Delidded my i7-3770K, loaded temperatures drop by 20°C at 4.7GHz

[24HZ]

Haswell delid results:

Got a lot of thermal but no real OC headroom. Going to get a new CPU if new bins show up and perform better.

 

[Idontcare]

Haswell delid results:

Got a lot of thermal but no real OC headroom. Going to get a new CPU if new bins show up and perform better.

True you are not getting any more clockspeed headroom, but at those volts and clockspeeds the reduced operating temperature will surely be a welcome relief to avoiding an early chip death from degradation.

(20C reduction will increase lifespan by at least a factor of 4x)

 

[24HZ]

True you are not getting any more clockspeed headroom, but at those volts and clockspeeds the reduced operating temperature will surely be a welcome relief to avoiding an early chip death from degradation.

(20C reduction will increase lifespan by at least a factor of 4x)

Sadly I really don't care about lifespan. Not on this chip anyway. Once I get a hold of something that OCs a bit better, then I'll hold onto that.

I must sound like a bad parent given its fathers day :\

 

[Grahamrkf]

I know this may be a stupid question, but would there be any problem with removing the IHS completely and having the CPU be in direct contact with the cooler?

 

[cytg111]

Haswell delid results:

Got a lot of thermal but no real OC headroom. Going to get a new CPU if new bins show up and perform better.

That shows the same pattern as IVB, perhaps just a little more clearer, temps are secondary(or non relevant if not for throttling) to oc potential of these chips.

look at

http://forums.anandtech.com/showpost.php?p=35116033&postcount=16

"I benched my 2600 and 2500ks past 5.4ghz with these boards.I first bought the gigabyte extreme udh of what ever is was and the pos was stuck at 4.3-4.4 GHz max"

Overclocking same chip, different mainboards.

 

[Idontcare]

I know this may be a stupid question, but would there be any problem with removing the IHS completely and having the CPU be in direct contact with the cooler?

Not a stupid question at all. It does add some logistical complications into the picture but nothing that a screwdriver and a bag of cheap washers can't remedy

Results of Direct-Die Mount with H100 (no IHS)

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

 

[24HZ]

I'll post a diagram later of what I did with my H110. It was literally 8 nylon washers I bought for $1.00.

 

[Idontcare]

I'll post a diagram later of what I did with my H110. It was literally 8 nylon washers I bought for $1.00.

I like :thumbsup: Nylon is a nice compressible polymer, basically a solid-state spring. Excellent idea, and budget friendly

 

[Concillian]

Supposedly thats on an Antec Kuhler H2O 620 closed loop cooler (120mm x 1). No idea if its direct die and what TIM is being used but you can bet its top shelf and more likely direct die.

Quick take aways

1)Haswell is still hotter than IVB delidded
2)Possible Haswell needs slightly more volts for the same freq compared to IVB
3)Delidding Haswell still nets big thermal gains a la IVB

Are you comparing golden Ivys to average Haswells? my Ivy was nowhere near capable of that de-lidded on a Noctua, which should be about equal to a closed loop water.

 

[24HZ]

Are you comparing golden Ivys to average Haswells? my Ivy was nowhere near capable of that de-lidded on a Noctua, which should be about equal to a closed loop water.

Not really. There's hardly any data points for Haswell. Look at my own Haswell delid test a few posts prior. I couldn't get that low of a voltage either but I'm able to hit 4.5.

 

[Concillian]

My point was not about the Haswells, it was about the Ivy you posted which were not typical results, but pretty exceptional clocking for <1.2v.

 

[Idontcare]

My point was not about the Haswells, it was about the Ivy you posted which were not typical results, but pretty exceptional clocking for <1.2v.

Tough to say because if you look at the screenshot you can see the gflops are not where they are supposed to be which means the CPU isn't pushing the circuits like it is supposed to at that clockspeed...which means it doesn't need the volts and it won't push out the watts which means the temperatures will be lower than in comparison to other IBT tests.

 

[Concillian]

Okay, so either he's giving the "average" Ivy too much credit or Haswell not enough credit.

I think the days of IBT as a real-world stability test are at an end though. You can tell some things from it, but it's starting to get a little "furmark-ey" in that the loads are starting to get to the point where there's a disconnect with real-world loads.

I suppose there is, was, and always will be a difference of opinion on what kind of loading is considered appropriate testing for a "24/7 OC". I suppose also different definitions of stable (24/7/365 stable vs just stable enough to finish 5 non-AVX IBTs after 10 boots so you can take a screenshot and post it.) I guess the takeaway is with various versions of IBT resulting in ~60 - 180 Gflops from the same chip at the same speed, you have to be really careful what conclusions you draw from IBT data.

I always consider stable to be stable enough I'd be comfortable giving my overclocked system to my mother or grandmother. I think a lot of people posting OC results don't agree with that definition of stability, as I always find stable to be WAY higher voltage and/or lower clocks than people post screenshots of. My de-lidded Ivy was stable only "grandma stable" at ~4.0-4.1 at 1.165... not a chance at 4.5. The question for me is sifting through all these results to try and figure where Haswell is "grandma stable". My impression is that it's probably a much quicker transition from "grandma stable" to "doesn't even boot" than Ivy and SB had, but I'm not convinced there's a ton of difference in "grandma stable" clocks from where Ivy was (yet.)

 

[Idontcare]

The benefit of IBT (and Prime95) is that they tell you when the math is coming out wrong. You don't have to wait for a BSOD or corrupted driver issue to surface in order for you to get the picture that maybe things aren't all perfect with your OC.

Kombustor sucks for the same reason. It relies on you and your eyes to be able to detect visual artifacts (a method that will be rife with false negatives) as the sole means of concluding "my OC is solid!"

 

[Concillian]

Yeah, but I was always pretty close to a reasonable heat threshold with IBT near where I reached instability (without AVX). If I set my threshold at 25C to Tjmax, for example, I'd usually hit my temperature limit before a voltage or stability limit. Adding AVX and now Haswell specific instructions pushes heat to the point where it's a limiter before stability... unless you consider 5-10C to Tjmax acceptable...

These are essentially building in margin, but then it's a judgement call anyway depending on how much heat you consider 'acceptable.'

There is DEFINITE value in the quick and dirty limit testing. Super Pi and IBT are both great at that quick limit testing. WAY better than the old Prime95 is all there is days when a quick and dirty test was 1-2 hours. They'll both reveal gross issues very quickly, but as far as finding ultimate stability? Not so great, IMO.

 

[moonbogg]

I think a lot of people posting OC results don't agree with that definition of stability, as I always find stable to be WAY higher voltage and/or lower clocks than people post screenshots of.

Agreed. I've seen folks who swears up and down that their OC is solid, but then when BF3 keeps crashing they blame it on everything but the overclock. It simply can't be the overclock because they somehow "know" that its good. Ask them to torture test the OC and they refuse because its pointless since they already know its good. There are quite a few other things that they simply "know" to be true as well but almost certainly aren't. People seem to come wired that way.

/rant

 

[24HZ]

If you get a specific crash when OC'd and it goes away when at stock then its the OC. Thats a pretty simple way to test. No need to run a litany stress test for hours and hours. Anyone who thinks that's not worth doing to verify is an idiot and shouldn't be taken seriously.

Now thats different from people who run OC'd and know its not 100% stable but they do it anyway because whatever apps they actually use won't crash and isn't critical. Thats fine.

 

[24HZ]

So someone (wink wink IDC) should update the wikipedia articles on IVB and Haswell with respect to the TIM/IHS gap. Right now the info is lacking for Haswell and incorrect for IVB.

http://en.wikipedia.org/wiki/Ivy_Bridge_(microarchitecture)#Thermal_performance_and_heat_issues

 

[crashtech]

I don't think Wikipedia accepts forum posts as information sources. If we could get Anandtech to publish IDC's work in article format, then the wiki could be edited and properly referenced.

 

[studdmufin]

So I have been reading up on delidding and have seen people do it correctly and others break their chips. I'm intrigued in doing this but I am not sure if it going to be worth the risk. Right now I am using a 3770k on a gigabyte up5th with an h80i. It is clocked at 4.7ghz and 1.215 volts with max temps at 83 running prime or occt which I have ran for 18 hours with no errors. I have noticed that core 1 runs hotter than the rest but it is usually around 5C hotter but sometimes up to 10c hotter than the coolest. From what I have read it seems that I got a lucky chip. So is it worth the risk of delidding?

 

[taq8ojh]

Judging by these numbers, I would say you don't need to delid at all.

 

[Idontcare]

So I have been reading up on delidding and have seen people do it correctly and others break their chips. I'm intrigued in doing this but I am not sure if it going to be worth the risk. Right now I am using a 3770k on a gigabyte up5th with an h80i. It is clocked at 4.7ghz and 1.215 volts with max temps at 83 running prime or occt which I have ran for 18 hours with no errors. I have noticed that core 1 runs hotter than the rest but it is usually around 5C hotter but sometimes up to 10c hotter than the coolest. From what I have read it seems that I got a lucky chip. So is it worth the risk of delidding?

1. Welcome to the forums studdmufin :thumbsup:

2. That is indeed a cherry chip, congrats

3. Not worth the risk of delidding, just enjoy what you got and don't look back.

 

[studdmufin]

Cool thanks for the advice guys! I may try to get up to 4.8 late, who knows. Thanks again guys!

 

[Haas360]

Wow, I just spent hours reading all the posts in this thread. Thank you Idontcare, Yuriman, and others for making this a very educational thread. I think I am going to convince my friend to let me delid his cpu and put some Liquid Ultra on it. (I cant to mine because I have the 3930k)

Question though, throughout the thread I have seen a variance of users using Liquid pro vs Liquid ultra. What is your stance on this IDC or Yuriman?

Also, first post, so forgive if I do something wrong lol

-Haas360

ps. Does this Voltage seem a tad high on my cpu?

http://valid.canardpc.com/2679559

 

[Idontcare]

Welcome to the forums Haas360 :thumbsup:

Ha ha, glad to hear you enjoyed it! Definitely a collaborative effort with these types of projects around here, so many awesome contributors

ps. Does this Voltage seem a tad high on my cpu?

http://valid.canardpc.com/2679559

1.5V is basically what I had to put though my 2600K to get it to operate 24/7 stable at 5GHz as well.

Effect of Temperature on Power-Consumption with the i7-2600K

i7-3770K vs. i7-2600K: Temperature, Voltage, GHz and Power-Consumption Analysis

So you are in the right ballpark. As to whether or not it is a good idea to keep it there as a 24/7 OC...no idea if it is a bad idea but I didn't mind doing it to my 2600K.

Question though, throughout the thread I have seen a variance of users using Liquid pro vs Liquid ultra. What is your stance on this IDC or Yuriman?

A fellow member, tw33k, contributed an awesome body of work regarding Liquid Pro vs Liquid Ultra, including going straight to the horse's mouth (the CEO of Coollaboratory) and getting the thermal conductivity data that no one else has officially published to my knowledge.

Coollaboratory Liquid Ultra vs Liquid Pro

Finally got a response from Coollabs...

The heat conductivity value for Liquid Pro is 32,6 w/mk and for Liquid Ultra is 38,4 w/mk.

I knew the figures that people kept quoting were not correct and now we have confirmation (from the CEO no less)

Conclusion: Liquid Ultra is superior (and preferred) over Liquid Pro...but if all you can get is the Liquid Pro then it is only a small loss over using the Ultra, and Pro is second only to the Ultra compared to the rest of the competition.