Bare-die testing: A delidded 3770k, an H100, and 9 different TIMs

[Nemesis 1]

IDC, SStephan does make a valid point about running without a lid / Its really not a problem tho. I use the lid But We machine out the part covering the die . Than we remove 1/2 of the material on the lid top. than we machine the base 1/2 material removed . Than we machine the lid top further to get .around a thousandth differance between lid above the die. Remember that material I removed from the base . That thickness has to have shims on the hold down where it contacts the lid.

 

[Nemesis 1]

Great topic - am doing mine tonight!



Also great info - something for us all to bear in mind!

What material are the lids made of that cover the die ???

 

[GreenChile]

What material are the lids made of that cover the die ???

I believe they are nickel plated copper. Could be wrong about the plating metal but this is typical. The core of the lid is definitely copper though.

 

[Idontcare]

IDC, SStephan does make a valid point about running without a lid / Its really not a problem tho. I use the lid But We machine out the part covering the die . Than we remove 1/2 of the material on the lid top. than we machine the base 1/2 material removed . Than we machine the lid top further to get .around a thousandth differance between lid above the die. Remember that material I removed from the base . That thickness has to have shims on the hold down where it contacts the lid.

Thats great and all but do you have any data showing that you actually need to go through all that time, effort, and expense?

I would put more stock into SStephan's concerns if there were actually more people than just myself who experienced a dead chip after having mounted it bare-die style. His proposed failure mode to account for a singular failure does not make sense in light of the statistics of the number of other cases which have not failed.

He proposes a systemic failure mechanism which should impact every single bare-die rig out there, whereas the statistics only support the occurrence of some kind of one-off chipkill event (like an ESD event).

I have no doubt your efforts to mill down the IHS and so forth produces one cool looking product, but having put together multiple delidded setups and what must have been close to 50 or more mounts, I see you cautionary efforts with the milled IHS as being entirely unwarranted and needless.

Did you just decide it needed to be done, and so you did it, or do you have some data and results which justify the approach you took to mounting the chip bare-die?

 

[sstephan]

i already did a "proper" bare die instal .
first the compression of the socket ILM was measured :

before closing the socket :

after

its even deeper then the datasheet .


alignment pins cut :

socket ILM modified , 0.4mm shim on top of the cpu , so the pcb wont flex much under pressure .
liquid pro on die .

waterblock instal , measuring compression ...

closed the bolts , until the right compression reached ...

my 5GHZ temps are around 63C , 2hrs prime 27.7

5ghz is for kids . 5.2 suicide run , under prime , 10mins

 

[Idocare]

In my experience there is no better thermal interface than my personal home-made one, pure syntethic diamond powder rated at 100000 mesh (roughly 150nm) or 0-1/4 (in um that equals from 0 to 1/4um or >99% that could pass through a mesh with holes of 150nm) and a dispersant, the better (affordable) till now being bidistilled water and an acceptable one PG (Propylenic Glycole), water requires to be worked very fast in order to not let the compund essicate while PG allows a confortable amount of time evaporating only after few hours of load on the working machine.

Unfortunately this cannot be used on the bare die cause diamond powder is diamond powder, with much more hardness than silicon and applying pressure on the die covered with even the thinnest powder of diamond is always like putting pressure on a glass with a diamond, it will eventually break...

I don't know what "they" use in IC Diamond but the specs are <40um that means >200 times bigger than what "I" use, furthermore, coupling this thing with "perfectly-flat-mirror-like" surfaces (like the ones Idontcare likes) leads as much as possible (unfortunately only) "near" the phonon-phonon coupling.

I'd like to find a better and permanent dispersant with high thermal conductivity but the best of the best I could find till now is triethylene glycol that's not, however, that much better than PG nor permanent and is somehow hard to find but is highly hygroscopic until it lasts and this means it could help a bit the curing of the diamond powder layer.

As already specified in my first post english is not my birth language so I ask you all to forgive mispells if any.

 

[Idontcare]

i already did a "proper" bare die instal .
first the compression of the socket ILM was measured :

before closing the socket :

after

its even deeper then the datasheet .

alignment pins cut :

socket ILM modified , 0.4mm shim on top of the cpu , so the pcb wont flex much under pressure .
liquid pro on die .

waterblock instal , measuring compression ...

closed the bolts , until the right compression reached ...

my 5GHZ temps are around 63C , 2hrs prime 27.7

5ghz is for kids . 5.2 suicide run , under prime , 10mins

Those are great results sstephan :thumbsup: Nicely done

In my experience there is no better thermal interface than my personal home-made one, pure syntethic diamond powder rated at 100000 mesh (roughly 150nm) or 0-1/4 (in um that equals from 0 to 1/4um or >99% that could pass through a mesh with holes of 150nm) and a dispersant, the better (affordable) till now being bidistilled water and an acceptable one PG (Propylenic Glycole), water requires to be worked very fast in order to not let the compund essicate while PG allows a confortable amount of time evaporating only after few hours of load on the working machine.

Unfortunately this cannot be used on the bare die cause diamond powder is diamond powder, with much more hardness than silicon and applying pressure on the die covered with even the thinnest powder of diamond is always like putting pressure on a glass with a diamond, it will eventually break...

I don't know what "they" use in IC Diamond but the specs are <40um that means >200 times bigger than what "I" use, furthermore, coupling this thing with "perfectly-flat-mirror-like" surfaces (like the ones Idontcare likes) leads as much as possible (unfortunately only) "near" the phonon-phonon coupling.

I'd like to find a better and permanent dispersant with high thermal conductivity but the best of the best I could find till now is triethylene glycol that's not, however, that much better than PG nor permanent and is somehow hard to find but is highly hygroscopic until it lasts and this means it could help a bit the curing of the diamond powder layer.

As already specified in my first post english is not my birth language so I ask you all to forgive mispells if any.

Welcome to the forums Idocare :thumbsup:

I suppose you could concoct a hybrid-TIM by combining your diamond powder by throughly mixing it into Liquid Ultra

 

[Idocare]

Welcome to the forums Idocare :thumbsup:

I suppose you could concoct a hybrid-TIM by combining your diamond powder by throughly mixing it into Liquid Ultra

Thanks. :biggrin:

I thought even a more questionable and maybe more efficient mix, combining it with Liquid Pro, since Liquid Ultra already has a copper-graphite matrix, maybe using it as a filler for the base alloy of LP could be really effective.

I only have to find a good way to mix the two things up without messing anything since diamond powder in that size is highly hygroscopic and tends to bind togheter forming pretty solid grains that needs to be broken down to powder again and in this process cannot be stopped so easily from scratching everything and collecting unwanted particles. If I'm not wrong your poor H100 knows something about the ability of diamond powder to collect everything is rubbed onto. :whiste:

With other materials it's not been a problem since I always mixed it directly on the IHS and the HSF with the help of a lattice glove, with a liquid metal things are a little more complicated.

However this is the next experiment on my list, I will report if something good will come out.

 

[Idontcare]

Thanks. :biggrin:

I thought even a more questionable and maybe more efficient mix, combining it with Liquid Pro, since Liquid Ultra already has a copper-graphite matrix, maybe using it as a filler for the base alloy of LP could be really effective.

I only have to find a good way to mix the two things up without messing anything since diamond powder in that size is highly hygroscopic and tends to bind togheter forming pretty solid grains that needs to be broken down to powder again and in this process cannot be stopped so easily from scratching everything and collecting unwanted particles. If I'm not wrong your poor H100 knows something about the ability of diamond powder to collect everything is rubbed onto. :whiste:

With other materials it's not been a problem since I always mixed it directly on the IHS and the HSF with the help of a lattice glove, with a liquid metal things are a little more complicated.

However this is the next experiment on my list, I will report if something good will come out.

Yeah I was thinking Ultra instead of the Pro simply from a viscosity perspective. Will be a lot easier to add a drop of Ultra to container of diamond power and mix to form a mostly-diamond paste.

With the thicker Liquid Pro I can't imagine you'll have much luck getting a high volume loading of diamond mixed into it is all.

(I do the same thing when smoking pork ribs, for the dry-rub I add a small amount of olive oil first to make a paste out of the dry rub, which then sticks really well to the side of ribs before they go in the smoker...that's actually what made me think of it when I read your post)

 

[Idocare]

Yeah I was thinking Ultra instead of the Pro simply from a viscosity perspective. Will be a lot easier to add a drop of Ultra to container of diamond power and mix to form a mostly-diamond paste.

With the thicker Liquid Pro I can't imagine you'll have much luck getting a high volume loading of diamond mixed into it is all.

(I do the same thing when smoking pork ribs, for the dry-rub I add a small amount of olive oil first to make a paste out of the dry rub, which then sticks really well to the side of ribs before they go in the smoker...that's actually what made me think of it when I read your post)

Never had the pleasure to play with the Ultra but I deeply think is no more than the Pro with a copper-graphite filler, anyway I found a way to mix the two things without adding unwanted particles, the solution was under the nose but I'm currently deep into other troubles that are unduly using my mind too much.

..a fingertip of a lattice glove is the perfect "high tech" stuff I need.

It's not that I'm stingy, it's they don't sell LP in barrels and the powder is not so easy to find unless knowing someone in the diamond cutting business, I have to mix few at once to see what I get...

I'll update (sooner or later).

P.S. For grilling steaks I usually only let them steep an hour into a mash of extra virgin olive oil (as a dispersant), sliced garlic and rosemary sprigs (as a filler) ..:biggrin:

 

[crashtech]

I just searched for this thread because I'm running out of TIM. I really think this thread, or a trimmed down version thereof, should be made a sticky. Very good information here.

 

[tigersty1e]

I want to see the Liquid Ultra results when applied correct.

On die u only use enough to coat the die there should be no run off or puddles.. like u painted it on very clean..
Like u see with the IHS i did as an example of how much Liquid Ultra to use.

You can do the same thing with the back side of the IHS very thin coat.. this is how the Liquid ultra works it is different then other tims..

As im sure u could see when u used it.

Try this and i bet temps will drop another 10c.
See look.. this is the IHS a much larger area and i used 1/4 of what u used on the tiny die..

VS

There is a huge difference there.. So im hoping u give it another go!

if you put this much, do you risk the extra dripping down into the socket in an upright case?

when I went to put my CLU on my die, the CLU looked like a giant puddle. It didn't flow to the sides, but probably did after I applied pressure from the ihs.

most of the pictures i've seen online had the clu look like it was aluminum. it looked more dry, but mine had more water in it.

 

[Idontcare]

if you put this much, do you risk the extra dripping down into the socket in an upright case?

when I went to put my CLU on my die, the CLU looked like a giant puddle. It didn't flow to the sides, but probably did after I applied pressure from the ihs.

most of the pictures i've seen online had the clu look like it was aluminum. it looked more dry, but mine had more water in it.

It could drip, I wouldn't recommend putting that much on it. Just brush a thin layer on, like you are painting. That is all you need.

 

[crashtech]

I'm pretty happy with the NT-H1 I just bought, it seemed the most idiot proof of the lot and close to the best. What a big change from the Arctic Silver I'd been mindlessly using all these years!

Thanks for all your hard work, IDC. My CPUs thank you, too, lol.

 

[tigersty1e]

I'm not sure how to post pictures here, but I posted my pics over on OCN.

Is that too much?


http://www.overclock.net/t/1313179/official-delidded-ivy-bridge-club/16120_20#post_19692966

 

[Idontcare]

I'm pretty happy with the NT-H1 I just bought, it seemed the most idiot proof of the lot and close to the best. What a big change from the Arctic Silver I'd been mindlessly using all these years!

Thanks for all your hard work, IDC. My CPUs thank you, too, lol.

LOL, happy to have helped!

I'm not sure how to post pictures here, but I posted my pics over on OCN.

Is that too much?


http://www.overclock.net/t/1313179/official-delidded-ivy-bridge-club/16120_20#post_19692966

Nah, that is fine. You could have used less, but you did not use "too much". Its good.

 

[24HZ]

This thread needs a bump for Haswell. I'm not sure if I want to do the IDC 'Home Depot' mount or buy $10 worth of replacement mounting hardware for the XS-PC block to retrofit on a H100.


Edit: Ok I'm going through some ideas.

1)Use a M4/M3 10mm standoff and a few thin washers. By my math, the IHS is about 2.56-2.58mm thick in the part thats right above the die according to IDCs measurements and accounting for the fact that he measured his lapped IHS (added 0.02mm).

The stock corsair standoffs are 13.45mm. 13.45-2.58 = 10.87. So a replacement standoff needs to be ~10.87mm tall.

I don't know the thread spec for the stock standoff but its probably M3 or M4. You can get M3/M4 standoffs in 10mm which is great. To get the extra 0.87mm you can simply use 1 or 2 very thin washers either at the base or at the top of the standoff or both. Just as long as the washers total thickness is around 0.87mm. Its a pain but if you have a very flat heavy object like a machinists 1-2-3 block, you can tape 4 washers evenly spaced on one side and lap them to an exact thickness.

 

[Idontcare]

I like your ideas :thumbsup:

 

[24HZ]

I'm actually leaning towards just buying this:

$5

and this:

$5

Hopefully the springs will let me fudge exact tolerances and I can always throw some washers in to get the right tension and depth.

 

[24HZ]

Well, I went a totally new direction. Bought a H110 and went direct die with a 4770K. Results were lackluster at best. 5-10C drop. Then I decided to pop off the H110 waterblock and put a razor on it. Sure enough the damn thing is extremely convex. *sigh*.

Didn't really plan on lapping but its going under the grind right now.

 

[Idontcare]

Well, I went a totally new direction. Bought a H110 and went direct die with a 4770K. Results were lackluster at best. 5-10C drop. Then I decided to pop off the H110 waterblock and put a razor on it. Sure enough the damn thing is extremely convex. *sigh*.

Didn't really plan on lapping but its going under the grind right now.

My H100 sucked for flatness as well. Seems to be a Corsair thing :\

 

[24HZ]

My H100 sucked for flatness as well. Seems to be a Corsair thing :\

Well, unlike your H100, mine was pretty uniformly convex in the center. Maybe even deliberately so since most of the current Intel IHSs are concave.

Mine had radial machine marks which would imply that it was possible to machine it convex deliberately.

 

[aigomorla]

My H100 sucked for flatness as well. Seems to be a Corsair thing :\

LOL...

wait... the edges were poof'd out more? (concave and not convex)
Looking at how u lapped it...

That is a really bad cold plate...

normally it should be bowed the other way....

 

[sxr7171]

This thread should be a sticky. It is the definitive answer to most if not all questions regarding TIM and delidding.


I'm wondering if there is a way to measure the pressure of the block on the die. Maybe across 3 trials. I suppose there's probably a thin tolerance between near optimal pressure and enough pressure to crack the die.

 

[24HZ]

It should take A LOT of pressure to crack the die if the pressure is evenly distributed.