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

Page 27 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Ferzerp

Diamond Member
Oct 12, 1999
6,437
107
106
We've seen here and there, some eager, young or otherwise less disciplined folk who haven't even read and studied their motherboard manual. You'd even wonder if some would ever bother scanning through Intel's own web-pages, spec-sheets, and other available information.

Frankly, I don't even trust to my own conclusions about some things without soliciting "second opinions" with a well-stated question, and then critiquing those opinions. Heck. That's why I stay plugged in to these forums.

Well, my theory is faulty since VID could at the very most, signal for 1.52V on SB and IB (see edit where I linked the PDF), so I have no clue where 1.55V came from. I "misremembered" the spec.
 

Ed1

Senior member
Jan 8, 2001
453
18
81
A little tweaking and i got voltage a little bit more respectable..

http://valid.canardpc.com/show_oc.php?id=2544243

The thing that gets me is i can do 5ghz with 1.38v but it takes 1.55 for 5.1.. huge jump in Vcore...
That is how it goes when you plot min vcore vs speed . there is a point you don't need any increase (say around 4.2-4.4 ) then as you ramp up speed, vcore needs to have higher slop up after 4.6 or so.
Also the temps goes way up and this means even more vocre if it is to be stable .

I am surprised by your low temps at those speeds and vcore .
 

hokies83

Senior member
Oct 3, 2010
837
2
76
That is how it goes when you plot min vcore vs speed . there is a point you don't need any increase (say around 4.2-4.4 ) then as you ramp up speed, vcore needs to have higher slop up after 4.6 or so.
Also the temps goes way up and this means even more vocre if it is to be stable .

I am surprised by your low temps at those speeds and vcore .
Im Delidded

Cool Labs Liquid pro on the die and Lapped the IHS.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
57
81
My last two cores are actually the same temp (and through all the remounting I did over time, were always so) I'm not sure why. The first two are quite a bit lower (more than 10C for the first)

Just checked last night, and peaks still 68C with IBT.
This is where things get quirky and I'll tell you what I have experienced...on my specific mobo it is not straightforward to get the iGPU to actually power down and remain inactive despite what the BIOS settings are set to.

In my MIVE-Z, as is the case with most people's mobos, the BIOS gives me the option of disabling or enabling the iGPU.

With Sandy Bridge (my 2600K) this bios feature works exactly as one would expect it to work. Set the option in the BIOS to disabled and the iGPU is powered off.

With my Ivy Bridge (3770k), if I set the BIOS to "disable" the iGPU it doesn't actually do that. Not at the functional level at least.

How do I know? Power consumption ends up being 10W higher and the 4th core (the edge core) stays just as hot (it is the hottest actually) as the middle core.

But if I go in the BIOS and enable the iGPU, but do not install the windows drivers for the iGPU (so no lucid software, etc) then my power consumption is 10W lower (be it at idle or at load) and the temperature for the 4th core ends up about 5-6C lower as well.

My working theory is that the BIOS is somehow preventing the PCU on the chip from doing its power-gating thing properly for the iGPU when the BIOS is set to disable the iGPU. But when the bios is set to enable the iGPU then the PCU on the 3770k is able to do its thing which is to powergate the iGPU (since it isn't being used by the OS after all) and it actually powers down and uses essentially no power.

The quirky here is that this is only true when I plug my 3770k into the MIVEZ (and I am using the latest BIOS). When I plug my 2600k into the same MIVEZ then the BIOS features perform as one would expect them to perform.

I've assumed this is an issue unique to the MIVE-Z though because until your post I have never seen anyone else mention observing anything like this. But it sounds to me like you are. So you may try and monkey around with your BIOS settings of enabling vs. disabling the iGPU and see if the 4th core's temperature shifts around.

iGPU disabled in BIOS:



Results in ~90W idle power usage (iGPU is disable in BIOS):



But if I enable the iGPU in the BIOS (but don't bother installing the drivers in windows 7):



Then the power usage drops by >10W to just under 80W and the temperature for Core 4 drops as well (iGPU is enable in BIOS):

 

C.C.

Member
Aug 21, 2012
28
0
0
Im Delidded

Cool Labs Liquid pro on the die and Lapped the IHS.
Can you provide a screenshot showing that you are stable for at least ten runs of IBT @ Xtreme Mode using all your ram? Or 5+ hours of the latest Prime95 build that uses AVX? How about some pictures of your cooling setup?

While I hate to call anyone a liar, I can easily boot into windows and get a valid CPU-Z screenshot @ 5Ghz as well..
 

C.C.

Member
Aug 21, 2012
28
0
0
That is how it goes when you plot min vcore vs speed . there is a point you don't need any increase (say around 4.2-4.4 ) then as you ramp up speed, vcore needs to have higher slop up after 4.6 or so.
Also the temps goes way up and this means even more vocre if it is to be stable .

I am surprised by your low temps at those speeds and vcore .
I must be blind, because I dont see any temps in the link..Also, after looking at your link, you are saying that you are stable @ 5Ghz+ on an H100 with reasonable temps under proper stress testing?

I am running 420mm of rad space with one of the most powerful pumps on the market, along with some of the best fans you can get (Gentle Typhoon AP015s), and I simply do not see it being possible with my setup, let along an "h100":rolleyes:
 

hokies83

Senior member
Oct 3, 2010
837
2
76
Can you provide a screenshot showing that you are stable for at least ten runs of IBT @ Xtreme Mode using all your ram? Or 5+ hours of the latest Prime95 build that uses AVX? How about some pictures of your cooling setup?

While I hate to call anyone a liar, I can easily boot into windows and get a valid CPU-Z screenshot @ 5Ghz as well..
I ran Prime95 for 7 hrs Stable at 1.53v

As u may see me with not much posts here..

Over on OCn i have about 3k and 130rep.. So when i say i ran Prime95 for 7hrs people trust in that cause i got no reason to lie.

I can boot into windows at 5.1ghz with 1.45v but im not claiming that is my Vcore cause it is not.. im having to push my Vcore to it's limits at 1.53v to be stable it is not by any means a great chip.

Also i would claim some adsurd overclock of like 5.5ghz and not just 5.1 lol.
 
Last edited:

rge2

Member
Apr 3, 2009
63
0
0
I have seen at least 50 3770k on 8 different forums that could bench 5 to 5.4 ghz, but yet to see one prime/linx stable at 5ghz, and yet to even hear someone make that claim, using normal ambients/water cooling.

Also if your cpu hits a wall at 5.1, the chance of it being prime stable at that speed is near zero.

My 3770K will do 5.1 bench, spi 32m all day with 1.488 vcore. See pic and screen shot, Im opt33 on ocn.

but prime crashes within a few minutes each time. I would have to see a linx shot that is still under load at 24/25 before I would believe it, given 100s that have tried with better cpus and failed. Not saying it isnt possible, just saying if it doesnt make sense, have to show proof. I have very good temps for a cpu not delidded. But running prime at 1.52vcore at 5ghz, temps reach 105C, have screen shot of such. I even brought my computer in 10c ambients, still no go. I would believe some on xtreme that have recent cherry picked batches doing 5.4 with less vcore than yours and mine screenshots...one of those might get 5ghz prime stable. But if 5.1 is a wall, 5.1 isnt 24/7 use.

mine on ocn 32m runs 5.1 post 160
http://www.overclock.net/t/1247869/official-the-ivy-bridge-stable-suicide-club-guides-voltages-temps-bios-templates-inc-spreadsheet/150#post_17174887

http://valid.canardpc.com/show_oc.php?id=2360164
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,118
57
81
I have to disagree with setting affinity.

The process may not know, but the OS scheduler does and keeps them on the proper logical processors.
I would love for that to be how the theory works out in practice, seriously it would be a godsend in terms of time saved if I didn't have to waste all that time monkeying around with setting the thread affinity, but that just isn't how it works out for me.

If I don't lock the thread affinity here is what I get at 4.8GHz (clockspeed doesn't matter though, of course):



^ no affinity locking, task manager shows the 4 threads bouncing all over the cores, and peak GFlops is 132.4.

But when I affinity lock the threads to the logical cores here is what I get:



^ affinity locking delivers higher GFlops, 134.1 in this case.

Now I've only shown you 3 runs here but that is because it doesn't matter how many runs I do. Whether it is 3 or 10 or 25, locking thread affinity for me always results in higher GFlops at any clockspeed versus leaving the windows 7 scheduler to manage the threads. (I am using win7 x64 ult SP1 with all the latest updates)

And what about temperatures, well if we are getting less work done per second without affinity locking then we would expect there to be lower power consumption and lower core temperatures to come from that.

Here is an example of the peak core temperature (the hottest core as reported by coretemp) with and without affinity locking:



Notice how the peak core temperature is rock steady for the "affinity locked" case but jumps around quite a bit for the "no affinity" case. Also notice that the run with the affinity locked finishes sooner (temperature drops right after 1000s, before the temperature drops for the no affinity case).

Taking a look at the average temperature of all four cores over time depicts the same picture but we see the average temperature for the "no affinity locking" case is even more sporadic:



Note that in both cases the max temperature is the same, but the case with affinity locking sees a sustained level of peak temperature that the run without affinity locking experiences.

Likewise the Kill-a-watt readings, with affinity locking the kill-a-watt reads 238W for this setup. Without affinity locking the kill-a-watt readout will hit 238W very sporadically (once every 5minutes, maybe) but tends to run down around 233-235W - commensurate with the fact that it is doing less work per second so it uses less power per second.

And that is why I go to all the trouble and effort of affinity locking my LinX/IBT run on my hyperthreaded chips - if you want your hyperthread CPU to experience the most challenging thermal environment that LinX has to offer it then you will use 4 threads (not 8) and you will affinity lock those threads to logical (physical) cores and not leave it the scheduler to manage.

In theory the Win7 scheduler might be designed to need no intervention, but in practice, at least for me, it is leaving performance on the table that can be recovered by manually managing the core affinities.
 
Last edited:

Ed1

Senior member
Jan 8, 2001
453
18
81
do you know which cores are the HT ones, you don't want enabled (Duh, you must I assume to be able to set affinity ) .
I never used linX but ever try in prime95 and close all the HT ones. that might be another way around (I don't know if linX has similar close option as prime95)
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,118
57
81
do you know which cores are the HT ones, you don't want enabled (Duh, you must I assume to be able to set affinity ) .
I never used linX but ever try in prime95 and close all the HT ones. that might be another way around (I don't know if linX has similar close option as prime95)
Technically there is not such thing as the "HT core", the HT circuitry itself on any given physical core doesn't come into use until the physical core is processing more than one thread.

In my case it is rather straightforward through experimentation, just affinity lock the every other "core" in task manager and then be on the lookout for parked core isses and move affinity around accordingly.

If you have locked the threads to the wrong cores (accidentally locked two threads to the same physical core, thereby leaving one thread to run on a virtual core via hyperthreading) then you'll know in short order because the GFlops value will tank.

With my mobo/cpu/os I can routinly lock the LinX threads to cores 0/2/4/6. Sometime windows decides to park core 6 in which case I have to affinity lock to core 7 in task manager.
 

Ferzerp

Diamond Member
Oct 12, 1999
6,437
107
106
I would love for that to be how the theory works out in practice, seriously it would be a godsend in terms of time saved if I didn't have to waste all that time monkeying around with setting the thread affinity, but that just isn't how it works out for me.

If I don't lock the thread affinity here is what I get at 4.8GHz (clockspeed doesn't matter though, of course):



^ no affinity locking, task manager shows the 4 threads bouncing all over the cores, and peak GFlops is 132.4.

But when I affinity lock the threads to the logical cores here is what I get:



^ affinity locking delivers higher GFlops, 134.1 in this case.

Now I've only shown you 3 runs here but that is because it doesn't matter how many runs I do. Whether it is 3 or 10 or 25, locking thread affinity for me always results in higher GFlops at any clockspeed versus leaving the windows 7 scheduler to manage the threads. (I am using win7 x64 ult SP1 with all the latest updates)

And what about temperatures, well if we are getting less work done per second without affinity locking then we would expect there to be lower power consumption and lower core temperatures to come from that.

Here is an example of the peak core temperature (the hottest core as reported by coretemp) with and without affinity locking:



Notice how the peak core temperature is rock steady for the "affinity locked" case but jumps around quite a bit for the "no affinity" case. Also notice that the run with the affinity locked finishes sooner (temperature drops right after 1000s, before the temperature drops for the no affinity case).

Taking a look at the average temperature of all four cores over time depicts the same picture but we see the average temperature for the "no affinity locking" case is even more sporadic:



Note that in both cases the max temperature is the same, but the case with affinity locking sees a sustained level of peak temperature that the run without affinity locking experiences.

Likewise the Kill-a-watt readings, with affinity locking the kill-a-watt reads 238W for this setup. Without affinity locking the kill-a-watt readout will hit 238W very sporadically (once every 5minutes, maybe) but tends to run down around 233-235W - commensurate with the fact that it is doing less work per second so it uses less power per second.

And that is why I go to all the trouble and effort of affinity locking my LinX/IBT run on my hyperthreaded chips - if you want your hyperthread CPU to experience the most challenging thermal environment that LinX has to offer it then you will use 4 threads (not 8) and you will affinity lock those threads to logical (physical) cores and not leave it the scheduler to manage.

In theory the Win7 scheduler might be designed to need no intervention, but in practice, at least for me, it is leaving performance on the table that can be recovered by manually managing the core affinities.

Odd, I'll have to check when I return home later in the week. I have to leave for the airport now and won't leave this PC on for any remote access.

Ack, ok. Ran a quick visual test which agrees with what you're saying.

Now I'm going to have to make sure 2k8r2 isn't doing the same thing. I know it doesn't with massive scales of procs (because the HT "cores" are always tagged as parked), but I'll have to check on some of the mid-range count systems.
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,118
57
81
All I can figure is that this is such a niche corner case where the scheduler is still leaving performance on the table but Microsoft isn't going to prioritize eradicating the gap.

Consider that we are talking about only ~1% performance difference. How many people are going to worry about that for a nich corner app that is essentially a power virus?

Still though, if they ever do address (will have to see in Win8) then it would make me and about 7 other people happy(ier) customers. The rest of the world won't even notice.
 

Ferzerp

Diamond Member
Oct 12, 1999
6,437
107
106
All I can figure is that this is such a niche corner case where the scheduler is still leaving performance on the table but Microsoft isn't going to prioritize eradicating the gap.

Consider that we are talking about only ~1% performance difference. How many people are going to worry about that for a nich corner app that is essentially a power virus?

Still though, if they ever do address (will have to see in Win8) then it would make me and about 7 other people happy(ier) customers. The rest of the world won't even notice.
Honestly, I rarely see a need for it. Most of our Windows OS instances these days are virtual and there is no visibility for the guest OS for the physical layout of the hardware. It gets the correct moddel number, but it has no concept of where its threads are actually executing. It's only in the case of extremely niche applications (we're talking 40c/80t systems that hit 70-80% CPU utilization) that we back off on virtualization. Those are the ones that I see with every other logical core parked most of the time.
 

PCWargamer

Junior Member
Oct 16, 2012
10
0
0
I have seen at least 50 3770k on 8 different forums that could bench 5 to 5.4 ghz, but yet to see one prime/linx stable at 5ghz, and yet to even hear someone make that claim, using normal ambients/water cooling.

Also if your cpu hits a wall at 5.1, the chance of it being prime stable at that speed is near zero.

My 3770K will do 5.1 bench, spi 32m all day with 1.488 vcore. See pic and screen shot, Im opt33 on ocn.

but prime crashes within a few minutes each time. I would have to see a linx shot that is still under load at 24/25 before I would believe it, given 100s that have tried with better cpus and failed. Not saying it isnt possible, just saying if it doesnt make sense, have to show proof. I have very good temps for a cpu not delidded. But running prime at 1.52vcore at 5ghz, temps reach 105C, have screen shot of such. I even brought my computer in 10c ambients, still no go. I would believe some on xtreme that have recent cherry picked batches doing 5.4 with less vcore than yours and mine screenshots...one of those might get 5ghz prime stable. But if 5.1 is a wall, 5.1 isnt 24/7 use.

mine on ocn 32m runs 5.1 post 160
http://www.overclock.net/t/1247869/official-the-ivy-bridge-stable-suicide-club-guides-voltages-temps-bios-templates-inc-spreadsheet/150#post_17174887

http://valid.canardpc.com/show_oc.php?id=2360164

@IDC – great thread. I have read through every page and enjoyed your careful testing and the informative input from you and others. This thread is the reason I have decided to register on these forums. I hope to try deliding this weekend if the CL Ultra arrives by then.


@ rge2, I get what you are saying and I can understand your being skeptical as hokies83 has yet to provide any 5.1 stability proof, but I do not think he saying that his top OC is 5.1, but that he is having trouble at 5.1 with being stable at vcore lower than 1.5+v and wanting to get ideas from this thread on how to get his vcore lower. (hope so anyway)

But what caught my eye the most about your post was that you had yet to find any prime stable IBs on multiple forums, and as you note, it is hard to find if there are any 5GHz IBs that are prime stable on air/water, but there are some, which can be encouraging to us all.

And interesting enough, two can even be found on the thread link you posted with your OCN 32m runs at 5.1, as that thread also contains several 4.8 & 4.9 IB prime stable runs of 12-24hrs, but also two 5.0 prime runs on the first page spreadsheet:

http://www.overclock.net/t/1247869/official-the-ivy-bridge-stable-suicide-club-guides-voltages-temps-bios-templates-inc-spreadsheet

alex_tpc 5000.4mhz 1.312v 24hrs 80-89-89-83 WATER - Swiftech XT CPU / Custom
Arkaridge 5001.1mhz 1.376v 13hrs 88-92-96-89 AIR - NZXT Havik 140

Alex_tpc has a 3770K, and Arkaridge has a 3570K. And great chips too by the vcors, and one is on air. And by the vcors and temps I do not think either of these are delided, although several on that thread are. So the good news is that there are at least two IBs that are prime stable at 5.0 on just that forum!

Also, I do not think hokies83 is lying, but it would sure clear things up if he would post something as proof for all of you to see to remove that doubt. And (sadly) I guess there are enough people on forums today to make all of us skeptical of such claims without sufficient proof.
 

rge2

Member
Apr 3, 2009
63
0
0
In time it took to post that, he could of posted a screenshot of a linx run so we could all learn something, ie if his delidded chip helps him to actually get it stable there or not.

I was referring to 3770k, but you found 1, and like on xtreme where I said it was likely some cherry cpus could do it, it is with a cherry cpu only requiring vcore of 1.312 for 5ghz. But thanks, I did not see it posted there even a week ago, though run was from a while back.

But those requiring 1.5ish vcore and have a wall at 5.1 like mine and his, I would have to see linx during load to believe it. And again, I did not say he was lying or even that it wasnt possible, just said would have to see it to believe it since it would be very unusual. And he had 5.1 24/7 in his sig implying stable for 24/7 use, just I have never seen a 24/7 stable cpu at the limit/wall of stability.
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,118
57
81
I think the point here is that extraordinary claims bear the burden of requiring extraordinary proof. Its just the way it is.

Come in here claiming your 3770k takes 1.5V to be 5GHz LinX stable and people will take you at your word because you are not making an extraordinary claim. (mine take 1.45'ish volts at 5GHz, I'm in the same boat as most)

But come in to the AT forums claiming you get 5GHz on <1.4V or even higher than 5GHz and now you are making extraordinary claims, and should realize that you are doing so at the time, and just accept the fact that your extraordinary claims are going to be met with extra scrutiny and suspicion until you provide extraordinary proof of your claims.

What part of life doesn't work like this? It is common, and it makes sense. Hey, someone should make a phrase out of that. Something like "sense common" or maybe some other linear combination of those two words...:p
 

PCWargamer

Junior Member
Oct 16, 2012
10
0
0
I think the point here is that extraordinary claims bear the burden of requiring extraordinary proof. Its just the way it is.

Come in here claiming your 3770k takes 1.5V to be 5GHz LinX stable and people will take you at your word because you are not making an extraordinary claim. (mine take 1.45'ish volts at 5GHz, I'm in the same boat as most)

But come in to the AT forums claiming you get 5GHz on <1.4V or even higher than 5GHz and now you are making extraordinary claims, and should realize that you are doing so at the time, and just accept the fact that your extraordinary claims are going to be met with extra scrutiny and suspicion until you provide extraordinary proof of your claims.

What part of life doesn't work like this? It is common, and it makes sense. Hey, someone should make a phrase out of that. Something like "sense common" or maybe some other linear combination of those two words...:p

Can't say I can disagree with either of you. Extraordinary claims do need some conformation. I wish he had produced them myself.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
57
81
@IDC – great thread. I have read through every page and enjoyed your careful testing and the informative input from you and others. This thread is the reason I have decided to register on these forums. I hope to try deliding this weekend if the CL Ultra arrives by then.
Welcome to the forums PCWargamer! :thumbsup:

Let us know how it goes, both the delidding as well as the results with CL Ultra. Very curious to hear more about CL Ultra.
 

PCWargamer

Junior Member
Oct 16, 2012
10
0
0
Welcome to the forums PCWargamer! :thumbsup:

Let us know how it goes, both the delidding as well as the results with CL Ultra. Very curious to hear more about CL Ultra.
Thanks IDC.

I too am interested in how the Ultra will do. Most of the delidders on OCN seem to use CL PRO, others IC Diamond, and a few various other TIMS. We have started a delidders thread a few weeks ago to try to assist in informing those who are interested in delidding, and also try to capture some info like TIM choice and temp effects.

http://www.overclock.net/t/1313179/delidded-ivy-bridge-club

I have included a link to your thread on the first page of that thread as the info on here is very good, as well as your important finding that the IB temp problems have more to do with the die to IHS gap than Intel&#8217;s TIM choice.

Not enough data points have been gathered yet to determine much of anything, or even if Ultra is similar to PRO, but it seems PRO lowers temps by ~25C (YMMV). I am using Ultra to see how it does on my system. We hope more data will be collected over time to (maybe) tell us something.

But no one is trying different TIMs on the same system to get more accurate results. I was wondering if such an undertaking was going to take place on this thread now or in the future? I know lots of delidding OC&#8217;ers would love to have some systematic results of TIM choices! :)
 
Last edited:

Yuriman

Diamond Member
Jun 25, 2004
5,530
141
106
Just ordered a tube of CL Ultra, will update with results when it arrives.
 

ASK THE COMMUNITY