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

[Maseo]

Hey Folks, New here to AnandTech but been following this delidding topic for a while now.

Here´s a thing i´ve been going nuts about:
I recently upgraded to a Sabertooth Z77 because of a dying P8P67, slapped everything back on the new board, fired it up, same old clocks / speeds as on the P8 and, voila... right?

Well here's the curveball: Using the same clocks, with LOWER vcore compared to the P67, the temps are all over the place. Idle is just fine and dandy, but the load temperatures are nuts. The P67 did 4.8 with 1.45-1.47 Vcore, where the Z77 does 1.39-1.141 for the same speed. I'm using a custom WC-loop.
Using LinX it goes to 75-100-105-99 where, as you might guess, throttling takes over, and where i stop the process short. The strange thing is though, on the P67 board it would pass 90-92° on the hottest core, using the same CPU cooling setup. At stock (100 x 35) LinX holds up to 48-60-60-60 as max temperatures with a vcore of 1.104 (CPUID)


I've tried cleaning the CPU many times, reapply TIM, with or without retention socket, but all to no avail.

What in heavens name could be going on here? Does a sensor malfunction? Anything else to look at? I'm at a loss..

 

[Idontcare]

Welcome to the forums Maseo :thumbsup:

Hey Folks, New here to AnandTech but been following this delidding topic for a while now.

Here´s a thing i´ve been going nuts about:
I recently upgraded to a Sabertooth Z77 because of a dying P8P67, slapped everything back on the new board, fired it up, same old clocks / speeds as on the P8 and, voila... right?

Well here's the curveball: Using the same clocks, with LOWER vcore compared to the P67, the temps are all over the place. Idle is just fine and dandy, but the load temperatures are nuts. The P67 did 4.8 with 1.45-1.47 Vcore, where the Z77 does 1.39-1.141 for the same speed. I'm using a custom WC-loop.
Using LinX it goes to 75-100-105-99 where, as you might guess, throttling takes over, and where i stop the process short. The strange thing is though, on the P67 board it would pass 90-92° on the hottest core, using the same CPU cooling setup. At stock (100 x 35) LinX holds up to 48-60-60-60 as max temperatures with a vcore of 1.104 (CPUID)


I've tried cleaning the CPU many times, reapply TIM, with or without retention socket, but all to no avail.

What in heavens name could be going on here? Does a sensor malfunction? Anything else to look at? I'm at a loss..

I can think of a few possibilities here, but it is just speculation at best on my part.

For starters, the reported voltage versus the actual voltage can vary widely from mobo to mobo, even within the same make and model. But the reported sensor temps should not, because the temperature sensor itself remains the same even when you change mobo's (because the sensor is on the CPU).

Is your CPU delidded? I ask because the voltages you are reporting using for 4.8GHz operation are really quite high.

The one concern I have is you mention a dying mobo as your motivation for changing mobos...it is possible that the dying mobo was doing bad things to your cpu (degrading it prematurely) and now you are seeing the result of that accelerated degradation. Or it is possible that your particular CPU did not fare well under the voltage duress you applied to it, and as the CPU degraded it took out your mobo as part of collateral damage.

In all liklihood the temperatures are real but not a product of the new mobo per se, I suspect you will observe similar if you put the same CPU into a different mobo.

At this point you are probably just going to have to dial in a new safe OC that keeps your voltages and temperatures reasonable.

Remember, the CPU was only designed to experience high temperatures when the voltage was kept at stock. When you overvolt, your safe max temperature decreases to a value that is well below TJmax.

Running at 1.4V and close to TJmax is degrading your processor faster than it was intended to degrade, and in your case it may have accelerated the timeline quickly enough to have already degraded the cpu to the extent you are observing.

If that is true then the only course of action for you to take is to back off on the clockspeed and voltage and try to keep your CPU from degrading any faster (or it will be useless to you far sooner than you were anticipating).

 

[Maseo]

Welcome to the forums Maseo :thumbsup:

I can think of a few possibilities here, but it is just speculation at best on my part.

For starters, the reported voltage versus the actual voltage can vary widely from mobo to mobo, even within the same make and model. But the reported sensor temps should not, because the temperature sensor itself remains the same even when you change mobo's (because the sensor is on the CPU).

Is your CPU delidded? I ask because the voltages you are reporting using for 4.8GHz operation are really quite high.

The one concern I have is you mention a dying mobo as your motivation for changing mobos...it is possible that the dying mobo was doing bad things to your cpu (degrading it prematurely) and now you are seeing the result of that accelerated degradation. Or it is possible that your particular CPU did not fare well under the voltage duress you applied to it, and as the CPU degraded it took out your mobo as part of collateral damage.

In all liklihood the temperatures are real but not a product of the new mobo per se, I suspect you will observe similar if you put the same CPU into a different mobo.

At this point you are probably just going to have to dial in a new safe OC that keeps your voltages and temperatures reasonable.

Remember, the CPU was only designed to experience high temperatures when the voltage was kept at stock. When you overvolt, your safe max temperature decreases to a value that is well below TJmax.

Running at 1.4V and close to TJmax is degrading your processor faster than it was intended to degrade, and in your case it may have accelerated the timeline quickly enough to have already degraded the cpu to the extent you are observing.

If that is true then the only course of action for you to take is to back off on the clockspeed and voltage and try to keep your CPU from degrading any faster (or it will be useless to you far sooner than you were anticipating).

Thanks for the response IDC, appreciate it!

Yes, it is delidded. The IHC has been sanded as well, but even die-mounted the temps are high. It's also fairly new, about 1 to 2 months.
Any lower Vcore and LinX starts to give errors and programs shutdown / crash.

Right after i delidded it my temps improved (this was on the old board). With AVX enabled it was sub 90, mid 80's on the hottest core, with a higher vcore.
This cpu hasn't been real keen on low voltages. i've settled for the fact that it needed more juice to keep stable at these high(er) clocks, where the temperatures would allow it to.

The exchange of the mobo's was due to the fact that two out of four RAM slots malfunctioned out of nowhere, and now is up for RMA as ASUS said i can send it back and they will examine it.
The strangest thing is though that, with the mobo swap, the cpu temps went up as well, that's the only coïncidence i have here.

The thing that bugs me the most is that, even with a stable stock clock, the temps are awefully high. I ordered some PK-3 and Ultra, as a final ditch effort, to see if it helps..

Edit:
4.7 @ 1.35v - Still hits 72-101-105-97. Somethings off here.. just hope it isn't already dying. I've seen better temperatures on air rigs..

 

[tw33k]

tw33k, What were you running to load the cpu with? For apple/apples comparison, you should stick to what's been used by others in this this thread - prime95 and ibt.

I don't use Prime to test stability and never will. Aida64 is much more reliable because it is closer to real world usage. I've lost count of the number of times I've seen people run Prime for 12 hours+ and think their system is stable only to have it crash after gaming for a while. I've found that if my system can run Aida64 then it will run anything I will ever throw at it and I've been using it for a lot of years (back when it was Everest)

 

[Maseo]

Welcome to the forums Maseo :thumbsup:



I can think of a few possibilities here, but it is just speculation at best on my part.

For starters, the reported voltage versus the actual voltage can vary widely from mobo to mobo, even within the same make and model. But the reported sensor temps should not, because the temperature sensor itself remains the same even when you change mobo's (because the sensor is on the CPU).

Is your CPU delidded? I ask because the voltages you are reporting using for 4.8GHz operation are really quite high.

The one concern I have is you mention a dying mobo as your motivation for changing mobos...it is possible that the dying mobo was doing bad things to your cpu (degrading it prematurely) and now you are seeing the result of that accelerated degradation. Or it is possible that your particular CPU did not fare well under the voltage duress you applied to it, and as the CPU degraded it took out your mobo as part of collateral damage.

In all liklihood the temperatures are real but not a product of the new mobo per se, I suspect you will observe similar if you put the same CPU into a different mobo.

At this point you are probably just going to have to dial in a new safe OC that keeps your voltages and temperatures reasonable.

Remember, the CPU was only designed to experience high temperatures when the voltage was kept at stock. When you overvolt, your safe max temperature decreases to a value that is well below TJmax.

Running at 1.4V and close to TJmax is degrading your processor faster than it was intended to degrade, and in your case it may have accelerated the timeline quickly enough to have already degraded the cpu to the extent you are observing.

If that is true then the only course of action for you to take is to back off on the clockspeed and voltage and try to keep your CPU from degrading any faster (or it will be useless to you far sooner than you were anticipating).

Well, i've got my PK-3 in, and guess what?

@ 4.9Ghz, now my max temps level out at 59-75-75-73 @ 1.43Vc. That other paste i had (genuine white thermal grease) had to be UTTERLY crap. I'm happy again.

 

[Idontcare]

Well, i've got my PK-3 in, and guess what?

@ 4.9Ghz, now my max temps level out at 59-75-75-73 @ 1.43Vc. That other paste i had (genuine white thermal grease) had to be UTTERLY crap. I'm happy again.

Very glad to hear it, and those temps are nice :thumbsup:

 

[chris89]

So where can we find a 216 mm2 copper shim? I need to get some better cooling out of my 2720qm which in turbo boost at 3Ghz. I pulled the blue pin from the fan so it runs at 4,700rpm continuously and runs a lot cooler than stock. However contact is poor and temps almost hit 90C while in game... I used to be able to pull off high 70s under full load in turbo boost 3Ghz. Now I'm having a hell of a time getting some solid temps...

Any thoughts on where to get the shim that will work on our 216 mm2 die size?

thanks

 

[notty22]

A Grainger would have it.

http://www.grainger.com/Grainger/ec...r+shim?Ndr=textsearchesinbase+true&sst=subset

 

[chris89]

Since it's rolled it won't be 100% perfectly flat...

 

[notty22]

That is what newspaper and other pressman use for plate registration (one use) it can be cut , at those thicknesses should form to function.

http://www.grainger.com/Grainger/Sheet-Metal-5MWN3?cm_sp=IO-_-IDP-_-BTM_BTB05209020&cm_vc=IDPBBZ2

 

[chris89]

So this sheet is 100% perfectly flat? Not rolled up at all? I may want to find a way to compress my finished shim to be sure it's 100% perfectly flat because otherwise all hell will surely break loose lmao

 

[BonzaiDuck]

Very glad to hear it, and those temps are nice :thumbsup:

At this point, and to save you the trouble for anything lengthy, I will say "Marvelous temps!" but the voltage seems excessive for 24/7 operation. You will say "Intel is more conservative in published specs than the testing shows they need be," or "the processors will last longer at higher voltages than had previously been thought."

I know the data have already been posted by other writers here. What's the estimate or expected voltage at 4.7 and 4.8? How close does it come to -- "say" -- 1.30V? Or even 1.35V? Well -- I'll go back and check your earlier posts and those of some others. Seemed to me you could get 4.7 at > 1.298V, though. I'll bet the temperatures on Maseo's rig would be insanely tame on air or water . . .

 

[chris89]

On my i7-930 at 4.4Ghz and 1.4v i get like 80C on intelburntest and thats with a special single use thermal interface from I forgot the manufacturer but it was $25 for 2 applications worth. At 4.8Ghz I can hit 100C. Running on the 1st application 3 years later still sitting at the same temps. Running a Cocage true spirit lapped and core i7 lapped and 120cfm fans... Not bad if u ask me... 4.2Ghz is more ideal hitting 75C on intelburntest though

 

[Idontcare]

So where can we find a 216 mm2 copper shim? I need to get some better cooling out of my 2720qm which in turbo boost at 3Ghz. I pulled the blue pin from the fan so it runs at 4,700rpm continuously and runs a lot cooler than stock. However contact is poor and temps almost hit 90C while in game... I used to be able to pull off high 70s under full load in turbo boost 3Ghz. Now I'm having a hell of a time getting some solid temps...

Any thoughts on where to get the shim that will work on our 216 mm2 die size?

thanks

I don't understand why you want the shim. Is this for a laptop?

Desktop's don't need the shim. I tried all kinds of shims, metal shims, paper shims, plastic shims, and they are unnecessary.

At this point, and to save you the trouble for anything lengthy, I will say "Marvelous temps!" but the voltage seems excessive for 24/7 operation. You will say "Intel is more conservative in published specs than the testing shows they need be," or "the processors will last longer at higher voltages than had previously been thought."

I know the data have already been posted by other writers here. What's the estimate or expected voltage at 4.7 and 4.8? How close does it come to -- "say" -- 1.30V? Or even 1.35V? Well -- I'll go back and check your earlier posts and those of some others. Seemed to me you could get 4.7 at > 1.298V, though. I'll bet the temperatures on Maseo's rig would be insanely tame on air or water . . .

I was getting 4.8GHz at just a smidge over 1.3V (1.316V to be exact).

How long will it last? Not long is my expectation. Maybe 2yrs, possibly 3yrs. Depends on the duty cycle at that voltage and clockspeed.

If someone overclocks their rig to say 5GHz on 1.5V but they mostly use it for web surfing and say 2-3hrs of gaming per day then the CPU's duty cycle is going to be extremely light and the CPU would probably last 5-10yrs at that setting.

If you OC to the same level but run DC or prime95 24x7 then the CPU is going to degrade at a much faster rate owing to the 10x increase duty cycle. Your 5 yrs lifespan on light duty cycle will be reduced to 6 months in that situation.

 

[BonzaiDuck]

I was getting 4.8GHz at just a smidge over 1.3V (1.316V to be exact).

How long will it last? Not long is my expectation. Maybe 2yrs, possibly 3yrs. Depends on the duty cycle at that voltage and clockspeed.

If someone overclocks their rig to say 5GHz on 1.5V but they mostly use it for web surfing and say 2-3hrs of gaming per day then the CPU's duty cycle is going to be extremely light and the CPU would probably last 5-10yrs at that setting.

If you OC to the same level but run DC or prime95 24x7 then the CPU is going to degrade at a much faster rate owing to the 10x increase duty cycle. Your 5 yrs lifespan on light duty cycle will be reduced to 6 months in that situation.

I'm assuming your longevity estimates are based on the 5Ghz/1.5V configuration. So it would seem that 4.7 or 4.8 can be a reasonable expectation.

 

[chris89]

I don't understand why you want the shim. Is this for a laptop?

Desktop's don't need the shim. I tried all kinds of shims, metal shims, paper shims, plastic shims, and they are unnecessary.

Yeah it's for a 2nd gen 2720qm under intelburntest I hit like 85C and I'd like to get it down to around 75-80C tops and I've seen it before but on a flawlessly re-paste.

 

[Idontcare]

Yeah it's for a 2nd gen 2720qm under intelburntest I hit like 85C and I'd like to get it down to around 75-80C tops and I've seen it before but on a flawlessly re-paste.

OK but I still don't get why you need/want a metal shim? My DELL laptop was overheating (throttling) and it turned out the thermal pad had deteriorated.

I replaced the pad with some NT-H1 and the existing heatsink apparatus screwed back down and sits right on the die without any gap issue.

Does yours have some sort of stand-off that is creating a gap that must be filled with the shim?

 

[chris89]

OK but I still don't get why you need/want a metal shim? My DELL laptop was overheating (throttling) and it turned out the thermal pad had deteriorated.

I replaced the pad with some NT-H1 and the existing heatsink apparatus screwed back down and sits right on the die without any gap issue.

Does yours have some sort of stand-off that is creating a gap that must be filled with the shim?

Yeah there's a gap because I can't tighten it down enough to get good contact like I can on the gpu for some reason... It's quite annoying and a shim would help it a lot.

What's the exact dimensions of a 216mm2 die size? Like H-W-L in millimeters? I found a guy who will cut one for me perfectly to fit my 2nd gen core i7 mobile cpu for really cheap on $5-10 on ebay if you guy's are interested I can tell you the seller name so he could make one for you guy's and whatever exact dimensions you require.

 

[Idontcare]

Yeah there's a gap because I can't tighten it down enough to get good contact like I can on the gpu for some reason... It's quite annoying and a shim would help it a lot.

What's the exact dimensions of a 216mm2 die size? Like H-W-L in millimeters? I found a guy who will cut one for me perfectly to fit my 2nd gen core i7 mobile cpu for really cheap on $5-10 on ebay if you guy's are interested I can tell you the seller name so he could make one for you guy's and whatever exact dimensions you require.

When I made my shims I just cut the metal with a bulky pair of scissors. It actually cuts just fine, like cutting thick paper.

The very edge of the cut will be turned up slightly, so you just cut it larger than the die itself so the turned up area hangs over the edge of the die and isn't a practical issue.

There's no need to have it cut to perfect size, the IHS on a desktop CPU isn't sized to be perfect either. You just don't want the shim so large that you can get it to sit inside the region that the CPU's PCB occupies.

Also, for this application you don't need perfectly flat shim stock either. You can use the rolled type because it is going to become flat under the compression forces of the cooler anyways.

I bought a HUGE roll of metal shim stock for something like $6 or $11 (is was on par with the cost of a tube of tim) and it worked fine. Flattens out ok.

 

[chris89]

When I made my shims I just cut the metal with a bulky pair of scissors. It actually cuts just fine, like cutting thick paper.

The very edge of the cut will be turned up slightly, so you just cut it larger than the die itself so the turned up area hangs over the edge of the die and isn't a practical issue.

There's no need to have it cut to perfect size, the IHS on a desktop CPU isn't sized to be perfect either. You just don't want the shim so large that you can get it to sit inside the region that the CPU's PCB occupies.

Also, for this application you don't need perfectly flat shim stock either. You can use the rolled type because it is going to become flat under the compression forces of the cooler anyways.

I bought a HUGE roll of metal shim stock for something like $6 or $11 (is was on par with the cost of a tube of tim) and it worked fine. Flattens out ok.

thanks man that's what I thought so whats the dimensions of 216mm2? like Length in mm and width in mm? the guy is going to ship me the piece of pure 100% copper cut to the exact size I want for $5 shipped... I appreciate your help a lot.

 

[Idontcare]

thanks man that's what I thought so whats the dimensions of 216mm2? like Length in mm and width in mm? the guy is going to ship me the piece of pure 100% copper cut to the exact size I want for $5 shipped... I appreciate your help a lot.

That I don't know, but why don't you just open the laptop and remove the HSF cooler, exposing the CPU die and PCB and just measure it out with a ruler? Then you'll know exactly how much room for margin you have given yourself in terms of the overhang and there won't be any mystery.

If I knew the dimensions of your chip I would tell you, but there are a lot of length x width dimensions of a rectangle that will give you the same area, so me guessing at length and width won't help you. But you can look at the chip and measure it since you have the chip in hand so to speak.

 

[chris89]

lol i would totally do that but i just removed it yesterday and cleaned everything and applied a super thin layer to the cpu and gpu and used a shim on the gpu. The shim dropped the temps on the gpu by 1-2C not much but it's something. The cpu hits 85C+ and when I ran intel burn test right away core 1 hit 89C and i hate seeing temps that high. Maybe the laptop's cooling capabilities is lacking. Maybe the heatsink isn't working efficiently enough. I have the fan modded to 5V and it runs at 4,700rpm continuously. I also cut a hole under the laptop so the fan can suck fresh cool air and that alone dropped temps 10-15C. Maybe I have dust in the cooler and it's reducing it's airflow. I'll look into it as soon as I get the new shim. Maybe I'll oil the fan to allow it to spin a tad bit fast maybe 4,750-4,800rpm. I've done it before and seen 50-100rpm increase with a nice decrease in bearing noise which is nice.

I told him to make it 30mm long and 15mm wide to be on the safe side so hopefully it will work if not ill just trim it up a little. One thing that reduces thermal conductivity is the core contact isn't copper if I remember correctly it's like aluminum welded to the copper pipe and that's not good, same for the gpu. If it was 100% copper construction it would run much cooler...

I plan on getting a new laptop anyway. I've been looking into one with a GeForce GTX670MX 3GB gpu which seems to devasatate among mobile gpu's and going for a 2670qm in it to reduce cost it's like 1,200 with the 2670qm or it's not even $100 more for the 3rd gen core i7. I mean I have found the core i7 even 2nd gen not even in turbo boost is far from bottlenecking the gaming performance. It's mostly gpu power.

I appreciate your help alot guy's and u especially Idontcare.

 

[Ed1]

I don't understand why you want the shim. Is this for a laptop?

Desktop's don't need the shim. I tried all kinds of shims, metal shims, paper shims, plastic shims, and they are unnecessary.

I was getting 4.8GHz at just a smidge over 1.3V (1.316V to be exact).

How long will it last? Not long is my expectation. Maybe 2yrs, possibly 3yrs. Depends on the duty cycle at that voltage and clockspeed.

If someone overclocks their rig to say 5GHz on 1.5V but they mostly use it for web surfing and say 2-3hrs of gaming per day then the CPU's duty cycle is going to be extremely light and the CPU would probably last 5-10yrs at that setting.

If you OC to the same level but run DC or prime95 24x7 then the CPU is going to degrade at a much faster rate owing to the 10x increase duty cycle. Your 5 yrs lifespan on light duty cycle will be reduced to 6 months in that situation.

those are pretty low min Vcore , those lower than 0.7 at say 1600mhz , idle times were stable ?
I kind of thought once you get in low or below 0.8v stability goes down, random lockup etc .

 

[Idontcare]

those are pretty low min Vcore , those lower than 0.7 at say 1600mhz , idle times were stable ?
I kind of thought once you get in low or below 0.8v stability goes down, random lockup etc .

Those are manual voltages, not offset voltages. In other words the voltages are the same regardless whether the CPU is idle and loaded in that graph.

 

[Ed1]

Those are manual voltages, not offset voltages. In other words the voltages are the same regardless whether the CPU is idle and loaded in that graph.

Ok, so that helps with stability or was it only stable under loads .
Maybe your saying there is problem with low idle voltage because of speedstep/turbo and how MB manages voltage dynamically ?

I always hear issues with 35xx/37xx when you try to lower to much, this is with offset and speedstep/turbo enabled .