fill the gap?


Haven't gotten to the overclocking part yet. You'll have to guide me to see (once I succeed in the hybrid modding) to overclock to above 460/1220.

Thanks, K6