Can we get a masssive Delidded thread going again?

[chris89]

I'm very interested in delidding vga cards among all the generations gtx 400, 500, 600, 700, 900 and beyond.

Right now my 480's IHS fell off and now I can't get decent temps using the stock cooler. I even used high dollar 17w/m K fujipoly between the die, ihs, heatsink but I think pressure is at fault. What's the thread specs of the heatsink if anyone knows? I wanna get some nice super duty screws that are Hex head and much larger so they don't easily strip/ reducing pressure.

I bought a 0.6mm thick 2cm x 2cm Shim, and potentially a 1.5mm thick 2cm x 2 cm Shim as well. I think the thickness is 1mm or so?

 

[EliteRetard]

You're gonna need some calipers:

http://www.amazon.com/Inch-Digital-...5?s=industrial&ie=UTF8&qid=1427133713&sr=1-55

Measure the IHS and find a shim to replace that thickness (or slightly thicker).

You need to be very careful about applying extra pressure, you can easily damage the die. Best to stick with the stock solution and just use a shim to replace the IHS...with new thermal paste your temps should be as good as new. The only hiccup might be if there were a slight gap under the IHS or paste was thicker to begin with...which is why I suggested slightly thicker shim (like .005 inch). Best to use a single shim, and not stack multiple thin sheets.

 

[Headfoot]

You could also consider using the G10 bracket and an AIO cooler. You'd have to be careful not to over tighten as the bracket doesn't have spring tensioners to stop you

 

[Red Hawk]

Wasn't it Idontcare who did a lot of the delidding? He's sadly not active on the forum anymore, from what I can tell.

 

[chris89]

Thanks guys I bought an assorted of shims... I guess they say the die is 529mm *2 which is converted to 0.8" basically squared so I picked up 20mm which is 0.787" Squared. I can't seem to pull off the pressure to get the temps in order with the stock screws...

Another thing is the IHS is exactly 2mm thick correct? It has the indention that reduces die to ihs contact to exactly 1mm. So I suppose the 1.2mm thickness may be ideal to insure the ideal proper pressure contact required?

2 of Copper shim 20x20x0.3 mm
2 of Copper shim 20x20x0.5 mm
2 of Copper shim 20x20x0.8 mm
2 of Copper shim 20x20x1 mm
2 of Copper shim 20x20x1.2 mm​

 

[chris89]

*Update*
Got the shims in and used the thickest 1.2mm i believe and the temps are still getting way out of hand in a hurry. I have a 17w/m K 0.5mm pad on the die, the shim (which is about 3mm smaller squared than the die) and another 17w/m K 0.5mm pad on top of the shim sitting tightly on the heatsink base.

I'm not sure what to do actually because it's fine in 2d like right now but in game it shoots up over 95C in less than 30 seconds.

Is it that the die dislike the pad and needs paste? I'm guessing I have poor contact, I'll pull it back apart and take a look.

Maybe i'll need to pickup some new screws to tighten it down further... maybe the thread length isn't allowing a very tight connection...

The exact dimensions of the die is 529mm ^ squared which is just about 0.81 inches converted to millimeter is 20.574mm but the Shim must be incorrect because it is slightly smaller than the die. It sits within the die with about a millimeter or so of die showing. Maybe I need 25mm shims to be safe... at 2mm possibly?

529mm^squared is 0.8199516399 square inch if I'm not mistaken that's 0.82 inches length/ width correct?

Any ideas?

 

[lavaheadache]

I imagine those pads aren't the best solution. Why aren't you using viscous tim?

 

[MongGrel]

Thanks guys I bought an assorted of shims... I guess they say the die is 529mm *2 which is converted to 0.8" basically squared so I picked up 20mm which is 0.787" Squared. I can't seem to pull off the pressure to get the temps in order with the stock screws...

Another thing is the IHS is exactly 2mm thick correct? It has the indention that reduces die to ihs contact to exactly 1mm. So I suppose the 1.2mm thickness may be ideal to insure the ideal proper pressure contact required?

2 of Copper shim 20x20x0.3 mm
2 of Copper shim 20x20x0.5 mm
2 of Copper shim 20x20x0.8 mm
2 of Copper shim 20x20x1 mm
2 of Copper shim 20x20x1.2 mm​

If I were shimming something like that I'd use G10 I imagine.

http://www.eplastics.com/G10-FR4-glass-epoxy-sheet

I'm going to stay out of this one in the future I imagine.

 

[chris89]

@lavaheadache
@MongGrel

Thanks guys. What would you guys do and can you explain the epoxy?

Glass epoxy sheet in replacement of the IHS and copper Shim? Wouldn't the glass expoxy require a high thermal conductivity rating?

I stopped using paste because of it's inconsistent low performance and disposable/ non-reusable nature. That and the performance of paste is far less than pads. The pad ensures perfect contact on every application because it's 0.5mm thick and compresses reducing thermal resistance.

Anyway the best paste normal doesn't exceed 10w/m K and that's unlikely. Usually between 5-6w/m K is the typical "high end" paste performance. Paste also performs worse as temperatures rise as it begins to lose it's properties. The higher dollars pads never lose their performance and actually perform better as they get hotter.

Anyway I did just buy some super thin 0.254mm thick graphite thermal material that has a rating of 240w/m K. Pads/ material are the best because you can remove a heatsink 100 times and never need to use another application. That's as long as you don't over-tighten and ruin the pad. Not only that but pads never fail like paste does, apply once and forget it as long as you did it perfectly that first time.

Running 0.5mm thick gap pad 5000 Bergquist pads on a GTX 580 on all ram, voltage regulators, and core allowed me to achieve clocks not known to be possible. I can pull off 900Mhz stable all day on Bergquist 5w/m K 0.5mm pad material. Bergquist gap pad 5000 is pretty fragile though and it's a 50/50 chance of pad corruption on every pull-apart.

I did just buy 25mm squared shims some 1.5 and 2mm thick. 25mm is 0.98 inches and I think will cover the die better.

Have you guy's ever delidded? I hear stories of success but thermals are hard to handle without the IHS and stock cooler...

Thanks

https://www.youtube.com/watch?v=oIr1Qyonm-I

https://www.youtube.com/watch?v=ZAEhyY1_czM

 

[utk193]

high end TIM like arctic mx4, gc extreme ,IC diamond have thermal conductivity in excess of 8W/mk.
the reason according to me as to why you are getting such high temps are because you are using a thermal pad.
take for example a TIM like mx4. The perfect TIM application shall have very low thickness ~0.003mm or in the worst case scenario say ~0.01mm.
the pads come in a min. thickness of 0.5 mm which in compressed condition go to say ~0.2mm. take thermal cond. of pad= 20 W/mk.

Q(heat transferred /unit area)=k(thermal cond) x Temp. gradient(T)/Length(direction in which heat is conducted)

So for paste Q1= 8xT/0.01(taking worst case scenario)
For pad Q2= 20xT/0.2( taking same temp gradient)

Q1/Q2= 8
so (taking a bit bad application of paste and exaggerating pad thermal conductivity) heat transferred by TIM is eight times than the pad.

I would advise you to ditch the pad completely and either use shims with least amount of TIM as possible to cover the die if there is big height difference between heatsink and die or just the paste if the height is low(~0.01mm or less).

As you are delidding , you could have better luck with CLU ( goes as a very thin layer)

 

[EliteRetard]

I still think you need to measure the IHS that was removed. Trying to guess what the thickness is, is just going to be a hassle or cause problems. Why do you think it's 2mm and not more, like 4mm or something?

You want the least amount of material between the die and HS/F without going less than the IHS that was removed. If you go to thin you'll run into contact issues (as it sounds like you are).

At idle the card down clocks and lowers voltage etc, that's why it's "fine" at 2D.

Hopefully you haven't damaged the screws/threads for the HS/F...they aren't really designed to be reassembled a whole lot...nor cranked down hard. If you have the thickness of your replacement IHS correct you should have all the pressure you need.

How flat is the surface of the HS/F? Since you're using a very flat shim, perhaps it would be worth lapping (sanding smooth) the heatsink surface. Looking again at your heatsink photo, is that the IHS sitting on top? Is the heatsink just the raw heatpipes normally? If so, you cant sand those much (maybe just a bit with 800+ grit)

As mentioned above, it may be worth trying thermal paste vs pads. Suppose you could try your thinner/better pad since you just happen to have it...but you still need to get the thickness of your shim right first.

 

[utk193]

^this too
find out the distance between your gpu and IHS ,where they contact.
Assuming your IHS inside to be flat, find the distance between them.
it should mostly be very less like ~ 0.1mm.could be less or more.depending on that use a paste or pad there.usually the distance there is less so just paste would be better.
the 2mm distance you cited is way too high and you are probably putting too much pressure on the die if you try to tighten the screws too hard. (would harm the screw threads too as stated above and crack the die).also,the actual distance IMO would be lower.
You effectively have ~0.8mm(just a guess) +shim thickness between the IHS and die, this could be much lower if you have paste there.

you could check whether the IHS top and heatsink surface are flat,if not then you could proceed with lapping.
So, to proceed
1.find the height diff.
2.choose the correct TIM.
3.lap if necessary.

Also, ive used the fujipoly 17W/mk pads, they compress very little.CLU is an excellent choice as a TIM. On another delidding thread on this forum ,someone posted the same temps after 2yrs using CLU betweent the IHS and die.(your heatsink or IHS should not have aluminum to use it).

 

[EliteRetard]

*snip*
the 2mm distance you cited is way too high and you are probably putting too much pressure on the die if you try to tighten the screws too hard. (would harm the screw threads too as stated above and crack the die).also,the actual distance IMO would be lower.
*snip*

I don't think you realize how thick an IHS is, it could easily be 2mm or even more. He's been trying various shims starting from .6mm and hasn't gotten good temps with any of them. It seems very likely that the IHS was actually quite thick and he hasn't gotten a shim thick enough to make up that void yet.

IDC has done a few de-lid projects here, and he did this with a GTX460. I can't find the exact dimensions of the IHS at the moment but it was like 2-3mm thick. It would take all of 10 seconds to measure it with the right tool, and then buy the right shim for the job. And the micrometer tool you need can be had for less than $15.

Doing the guess and check can work, but it's a lot more hassle and material...and there is more potential for problems and damage as well.

 

[EliteRetard]

Well I found some info on a CPU that IDC de-lided, it seems the IHS thickness just above the die (NOT the total thickness of the IHS including the lip/legs of it) was about 2.6mm. I say about, sine he had sanded the IHS before measuring.

Basically, if the GPU IHS is of similar thickness (and I think it is), then even the 2mm shim may not be thick enough to get good temps. Perhaps the thickness of the thermal pads will be enough when added on the top and bottom of the shim. You may need to look for a .1 inch shim (2.54mm) or even a 3mm shim. Perhaps a thicker shim with much thinner paste (vs pad) will be the ideal combo...

 

[utk193]

yeah, the ihs thickness+gap can indeed be greater than 2mm.my bad.
you would still want to do what ER says and take the measurements , then proceed from there.

 

[EliteRetard]

At this point he has tried enough shims and has enough sizes it probably wouldn't take to many more tries to get a good fit. Use several thinner sheets to test with and then finally buy the right size (unless the 2mm is just right *crosses fingers*).

 

[kmmatney]

I had a cpu bracket break on me several years ago, so I couldn't mount the heatsink on the cpu anymore. I ended up gluing the heatsink directly to the cpu with silver epoxy, and it ran great for another 3 years. It was a large tower heat sink, and luckily the cpu never came out of the socket with the system in the vertical position. SO silver epoxy worked well for me - there wasn't even any pressure between the heatsink and the cpu - just the silver epoxy holding the heatsink on.

 

[chris89]

Thanks guys!

I just got in some of the best thermal material on the planet at 240w/m K. It's so conductive you can hold the material and cut through ice like butter.

Anyway I used it on my other systems cpu and saw 25C reductions instantly. Idling at 20C and the load to idle time is like 1ms from load temps to idle it's crazy.

On the GTX 480 though I cut a square of the exact size of the die, and another the exact size of the IHS (Stock) and bolted them down.

Temps are still way too high in 3d.

So I've determined from trial and error, that it's the pressure needs to be great between the die and IHS apparently... Because that's where the fault is, between the die and ihs regardless of thermal interface. It just needs to be tight as hell on the die like extremely tight just like stock factory...

I have thick 25mm wide 2mm thick copper shims on the way.

I'd prefer to bond the copper shim to the die so I can bypass the IHS all together. Maybe I'll see if there is some adhesive double sided Graphite or Graphene thermal interface out there at 200w/m K+ rating. Maybe there's a 1500w/m K double sided adhesive thermal interface. Then just mount-up, it'll be semi-permanent between the heatsink/ die/ shim but that's okay as long as that first time around keeps temps extra low I'll be happy.

Come to think of it maybe I could bypass the copper shim as well and get thermal highly conductive material at 2.5mm thickness to place between the die and heatsink and bypass them all together. I guess at that point I'd be relying on great thermal resistance but I'll do some research.

I need to let the adhesive bond at high pressure like have it clamped for 24 hours before testing...

This all comes down to how much money am I willing to spend? I guess I'll just go about it as if it were a learning process.

-------------------------------------------------------

As for one side adhesive material especially at 2.5mm thickness is $143 and just 6w/m K but is an 8"x8" piece. Far too much. I'll need to call them and see if they can cut a custom size. They have a one sided adhesive material at 700w/m K at 2.54mm or 2.79mm or 3.06mm

Actually Panisonic has the 700w/m K material which would require the shim one side adhesive... Maybe I need double side If I'm going to use the shim but it's cheap at like around $20 shipped. It's a 2.5x3.5" sheet.

 

[Techhog]

I wouldn't count on this thread becoming "massive," given that modern GPUs don't have lids to remove.

 

[chris89]

Just a small update. I ordered the parts I needed to get this working and took it to the next level. As we know Fujipoly 17w/m K pads are $150 for like 8"x8" piece.

I am going to run Graphite one-sided ahesive on the VRM's, and Ram at 240 Watts Per Meter Kelvin conductivity.

Since double sided adhesive Graphite at 1850 Watts per meter kelvin was out of stock I had to use the best available Silicone Elastomer double-sided adhesive at 6 watts per meter kelvin on the Core. Then I'll add the Copper shim either 1/8" or 1/4" piece of copper to replace the IHS. However my Shims are a bit too big in dimensions so I'll need to trim them down... I think 1/8" is plenty.

Anyway for the Core it goes at the moment Silicone Elastomer to copper to 400 Watts per meter kelvin One-sided adhesive so I'll adhere the Core to the Shim, then Adhere the graphite to the shim which will connect with the heatsink for Temperatures beyond what I could imagine at the moment. However the 6 watts per meter kelvin Silicone Elastomer will be a massive bottleneck since the numbers speak for themselves.

As you can see the Mounting Bracket is modified and I do that to all my graphics cards. Since at peak potential the thermals will benefit from the Wide-Open free-flowing Restrictionless fan operation. You can also feel airflow up to 15 feet off the back of the computer. Compared to 6-8 Inches with the stock high restrictive bracket. Not Even MSI or EVGA's "Special" high flow bracket will even come close to the performance from this modification.

 

[PhIlLy ChEeSe]

Why even bother, people put liquid nitrogen on there GPU'S you may lower the temps a bit(-25c is a stretch)but your not gonna set any world records.

 

[chris89]

I mean has anyone been able to run a 480 at 900 Mega Hertz on the stock cooler?

It's a matter of keeping the temps from going up. On the 480 at the volts required for 900Mhz operation it is stable per-say until the temps exceed 88C on 100% fan and temps continue to rise. That's on 5 Watts per meter kelvin thermal interface material from bergquist. Very good stuff far better than any paste. With 240 to 1850 Watts per meter kelvin is all we may need to stabilize at 85C and not continue to rise totally stable at 43 Billion 200 Million Pixel's per second. The GPU Ram Too At 384-bit can see Clocks as high as 2,240Mhz which results in 210GB/s memory bandwidth. However 2133Mhz is tops for The Bergquist which yielded 200GB/s at 2133Mhz. Graphite may allow true Stability at 210GB/s.

Also the math is quite impressive on the 480 as you see the Giga Pixel Throughput of the Uber-Old Ancient Geezer GTX 480. The GTX 480 has 48 ROP's and all you need is a good Clock to push past very expensive video card's pixel output... Clocking the 480 as follows...

GTX 480 to 900 Mega Hertz Multiplied by 48 ROP's ... That's 43 Billion 200 Million Pixel's per second...

Just to name a few cards that can do that or more stock are as follows ...

GTX 960 is .............39 Billion 200 Million Pixel's per second...
GTX TITAN BLACK is 42 Billion 700 Million Pixel's per second...
GTX 780 Ti is .........42 Billion Pixel's per second...
GTX 970 is .............54 Billion 600 million Pixel's per second...

I don't believe anyone has done that while maintaining stability.

With this material we are seeing conductivity rates exceeding the best pads by 23 Times better performance. It's all about perfect contact and pressure. As you can see the graphite gave good contact on all ram, and vrm's.

To be honest the graphite is really cheap like $8 for a massive sheet.

So to get back to your question.. Why Bother? Why bother fixing a perfectly fine video card... Or leave it on the shelf to collect dust?

Why not at least get it working and see the gains of something no one has ever tried on this planet?

 

[chris89]

After further investigation the gpu diode is not flat AT ALL.

It only contacts around the edges of the chip about 1/1000th of a millimeter of contact.

So I need to LAP the stupid GPU diode... Wow

 

[PhIlLy ChEeSe]

You could just volt mode the card flash the bios and run full fan, I just don't see the point of it. Didn't the 480's have a higher failure rate then the 580's(580 had built in reduction)at a certain voltage draw where as the 480's did not(see the link) http://forums.anandtech.com/showthread.php?t=2119578
Hey if you can reduce the temp's 25c, and run full balls out all the time cool!!! For all the effort I'd suggest a LN2 POT and a voltage mode is all.
Now if you could do something about the junk ass fan's Asus put on the 780TI DCII OC, or mode a nice fan then I'd be happy!!! I don't mean to high jack yer thread, please move on!

 

[RussianSensation]

The GTX 480 has 48 ROP's and all you need is a good Clock to push past very expensive video card's pixel output... Clocking the 480 as follows...

GTX 480 to 900 Mega Hertz Multiplied by 48 ROP's ... That's 43 Billion 200 Million Pixel's per second...

Just to name a few cards that can do that or more stock are as follows ...

GTX 960 is .............39 Billion 200 Million Pixel's per second...
GTX TITAN BLACK is 42 Billion 700 Million Pixel's per second...
GTX 780 Ti is .........42 Billion Pixel's per second...
GTX 970 is .............54 Billion 600 million Pixel's per second...

I don't believe anyone has done that while maintaining stability.

That's not how it works though. Theoretical pixel fill-rate and actual pixel fill-rate performance are completely different.

680 is miles faster than a GTX570 in pixel fill-rate.

780Ti is miles faster than a 680

980 and Titan X are way faster than a GTX580.

At this point, even if you overclock GTX480 to 900mhz, it will add 2-3 fps in games and at best get you to GTX580 level of performance which is barely faster than a $99 R7 260X

http://www.computerbase.de/2015-05/...ergleich/2/#abschnitt_leistungsratings_spiele