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

[Idontcare]

Edit: Adding direct links to specific posts within this thread that contain specific updates, a table of contents if you will:

• Bare-die testing: A delidded 3770k, an H100, and 9 different TIMs (that's this post, read below )
• Socket Preparation - Preserving the Backside Socket Support Bracket
• Crafting a robust replacement for the H100 mounting bracket
• Final prep work on the H100 mounting bracket
• Results with MX-4
• Results with IC Perihelion
• Results with Céramique
• Results with AS5 (Arctic Silver 5)
• Results with Tuniq TX-1
• Results with Tuniq TX-2
• Results with IC Diamond
• Results with NT-H1 (Noctua)
• Results with Liquid Ultra (Part 1 of 2)
• Results with Liquid Ultra (Part 2 of 2)
• Results with NT-H1 (the silly, way-too-much, test)
• Results Summary

Original Opening Post:

I happen to have a collection of thermal pastes that I figured would make for some fun testing with bare-die mounting (no IHS) on a delidded i7-3770K.

~$200 worth of assorted TIMs :whiste: Let's see if we can put them to good use

The tests here are all conducted with an H100 (4 fans in push-pull, lapped to 3000 grit) on an Asus MIVE-Z.

I've already removed the IHS from the 3770k, and the bare die is fully exposed (technically we are looking at the backside of the die, the cpu sits upside down with the frontside soldered to the green PCB).

If we attempted to mount the H100 to the bare-die 3770k at this point the socket's retention bracket would get in the way because the bracket hardware sticks up higher than the cpu die itself (normally the IHS pokes up higher than the retention bracket so this isn't an issue).

If we left the socket's retention bracket in place with our lidless 3770k the HSF would not be able to contact the CPU and that would be bad for temperatures

So the next step here is to remove the socket's retention bracket hardware. It might sound like a daunting task but it really is quite simple to accomplish. There are three bolts which use a standard star-key bit available in most $5 bit-kits you would get at Home Depot.

Retention bracket successfully removed

We aren't close to being ready to start our tests though, we need to prepare the socket for test as well as deal with the H100 mounting bracket itself which is now set to the wrong height owing to the now missing IHS.

 

[Idontcare]

Now we have to be careful here because while we do want to remove the socket retention bracket we still want the motherboard to benefit from the backside support bracket (on the reverse side of the mobo, directly behind the socket) to remain in place since it helps support the socket and motherboard from bending and deflecting under the applied load of the HSF mounting pressure.

In order to keep the backside support bracket in place we must rebolt it back to the motherboard using the same three bolts we just removed from the mobo in removing the socket's retention bracket.

However care must be taken to ensure the bolts aren't going to abrade the mobo's PCB now, remember the bracket itself essentially served as a washer for the bolts to keep the abrasion to a minimum.

We have a thickness budget of ~1.3mm and we can see from the retention bracket that the designers built-in a black plastic shim to prevent the bracket from abrading the mobo. To create a suitable proxy for this situation I elected to employ your everyday basic SAE #6 flat washer and some black electrical tape:

^ you'll note the two washers on the black tape are not circular, they have both had one side of the flattened with a dremel tool and some sand paper because they were too wide to fit between the bolt holes and the socket itself on the motherboard.

Cut out the washers from the tape and we are good too go with our poor-man's retention bracket mount substitute:

Checking the taped washer for its final thickness (~1.2mm), we are right in the ballpark of the 1.3mm initial measurement for the bracket + plastic shim:

Next step is to bolt the support backplate back onto the mobo:

And then a confirmation check to verify the CPU silicon die really does extend above and beyond the height of the socket's retention bolt by using a razor blade and confirming visually that the head of the bolt was below the blade by sweeping the blade back and forth (it cleared):

Now we have a prepared CPU and a prepared MIVE-Z for our bare-die testing. We just need to get the H100 mounting bracket squared away and then we can start the actual tests

 

[Idontcare]

Ordinarily the H100 is mounted to a motherboard with the assistance of fixed height metal standoffs. Takes all the guesswork out of torquing down the bolts, you just install the standoff bolts, put your H100 on top of them, and then turn the thumb-heads to full stop and you are done.

The standoff measures out to be ~13.5mm, but remember this height was set with the expectation that the H100 would be touching down onto an IHS, so 13.5mm is now too tall by an amount almost equal to the thickness of the IHS:

To construct our replacement H100 standoff-bolt I could use a simple compression spring, akin to what I did when I delidded my GTX460, but the problem with going that route here is that it doesn't enable us to have a repeatable mounting pressure from test to test because we won't know if we are compressing the springs the same each time we test a new CPU TIM.

Instead, if we opt to put in place something very similar to what we are replacing - a tall metal standoff - then we can expect to have repeatable mounting pressure by simply turning the thumb-nuts to full stop.

My solution was to head to Home Depot and buy a $3 pack of 100 SAE #6 flat washers:

A stack of these will be minimally compressible and they are ~1mm in thickness so they can be easily added or removed to change the effective standoff value.

^ the flat washers stack up nicely in good approximation of our original H100 stand-off bolt. But we need to take some of them away to account for the loss of the IHS over the CPU die, thus lowering the H100 itself by a commensurate amount.

The IHS itself measures right near 3mm:

...don't forget that this includes the perimeter lip of the IHS, it doesn't capture the fact that the inside surface of the IHS form a cavity that sits on top of the CPU die itself (which we must also remember does not sit flush with the surface of the PCB either).

Putting a 1mm thick washer inside the IHS cavity and measuring it again shows us the cavity depth:

Basically we need our replacement standoffs to be 2 washers (2mm) shorter than the stock H100 standoff bolt.

Bingo! That will work quite nicely as a substitute stand-off bolt while accounting for the absence of the IHS

 

[Idontcare]

One concern I had in setting myself up to deal with dozens and dozens of washers was the prospect of miscounting them and getting an extra washer, or lacking one, in any one of the four stacks of washers that I would need for the H100.

A while ago when I first tried to mount my H100 on a delidded 3770k I found out after the fact that I had an extra washer on one of the standoff bolts, which explained why I had so much trouble with the reboots and so forth stemming from very slight adjustments in mounting pressure.

To avoid that fate this time around I devised a simple tool for quickly verifying that the stacks of washers were indeed the correct height:

A toothpick which has been marked in blue to the desired height and washer count. With this handy tool I could quickly touch it next to each of the four stacks of washers and visually verify no stack was missing a washer or had too many washers.

And with that I finished the preparation work for the H100 mounting bracket replacement:

Time to put the setup to the test, add a little NT-H1 to the bare silicon die:

And bolt down the H100 accordingly, tightening the thumb-nuts to full-stop, same as I would do if the IHS was still present:

Now we are ready to generate some temperature data! :biggrin:

(actually the tests are already finished, for all 9 TIMs, I'll post up the results asap tomorrow, hang tight till then)

 

[Avalon]

Very nice, looking forward to your results. Detailed as always, IDC!

 

[BrightCandle]

The approach looks sound, should produce decent results. However that last picture shows about 10x too much thermal paste and is a bit of a concern.

 

[BD231]

I to use surgical gloves and decontamination suit for maximum mhz.

 

[WhoBeDaPlaya]

$10 says IDC is wearing a bunny suit

 

[Haserath]

Can't wait, another bookmark!

 

[MisterMac]

There's a reason IDC is king, there's a reason.

 

[PlasmaBomb]

Looking forward to the results. Though I agree that looks like a lot of TIM. Though I'm sure a lot will be squeezed out with a high mount pressure.

 

[dma0991]

IDC should get paid for doing this. He's practically doing a free advertisment on which TIM is the best. They should sponsor him a couple of tubes instead.

 

[Idontcare]

The approach looks sound, should produce decent results. However that last picture shows about 10x too much thermal paste and is a bit of a concern.

10x too much? You haven't seen 10x too much until you've seen this!

I really did test the above just for fun

Seriously though, you are right to be concerned about the TIM quantity but I assure you I performed tests with less and more to "box in" the right amount of TIM to apply for reaching optimal temperatures with all the TIMs.

What I found was that it was very easy to apply too little, and very difficult to apply "10x too much". The die itself is so narrow, and so long, that the extra just squeezes out over the side without issue but getting the the die covered from corner to corner was the challenge.

Here's an example of what I mean:

^ whenever the corners were left uncovered, an easy thing to do with too little application of TIM, the peak temperatures were typically 2-3°C higher than the peak temperatures for the same TIM if I applied even more and got the entire die surface covered. (no surprise there)

Now going back to the original NT-H1 pre-mount photo which you were concerned had 10x too much TIM, here are the post-mount photos:

As you can see, very good contact was achieved between the cpu die and the H100 block, all four corners were covered and the extra TIM just squeezed over the side without issue.

I checked the mounts, pre and post, for every TIM and I also varied the TIM quantity (did 2-3 mounts per TIM) just to make sure I wasn't seeing temperature results which were due to too much TIM or a bad mount versus being due to the thermal performance of the TIM itself.

 

[Idontcare]

I to use surgical gloves and decontamination suit for maximum mhz.

$10 says IDC is wearing a bunny suit

LOL. My hardware is always trying to sweet-talk me into just trying the rhythm-method, and my response is always "no glove, no love!"

(well, mostly anyways )

I can't afford another "oops" like what happened that summer of '98 :| I had little Malaysian and Costa Rican celerons (they were Irish twins, poor mother) running around my office making a mess of everything, total mayhem I tell ya, total mayhem

 

[OVerLoRDI]

Awesome thread already. Can't wait to see the results.

 

[hokies83]

LOL. My hardware is always trying to sweet-talk me into just trying the rhythm-method, and my response is always "no glove, no love!"

(well, mostly anyways )

I can't afford another "oops" like what happened that summer of '98 :| I had little Malaysian and Costa Rican celerons (they were Irish twins, poor mother) running around my office making a mess of everything, total mayhem I tell ya, total mayhem

LoL waiting to see the Liquid Ultra beating the next best tim by 5c.

 

[Sheep221]

If I want to delid a CPU, the only thing I need to do is to put razor blade beneath the IHS and just remove it by force?

 

[Haserath]

If I want to delid a CPU, the only thing I need to do is to put razor blade beneath the IHS and just remove it by force?

Yes.

Carefully applied force.

It can only be done on Ivy Bridge. Earlier chips were soldered to the IHS.

And here's IDC's post about removing his:
http://forums.anandtech.com/showpost.php?p=33781277&postcount=2

 

[KingFatty]

It's refreshing to hear that we don't really need to worry about applying a little too much paste, and that it's more of a danger to under-apply and miss the paste spreading all over.

Also, that one picture above with "10x" looks like some small animal took a crap on the CPU.

 

[Haserath]

10x too much? You haven't seen 10x too much until you've seen this!

IDC, I think your computer has a worm!

 

[GPz1100]

IMO, the only paste one needs to be concerned about is that really liquid type, like liquid ultra. That stuff is literally like mercury. I applied about 3/4 of a rice kernel to the die, used the included brush to spread it, then called it a day.

Water block (xspc raystorm) went directly over it.

I actually had to redo this because I left off the standard socket backplate. Completely forgot about it. Although the system worked just fine without it, I felt IDC was correct in have it present to help distribute the force over the board.

After a week or so of having the system together, the cooling block was not *stuck* to the die at all, and came right off. The cpu was stuck to it because of no IHS/holddown. Gently wiggling separated the two. The copper block did have some staining however. Used an alcohol swab best I could to clean up both the die and block, then reassembled. Ironically enough, temps went down about 2C on average.

The imprint on the cooling block did not show that the entire die was making contact, so perhaps a bit TIM might of been better. Still, with these temps, im not losing one bit of sleep over it.

Before the reseat, I was peaking out at ~55C on the hottest core running IBT/prime95, after the reseat, 53C. Ambient temp is the same (~72F). Cpu oc's to 4.6Ghz @ +.050V offset mode, LLC set to high, everything else set to auto.

 

[Yuriman]

No offense IDC, but your retention bracket seems overly complicated to me. Why not use springs instead of a huge stack of washers?

BTW, I plan to pull of my waterblock and check to see if my CPU is cold-welded to it in a few minutes, will report back.

 

[EliteRetard]

Remember, with the IHS you don't need full coverage with the TIM...the IHS is much bigger than the die. Full coverage is important if you are going bare die, but otherwise coverage in the center of the IHS should be fine and you actually don't want any spill over of the TIM using the IHS (small chance it could get where you don't want it).

Question for IDC, any chance you could get a hotplate set to 60c and drop some of the TIM on there for a minute? I've always been curious to see what happens to the stuff when its pushing a heavy load. Does it get thinner or stay the same?

Edit here's some pictures:

This is about what you want (a bit to much here), though you can see he still got some on the lip of the IHS and such...this is why I evenly spread TIM with a plastic wrapped finger.

This is what you don't want:

 

[EliteRetard]

No offense IDC, but your retention bracket seems overly complicated to me. Why not use springs instead of a huge stack of washers?

BTW, I plan to pull of my waterblock and check to see if my CPU is cold-welded to it in a few minutes, will report back.

As he explained in his post, because he is testing the variance of the TIM he couldn't have any variance in the mounting hardware or it would skew the tests.

 

[Yuriman]

As he explained in his post, because he is testing the variance of the TIM he couldn't have any variance in the mounting hardware or it would skew the tests.

Thanks, I missed that paragraph.

To construct our replacement H100 standoff-bolt I could use a simple compression spring, akin to what I did when I delidded my GTX460, but the problem with going that route here is that it doesn't enable us to have a repeatable mounting pressure from test to test because we won't know if we are compressing the springs the same each time we test a new CPU TIM.

^ I'm inclined to believe compression with springs probably *won't* give less repeatability than bolting on directly though, so long as you repeat your mounting method and don't switch which spring goes in which corner between mounts. At least with springs you have a good idea of how many N/m² you're putting on if you know the spring rate.