Delid everything!

[Mesterial]

 

Hi there! I know there are a ton of stories about it, but I just want to share my on-going experience with this.

 

Soooo…. After watching so many videos of LTT and Gamers Nexus, and being so angry about my i7-7820X temps for so much time, especially having 20°C max delta between cores (which appears to be normal, but whatever), I went for some shopping.

 

 

Yes, I feel confident enough to delid my little big boy CPU! 

 

Those Rockit tools look and feel awesome quality, despite being 3D printed! It's very reassuring. 

 

They are made in very hard plastic, I was very surprised about their hardness.

 

I bought of course the Rockit 99 for the LGA2066 socket, but also the Rockit 88 for both LGA1150 and 1155 sockets...

Why? Well, to train first on my "TV" PC, equipped with a small i3-7100T! And then, maybe I will also do it for my mother's computer (i5-3350P) so her CPU stays cool whatever she does (those fff**fff webpage ads can be CPU hungry sometimes....). I mean, an i3-7100T is cheaper to replace than an i7-7820X in case something goes wrong…

Also, I chose the Rockit tools over the tools from Der8auer, because although his tools are made of steel, let's face it, I could buy both the Rockit tools for about the same price as just his Skylake-X tool… It's not to criticize of course, but it's a little expensive. Anyway, I bought all the tools from Caseking.de, so in a way, he has my money already! 

 

I got every tools needed: an iFixit toolkit to help removing the silicon adhesive once delided, some UHU high temp silicone for reliding, some Noctua NT-H2 paste to put on the IHS (why not replacing my Artic MX4 by this all brand-new even higher performance paste), some Thermal Grizzly Conductonaut of course, and some transparent nail polish, which is the good sh*t according to Gamers Nexus! 

 

I read every transparent nail polishes in the store! Beware those Maybelline ones are "rich in iron", that's too bad for a non-conductive purpose! 

 

I will start working on it on Sunday, I will then feed this post with pictures, results and stuff. I'm so excited about it, I had to share this post

 

[Trippsy]

Good Luck!

[Mesterial]

So, I had some time before Sunday to start the deliding of my good old SilverStone "TV" 

 

Hi there, smallest PC I own!  Notice the Intel stock smallest cooler, as the i3-7100T has a TDP of only 35W, IGP included!

 

Aligning the triangles...

And closing the box...

Hang in there, little one!

This part is so scary… I had to open and check three times… And then...

 

It's done! OMG! There's no going back now...

 

So let's open it, and...

OMG it's soooo small!!!  Ain't it cute?!

 

Time for some cleaning. For the TIM, I use some alcohol. For the silicone seal on the IHS, I use some "petrol F" (or zippo fuel?), it helps removing adhesives, and then some alcohol.

I let a little bit of silicone at the "opening" of the seal to mark it, as a relid guidance.

 

For the PCB part, I just scratch the seal with the iFixit tool. It is very practical, I was impressed!

No need to wipe it completely with some chemicals.

 

Then, even if it's clearly not useful, I put some nail polish around the die as a barrier, just to test the nail polish brush.

I read somewhere that the 4 dots are here to check the temperature, so I did not touch them, just in case.

 

And now, the liquid metal...

I'm aware of the very impractical syringe and the mess it can create… So here is my Haagen-Daas cap liquid metal battle station! 

OMG I'm so glad I was warned about the impractical thing… The first drop just dropped right away in the cap, and was too big! I could suck it back in the syringe, and start over. The applying cotton candy tool is very good at spraying the drop into a very thin layer.

 

I applied it a little bit too much area on the IHS, but not too much thickness, as Steve's Gamers Nexus teaches  So it's ok I guess.

 

Now time for the relid. I choose just to put 4 silicone drops at the corners, and 2 lines for the socket grips. I'm sorry, I missed to take a picture before reliding.

 

And finally, the rocket press. The silicone did not make a mess, so I'm glad!

 

Now time for the seal to rest. I'll reassemble the PC tomorrow, and do the same benchmarks I did this morning. But I'm not expecting rocketting performance, this CPU is really small, and with the Intel stock cooler… But we will see that! 

 

Pro tips: on the floor, you can't drop things! 

[Cbennett0811]

Nice play by play. Misread for a moment and thought you said you used alcohol as the tim lol.

[Mesterial]

So, a little drop of Noctua NT-H2 and reassembling.

Let's check the results!

Remember I use the 1 cm height Intel stock cooler… I report the temperatures from Coretemp, so only core temperatures (core0 and core1).

 

i3-7100T, 2 cores, 35W

 

Max temps at Windows idle: 45°C 47°C -> 39°C 41°C (-6°C -6°C)

Min temps at Windows idle + fan at 100%: 31°C 31°C -> 28°C 27°C (-3°C -4°C)

 

IntelBurn HPL (AVX), 2 GB, 20 times in a row, fan at 100%

Max temps: 60°C 59°C -> 54°C 51°C  (-6°C -8°C)

Max score: 48.54 GFLOPS -> 49.20 GFLOPS (89.2% eff. -> 90.4% eff. !)

 

Extreme HPL (AVX2), 3 GB, 5 times in a row, fan at 100%

Max temps: 61°C 61°C -> 54°C 53°C  (-7°C -8°C)

Max score: 78.12 GFLOPS -> 82.14 GFLOPS (71.8% eff. -> 75.5% eff. !)

 

Extreme HPL (AVX2), 3 GB, 5 times in a row right after previous test, fan at auto

Max temps: 66°C 66°C -> 61°C 60°C  (-5°C -6°C)

Max score: 81.90 GFLOPS -> 83.04 GFLOPS (75.3% eff. -> 76.3% eff. !)

(best scores compared to full fan due to cache I suppose)

 

WAW!!! I am really impressed for such a small CPU, using the Intel stock crappy cooler.

 

In few hours, I will start the process for the big one… 

[Jarsky]

I got the same RockIt kit some time back. Delidded my 3770K 

I used the Thermal Grizzly and sealed the IHS with Loctite Gel Super Glue. 

With it overclocked from 3.5Ghz to 4.6Ghz I was seeing approx -15C difference @ 100% load

[Mesterial]

OK, but you have an AIO, and not a midget CPU. The temperature is more important than the delta. At some point, you can't go bellow a certain temperature with the same conductor. The heat generated is kept the same, it's just the speed at which it is evacuated that changes. 73 W/mK, I mean, the i3 is just 35W max TDP, and around 21W during HPL! And with the Intel crappy cooler. 

 

Now this, is the real deal:

[Mesterial]

I'm done for today

 

So, the delid was very different from the small CPU: the silicone was somewhat softer, so I first thought it didn't work because the IHS was back at its place when opening the tool. 

 

Everything cleaned:

 

It was very time-consuming to clean VERY carefully the seal, because of the tiny components. They almost touch the silicone, so I started scratching from the opposite. It's not hard, it just requires some patience (along with the ifixit tools !). 

 

Then, the nail polish on every components:

 

And finally, the liquid metal, the silicone, and the IHS! 

[Mesterial]

And now, the results!

 

i7-7820X, 8 cores, 140W, 4 cores @ 4.6 GHz (0, 2, 5, 6) and 4 cores @ 4.4 GHz (1, 3, 4, 7)  (cores chosen by Intel ranking in the BIOS)

Corsair H115i 

 

Max temps at Windows idle: 54 40 39 39 40 45 40 38 -> 45 35 32 32 31 33 31 32 (-9 -5 -7 -7 -9 -12 -9 -6)

Min temps at Windows idle + fan/pump at 100%: 31 29 29 30 29 30 30 30 -> 31 28 29 30 28 30 30 30 (0 -1 0 0 -1 0 0 0)

 

IntelBurn HPL (AVX), 4 GB, 20 times in a row, fan/pump at 100%

Max temps: 82 88 88 94 92 91 83 79 -> 72 74 76 76 74 75 68 68  (-10 -14 -12 -18 -18 -16 -15 -11)

Max score: 210.15 GFLOPS -> 210.40 GFLOPS (91.2% eff. -> 91.3% eff.)

 

Extreme HPL (AVX2), 3 GB, 5 times in a row, fan/pump at 100%

Max temps: 82 89 88 95 92 92 83 79 -> 72 74 76 78 74 77 68 68  (-10 -15 -12 -17 -18 -15 -15 -11)

Max score: 427.29 GFLOPS -> 430.47 GFLOPS (92.7% eff. -> 93.4% eff.)

 

Extreme HPL (AVX2), 3 GB, 5 times in a row right after previous test, fan/pump at auto

Max temps: 83 95 88 100 97 98 83 82 -> 71 77 77 82 78 81 70 69  (-12 -18 -11 -18 -19 -17 -13 -13)

Max score: 426.94 GFLOPS -> 431.03 GFLOPS (92.7% eff. -> 93.5% eff.)

 

Cinebench R20, fan/pump at auto

Max temps: 87 95 89 100 97 98 83 82 -> 75 72 77 74 71 77 70 67 (-12 -23 -12 -26 -26 -21 -13 -15)

Score: 4586 -> 4599

 

Cinebench R20, fan/pump at 100%

Max temps: 83 79 84 86 80 88 78 74 -> 71 69 73 71 68 73 67 64  (-12 -10 -11 -15 -12 -15 -11 -10)

Score: 4565 -> 4594

 

 

Stunning results… 

 

What is funny, is that min idle temps at full blast fans are the same!

Also, what is amazing, is that I can now run a surprise HPL without fearing a meltdown lol! I guess this is where watercooling shines, now that it should be the "limiting" factor.

Maybe I will put the 8 cores at 4.6 GHz, of even 4.7 

 

So, even for the small 7100T, I am very happy with this delid frenzy that got into me. Of course I will do it on my parent's computer (3 series), but too bad my computer I got there is a soldered i7-960 though!

[Mesterial]

Oh, and the best part: the max delta between cores went from 16 to 10 degrees!! (at 100% fans and pump)

[Mesterial]

In conclusion, I went as far as 2*4.8 GHz + 2*4.7 GHz + 4*4.6 GHz, at 1.28 Vcore, letting all power savings, for 24/7 safe use. I'm happy with it