TR Forums

I'd like to do this myself when I build mine, but I dunno if I've got the guts to actually go through with it, heh.

vargis14 wrote:

Grats chuck! If you want it to last 5 years it would probably be a good idea to run her at 4.5 ghz and have a lower temp and voltage for 24/7. But if you need that 200+ mhz 24/7 by god its your rig and your under 1.3v.

I also have those cougar fans...great build quality, I dont know about lasting 300,000 hrs but they are nice

The Cougar fans have been pretty nice. The machine isn't dead silent, but it's not too loud and the sound is a low whirring that isn't annoying like when the high-RPM fans would ramp up on my old system.

I was able to touch 80C+ using the most intense parts of the mprime test, but in day to day use (even compiling using all hyperthreaded cores at once) I don't even break 60C. That's why I'm very happy with the results and can use 4.7GHz as a regular overclock.

kumori wrote:

ColeLT1 wrote:

Chuckula, I read an interesting article, and trying my best to find it again, that they tested all forms of different TIMs, and the intel's ranked up with the best. They posted temps and results, then culprit was not the TIM, but the black epoxy gluing the IHS to the board creates too large of a gap between the IHS and core. They removed all the epoxy and put the IHS back on without touching the TIM, then removed the IHS and tested with high quality TIMs between the IHS and core, and saw equal temps to high quailty TIM, and much lower temps than stock.

Granted, solder would be better than any TIM.

I think this is what he was referencing.

http://forums.anandtech.com/showpost.ph ... tcount=570

Thanks Kumori, that that looks like it!

ColeLT1 wrote:

kumori wrote:

ColeLT1 wrote:

Chuckula, I read an interesting article, and trying my best to find it again, that they tested all forms of different TIMs, and the intel's ranked up with the best. They posted temps and results, then culprit was not the TIM, but the black epoxy gluing the IHS to the board creates too large of a gap between the IHS and core. They removed all the epoxy and put the IHS back on without touching the TIM, then removed the IHS and tested with high quality TIMs between the IHS and core, and saw equal temps to high quailty TIM, and much lower temps than stock.

Granted, solder would be better than any TIM.

I think this is what he was referencing.

http://forums.anandtech.com/showpost.ph ... tcount=570

Thanks Kumori, that that looks like it!

Thanks to both of you guys for the info!

chuckula wrote:

Lemme know how those OCs go for you!

1.290v was 30hrs prime stable, max core temps under prime were 80-82c, but most of the time it was in the 70s in prime. I set the voltage to adaptive mode 4.6GHz @ 1.290, gamed on it one night, then let unigine heaven run for about an hour, and I shipped it out UPS Wednesday.

I may have to do this to my ivybridge one day

I have looked at the link kumora posted http://forums.anandtech.com/showpost.ph ... tcount=570 and using a paper shim is stupid in my eyes since it does not give you the closest distance between the IHS and the CPU die. Plus the way the Intel CPU retention bracket really pushes the CPU die hard into the 1150 socket I think the extra pressure on the CPU silicon die itself plus the liquid metal I would think it will give you much better heat transfer. If the IHS does not seem to be clamping down hard enough on the IHS i would put the paper shim on the outside of the IHS in between the IHS and retention Mechanism.

Chuck I have another question if you do not mind me asking? How was the Coollaborotry ultra to work with? Easy? Hard? Also how was the imprint from the IHS to the K60 using it. I am soon going to be replacing my TIM on my AIO liquid cooler and since i am running Sandy with the SOLDER. I just might replace the whole cooler since it has over 2 years on it. Also Chucky This question has been killing me and since you have the hardware, Have you tried messing with the base clock settings/strap speeds and running a lower multiplier with a higher base clock @ the same overclock speeds say 4.5ghz to see if it makes a difference in performance, temps, Voltage in that does it need more or less to reach the same OC ETC. As of yet no one has said a word about it. Thanks for your replies.

Now my rant/opinion on Why Chipzilla DID NOT SLODER IVY and HASWELL

"Since I feel Intel Purposely put Horrible TIM on IVY bridge and Haswell to limit Overclocking on non Extreme cooling solutions like Phase Liquid Helium ETC, so it would not cut into Sandy Bridge-E sales, Since If you could safely cool a IVY or Haswell with a AIO liquid cooling solution or a Good Air cooler and have 5ghz plus Ivys and Haswells I feel that unless a person really needs all the extra PCI-E lanes ETC Sandy bridge-E Gives you. I think it would have cut into Sandy -E sales bigtime since there are many Work Station 1155 motherboards that support more then the 16 PCI lanes thanks to Plex chips and other solutions. Now i pray they Solder the new IVY bridge -E IHS on! since the new Ivy Bridge-EP processors will each have 80 PCIe lanes (an increase of 40 lanes).
I also think that Soldering the IHS to the Die increases Surface area since thee fluxless solder does run down the side of the CPU die itself say a mm or so. Therefore it pulls heat from the side of the CPU die itself. One mm of solder running down the side of a CPU die would substantially increase surface area along with puling heat from the side and not just the top of the CPU die since a thick layer of solder conducts heat 1000% better then a thick layer of TIM paste that would be on the side of a CPU die. I think that is why when people removed the IHS altogether and mounted there cooling solution directly onto the CPU dies there was not as much of a increase in cooling since the surface area was so small. Hope this make sense....But i really have the feeling Intel did not want IVY and Haswell lowering there Sandy-E sales for desktops. Just look How Sandy -E overclocks Damn good! Why? Solder:)

OK i do not know triganomatry so, Does anyone know how would i caculate the increase in CPU die sizes increase 1mm was added to all 4 sides of a 159.8mm2 die(this is a 3770k die). How much surface area would be gained.

I want to know this since I posted my hypothesis on my Intel did not solder Ivy and Haswell in the last post of this thread. Thank you ahead of time to anyone that answers this.

vargis14 wrote:

I have looked at the link kumora posted http://forums.anandtech.com/showpost.ph ... tcount=570 and using a paper shim is stupid in my eyes since it does not give you the closest distance between the IHS and the CPU die.

That's the point. The shim was there to prove that the glue is the issue, the shim is representing the removed glue.

Ahh ok Cole I got you now...thanks.

vargis14 wrote:

Does anyone know how would i caculate the increase in CPU die sizes increase 1mm was added to all 4 sides of a 159.8mm2 die(this is a 3770k die). How much surface area would be gained

Sqrt(159.8 ) = 12.64mm x 12.64mm => 4(12.64mm)(1mm) = 50.56mm2 => 159.8mm2 + 50.56mm2 = 210.36mm2 = 31.6% increase in contact surface area with the CPU die.

New update with more good news... I shaved another 6-7C off the worst-case temperature just by lowering the voltage offset value in the UEFI interface (still stable). My worst-case temperature even after I took steps to cutoff ventilation to my man cave barely hit 75C on the hottest core @ 4.7 GHz in the 8K length mprime test! While the exact Vcore is hard to measure with the UEFI in "adaptive" voltage mode, I'm in the 1.25 - 1.3 range running flat-out with much lower voltages at idle. I'm actually pretty happy with the temperatures at this point since my hottest core is barely hitting 75C when I really try to pull out all the stops to get the temperature up.

I'm pretty much 100% certain that I could get this guy up to 4.8GHz for benchmarking if I felt like it, and I could probably do a 5GHz suicide run that would run long enough to gin up some useful benchmarks.. but I'm just really stoked to have a solid 24-7 OC configuration.

Vargis14 had some questions so here are my answers:

Chuck I have another question if you do not mind me asking? How was the Coollaborotry ultra to work with? Easy? Hard?

It's not the easiest stuff to use but it's not unamanageable either. If you have experience with soldering electronics, I compare it to a ball of liquified solder... except that it is liquid at room temperature. It's relatively viscous and tends to bead up. The techniques I learned from Youtube are that you form a small (and I do mean small) bead on a piece of paper and use the small paintbrush that they throw in with the kit to paint a very very thin (no really, THIN) layer of the stuff on your IHS/CPU die/surface to be treated. All the while you need to be extremely careful to prevent it from getting onto something that could cause a short.... it's electrically conductive! You can wipe the stuff up if it gets where it shouldn't be so just watch carefully where it ends up. This is different than my normal drop & pressure spread technique for using normal TIM componds, but it seems to work very nicely.

Have you tried messing with the base clock settings/strap speeds and running a lower multiplier with a higher base clock @ the same overclock speeds say 4.5ghz to see if it makes a difference in performance, temps, Voltage in that does it need more or less to reach the same OC ETC.

That's an excellent question. I honestly don't know the answer For the record, I have done only multiplier overclocking using the "classic" method without messing with the multipliers or clock straps. I really wish that Intel had enabled the clock-strap overclocking on the non-K series parts since it would be interesting to try it out on a non-K part. On a K-series part, I'm not sure there is much of an advantage to it outside of people doing crazy LN2 overclocking.

Another update: This time I have an actual benchmark to share

It turns out that the 64-bit version of Cinebench 11.5 runs just fine under Wine on my Linux system. I ran the CPU test a few times and got scores in a range of about 10.1 - 10.25. From what I can tell, that's pretty much the score you'd expect to get on Windows for the same clockspeed, so I'm pretty excited that Cinebench can work on Linux too. The multithreading works too, all 4 cores + 4 hypethreads get used. Core temperatures really never jump above the low-mid 50C range.

Here are three screenshots I took of a run including readouts from the i7z utility of my real CPU clockspeed. I'm using Ronch's desktop background for the new FX processors as a bit of an inside joke, and to prove that my results are either legit, or I've gotten a whole lot better at faking screenshots recently


First shot taken near the beginning of the run
Second shot taken farther into the run showing CPU temps in the 50's
Third shot shown in fullscreen after the run completes with a score of 10.23

Nice score!!! my 2600k might have hit 9.1 @ 5+ ghz i forget and i am not in the mood to pump 1.5 volts through my chip in the summer.

In the winter i have a small dual fan for my window and my haf922 case is 6 inches from the window Perfectly lined up with my side intake fans. The window fan runs slow enough that I can get below zero ambient temps in my case without freezing my computer room. So far thats the only time i have run a 5+ghz for any extended amount of time since I am sporting a little 120mm AIO cooler like a h-50. On top of that I have a wood stove in the basement that when its kicking my heat does not even run since the basement breaks 90 degrees with fans blowing the heat upstairs thanks to a few fans(plus a custom doorway fan) forcing the hot air upstairs into the living room But since the thermostat is in the living room the upstairs averages 55-60f when its below freezing outside. Me and the wife love it cold in our bedroom.....Lotsa covers and spooning if necessary Plus one 150 lb female great dane in our king size bed. Yes she gets under all the cover with us in the winter. But in the Summer with the AC set at 65 she just gets under the comforter...not the sheets and velux...hehe
No our Dane's do not have fleas. I have not had a dog with fleas in the 18 years i have been married. That includes 3 great danes, 2 Basenjis, 1 mastiff, 1 Sheperd dane mix. We keep 2 dogs at a time...but when we first got married we did have 3 the 2 little basenjis and a dane:)

DPete27 wrote:

vargis14 wrote:

Does anyone know how would i caculate the increase in CPU die sizes increase 1mm was added to all 4 sides of a 159.8mm2 die(this is a 3770k die). How much surface area would be gained

Sqrt(159.8 ) = 12.64mm x 12.64mm => 4(12.64mm)(1mm) = 50.56mm2 => 159.8mm2 + 50.56mm2 = 210.36mm2 = 31.6% increase in contact surface area with the CPU die.

Thanks Dpete!!! i knew it would be a substantial increase in surface area....I was guessing 25% but 31% is dramatic...Now if Intel soldered Ivys annd haswells I think they would be hitting much higher overclocks, at least a few 100mhz more with approx 30% more surface cooling contact area with a wee bit from the sides of the CPU dies. Temps would be 20+c cooler