Getting ready to delid Haswell..

[JimmiG]

Seems high. Was that with the .1v avx spiking?

Are your core temps more consistant now?

I've decided againt delidding mine. I get 4.4ghz just under 1.2v's without throttling in Linpack 11 with the H100i. Just need to replace the fans with quieter ones. Using quiet profile is pretty quiet but temps will hit 80's with IBT or 90's.with Linpack 11.

Will have to clock back to stock to make sure

Definitely more consistent as you can see in the Linpack AVX2 results - all cores peaked at 81C. Previously they were at 93 / 98 / 97 / 89 C

I think about 1.2V is the "sweet spot" for a non-delidded CPU using high-end cooling. Beyond that, temps become a concern with the most demanding stress tests.
Unfortunately mine was not quite stable at 4.3 GHz / 1.20V. Had to increase it to 1.218V to get it completely stable, which increased temps by several degrees and caused throttling in Linpack 11 before the de-lid.

 

[Sleepingforest]

Nice test! It's glad to see that people with Haswell can make thermal gains roughly equal to those of Ivy Bridge via delidding.

 

[JimmiG]

One more thing. I was surprised (or not really) by the poor quality of the stock TIM application. When I lifted the IHS off, almost all of of the TIM stuck to the underside of the IHS. The die barely needed cleaning (of course I did anyway), it was already shiny. The layer they applied was very thick and most of it was around the outside of the die. It was also hardened and crusty.

I think that when they assemble these CPUs in the factory, they don't actually apply the TIM to the die. They just dab a big lump of thermal paste at the underside of the IHS and then plonk it down.

 

[Aikouka]

I've been reworking my PC build, and while I had my CPU out last night, I was debating whether I should delid or not. Unfortunately, I haven't really done much overclocking yet as my build has never truly been finished, so I'm not really sure if my thermals demand it. I did run Cinebench at stock speeds, and I only hit 50C on water, but I also get around 45C just playing a game, which only hits maybe 20-50% utilization (the higher amount is usually loading screens and such). Although, my epoxy gap seems pretty small already. A good visualization is that it's about as thin as 2-3 sheets of paper.

 

[sm625]

Looks to me like you've gained 17C. CPU temp went from 88 down to 71. Also, your max core temp was 98 and is now 81. That's also 17C. Yet you mention 12C. Perhaps we need to standardize our method of determining the before and after temperatures.

 

[JimmiG]

I've been reworking my PC build, and while I had my CPU out last night, I was debating whether I should delid or not. Unfortunately, I haven't really done much overclocking yet as my build has never truly been finished, so I'm not really sure if my thermals demand it. I did run Cinebench at stock speeds, and I only hit 50C on water, but I also get around 45C just playing a game, which only hits maybe 20-50% utilization (the higher amount is usually loading screens and such). Although, my epoxy gap seems pretty small already. A good visualization is that it's about as thin as 2-3 sheets of paper.

I'd say wait until you've seen how the CPU works out of the box. Even though the individual steps to de-lid are safe as long as you do them correctly, there are so many moments when a small mistake could ruin your CPU.

Pre-delid I used to hit no more than 65C in most games when overclocked - there's a massive difference between stress-tests (especially those that use AVX or AVX2) and actual games. If you don't care about stress tests, it's possible to have a system that will run any normal game or application with 100% stability even though it would overheat running Linpack.

Looks to me like you've gained 17C. CPU temp went from 88 down to 71. Also, your max core temp was 98 and is now 81. That's also 17C. Yet you mention 12C. Perhaps we need to standardize our method of determining the before and after temperatures.

It's variable and depends on the application. The reduction is greater the more heat the CPU generates. With Linpack AVX2, the max temperature must actually have been over 100C, because the CPU throttled by 6% pre-delid (red line in the Aida64 screenshot). The peak must have been too brief for Aida64 to pick it up, since it only recorded 98C max. Now it's 81C, so at least a gain of 19C. In Prime95, the drop was smaller, ~87C to ~73C or 14C.

 

[Elfear]

OP - What did you use to clean the black sealant off with?

 

[JimmiG]

OP - What did you use to clean the black sealant off with?

Credit card, my fingernails, lint free cloth, rubbing alcohol and lots of patience

 

[Elfear]

Credit card, my fingernails, lint free cloth, rubbing alcohol and lots of patience

I think I have all of those on hand.

 

[Dahak]

Credit card, my fingernails, lint free cloth, rubbing alcohol and lots of patience

Credit card? that for when it does not work?
fingernails? for biting when nervous?
lint free cloth? to wipe sweat off forehead from nerves?
rubbing alcohol, may need to substitute real alcohol

and in case it did not come across with all the smilies, this was meant to be funny

 

[Kenmitch]

I'd say wait until you've seen how the CPU works out of the box. Even though the individual steps to de-lid are safe as long as you do them correctly, there are so many moments when a small mistake could ruin your CPU.

Pre-delid I used to hit no more than 65C in most games when overclocked - there's a massive difference between stress-tests (especially those that use AVX or AVX2) and actual games. If you don't care about stress tests, it's possible to have a system that will run any normal game or application with 100% stability even though it would overheat running Linpack.



It's variable and depends on the application. The reduction is greater the more heat the CPU generates. With Linpack AVX2, the max temperature must actually have been over 100C, because the CPU throttled by 6% pre-delid (red line in the Aida64 screenshot). The peak must have been too brief for Aida64 to pick it up, since it only recorded 98C max. Now it's 81C, so at least a gain of 19C. In Prime95, the drop was smaller, ~87C to ~73C or 14C.

Aida64 by defaults polls cpu every second, temps every 5 seonds. You can change it on the preferances tab > general tab? On phone so can't verify currently.

 

[JimmiG]

Aida64 by defaults polls cpu every second, temps every 5 seonds. You can change it on the preferances tab > general tab? On phone so can't verify currently.

That may be worth looking into. The temps with Haswell, especially running AVX/AVX2 stuff are extremely spiky. With my Phenom II the temp would stay constant within 1 or 2C throughout a Prime95 run, but with the 4770K it can change by up to 10C from one sample to the next...Of course if you run it long enough, the statistic page in Aida64 will eventually catch one of those peaks.

Credit card? that for when it does not work?
fingernails? for biting when nervous?
lint free cloth? to wipe sweat off forehead from nerves?
rubbing alcohol, may need to substitute real alcohol

and in case it did not come across with all the smilies, this was meant to be funny

That was pretty funny My forehead and palms were indeed very sweaty throughout the process. Repeatedly whacking your $330 CPU with a hammer until it splits into (hopefully no more than) two pieces, scraping it with hard plastic, dousing it in alcohol and finally applying a dangerous, conductive material that looks like mercury to it tends to do that.. :biggrin: I sure hope Haswell-E will use solder.

 

[Chipfiref]

When delidding and using CLU for a direct die to IHS contact you need to keep in mind:

-The black silicone glue needs to be effectively removed from the chip PCB without damaging the PCB. This black glue is your enemy - it lifts the IHS off the pcb causing a gap. The vertical displacement caused by the glue's application and curing seems to be one inconsistent variable between all the differently performing chips coming from manufacturing.

-With the CLU method of delidding, you want the IHS to skate freely on the top surface of the CPU die, to be held in place only by the motherboards retention mechanism, when you have finally cleaned the underside of the IHS, applied CLU to the die top surface ( and possibly a little to the unsderside of the IHS ), and set the IHS directly onto the cpu die. DO NOT ever let the IHS skate very far after CLU application (you should test that is skates freely before CLU application) so that you do not smear the CLU too much after you have applied it. It does not "run" like a liquid but you do not want to smear it too much after you get the proper amount liquid metal spread thinly on the top surface of the die.

-When the retention mechanism is clamped down on the IHS, it literally flexes the PCB between the socket edges and the IHS pressure on the die (just as your MB flexes when you insert RAM). If there is any significant silicone glue on the PCB or underside of the IHS, it can affect the contact between the underside of the IHS and the top surface of the CPU die.

-The reason to have direct contact between the IHS and the chip die is that CLU application is a very thin film, which will literally fuse to the IHS and contact directly the CPU die surface, IF the gap is very small. CLU is several kinds of exotic metals in a liquid state that may become more solid at about 50-70C. The CPU die surface is pretty hard (up to a point) - not sure if it actually fuses with the CLU.

-You will need to manually adjust the IHS over the CPU die when applying the retention mechanism clamp. It may take 3 or 4 attempts to get the chip aligned correctly as the mechanism lowers into place on the IHS, so that the IHS ends up perfectly square and in the right location on the PCB - consult a pre-delidding photo of the chip to estimate when the IHS should end up.

 

[rvd123]

Nice work.

 

[tweakboy]

Ya let it settle, I bet temps will go down even more.