Separate names with a comma.
Discussion in 'CPUs and Overclocking' started by Idontcare, Aug 3, 2012.
What program are you using to monitor the temperature?
I wonder how/why that happened?
In theory a 100W lightbulb burns a 100W regardless the light fixture, provided it is providing the correct voltage to the bulb of course.
Your CPU should run at the same temperature, use the same power, same voltage, etc regardless the mobo (the mobo itself won't be at the same temp, or use the same power, of course).
To have a CPU drop 12C just in relocating from Mobo A to Mobo B...that sounds more like a questionable HSF mount on the CPU when it was installed in Mobo A to me, versus an impressive double-digit decrease.
If the HSF mounts are not in question, then the CPU voltages should be. My MIVE-Z changes the "default" voltage for my 2600K depending on which BIOS I install. An earlier BIOS defaulted to 1.46V D: (verified by voltmeter)
Naturally power usage was through the roof as well. Thankfully ASUS finally got that fixed with one of the later BIOS revs for the MIVEZ.
Otherwise the temperature of your CPU should be mobo-agnostic. The mobo may be screwing up the voltage delivered, or messing up the power-states (bad BIOS is bad), but if it is operating to spec then you shouldn't be able to see a difference in CPU temp or power usage.
Promising results: I'm at around the 18 hour mark since I mounted my waterblock directly to the die and temps have come down from where they were by about 4c. Currently Prime'ing at 4.4ghz (CPU-Z reported voltage is ~1.200) and my hottest core is hovering around 73-74c, and the coolest around 65c. Before delidding, the hottest typically ran around 78-79c with the same settings. Ambient isn't more than 1c different than it was yesterday.
An experienced overclocker posted a guide to IB OCing here. One thing he said was, "When overclocking on air the only two voltages you should need to touch on an Ivy Bridge setup are the Vcore (which you increase) and the CPU PLL( which can be decreased to help temperatures)." He is also using a Gigabyte board.
He seems to have published a copy or near copy here.
That's far too hot for direct die. My system doesn't even pass 68C on the hottest core with ~1.24V and IBT with only an H100 and the IHS still on (but 0 gap, as I sanded the thing down and scraped all the black goop off within about 2 weeks of them being released). Check your mounting or something.
The mount is fine. I'm basically almost out of ideas... the next item on my list is to lap the waterblock itself. It has a fantastic looking finish and it looks flat when I tested it with a razor, but who knows, maybe there's some old paste baked into it that's hurting its ability to transfer heat away?
My liquid temps are fine, and I have a GPU in the loop too that doesn't get above around 40c fully loaded (using the same thermal paste too!) so that rules out other issues like flow rate.
Bum chip? Worth trying out some Liquid Pro or Liquid Ultra. It's not nearly as hard to apply as people make it out to be.
There's always the unlikely possibility that he has one or many core thermal sensors that are out of whack in the same direction from the "real" temperatures. Or -- that the temperatures are not being read properly at the BIOS level. And I remember the fiasco about Q6600's TjMax.
I'm idling at approximately 26-27c with a room temperature of around 24c, so if it's a matter of the sensors being off they aren't off linearly.
Easy to check, let your CPU overheat while watching the clockspeed with Realtemp or CPUz. You know the chip will throttle when it hits TJmax, so watch to see when it throttles.
The thing with temperature is that from our side of the equation there are only two things that are going to determine temperature - cooling (heat loss) and power-consumption (heat generation).
If this CPU is hitting these kinds of temperatures despite the improvements in cooling then it must be consuming prodigious amounts of power.
Yuriman, do you have a killawatt to measure power usage?
Unfortunately no. I used to use a UPS but the battery died a few months back and I haven't replaced it.
I tried to determine power using by limiting my TDP via the Intel software a few weeks ago and came up with these numbers (for what it's worth):
4.4ghz ~ 92w
4.5ghz ~ 104w
4.6ghz ~ 120w
I'll see about getting a meter this afternoon.
IDC, Very true assessment. I did not look into it too much but I think there could be many variables to this. First of all the heatsinks are much larger and the board is more spacious on the Gigabyte, more power phases and a true digital power phase system unlike the asrock. For some reason the asrock board held load voltages for longer periods of time and for some reason my dimms were hot to the touch unlike this gigabyte board. It could be that my memory was spewing heat inside the HSF since I did remount 4 times on the Asrock board.
I do believe that having a good motherboard layout can have a small effect on temps.
It is not surprising that people get different temp improvements when delidding, if a lot of the effect is the gap, that gap is going to vary hence effect should vary. Not only from manufacturing variance in IHS, but the IHS is attached with an adhesive, and that adhesive definitely going to vary in thickness, hence cause corresponding variance in size of gap.
Although pudget systems delidded and tested at stock volts, which will show the least effect, they did state one cpu got 10C better temps by delidding, another only 2C, even though they have a dog in race to show as little benefit as possible. Their conclusion was bad tim application by intel on 1 cpu, but more likely just large gap on one, smaller gap on the other.
Had they tested at overclocking vcore, that 10C would have been closer to what IDC got, and the 2C would have been higher but maybe closer some others that have listed 7-8C better temps.
Yurimans might have been the other extreme, where his gap was very small to begin with, plus testing at lower range 1.2vcore, getting only 4C improvement. But the one issue with that, is like others, those temps are pretty high. I turned off my HTing so running 4 cores, and with 4.4ghz, 1.2 load vcore cpuz ran prime small ffts for 30 mins (I know Yuriman used blend, but I was trying to make mine as hot as possible), and max temp was 65C, most time in 63C range, so at least 10C cooler. And I still have my IHS on.
As for wattage, download HW monitor, if your mobo supports it, will read total package and core watts. On mine at 4.4, 1.2v cpuz load, total package said 77W.
I normally use small FFT to get my load temps unless stated otherwise. I'm down to 73-74c small FFT loaded (delidded) now that the paste has had time to settle in, down from about 78c before delidding and 79-80c after popping the lid and replacing it with new paste.
Still though, from one enthusiast to another fellow enthusiast, I am sorry you didn't get the to experience the rewards of the thrill of the chase that many of us got to partake in witnessing our peak temps drop by 10, 15 or 20 degrees
I take heart in your stamina to keep with it and find 1 and 2 degree improvements, that's the spirit! But such a spirit deserves so much more a return on its investment, and I really hope you find that magic bullet that is going to deliver you the drop in temperatures that you are seeking as justification for all your efforts.
Makes me feel a bit guilty, all I did was bang away with a hammer and a blade to unlock the potential of a 20C decrease in temps for my CPU. How lazy and easy was that? And here you are doggedly pursuing every degree in temp reduction you can find. Now that is inspiration mate, cheers to you and godspeed in your pursuit :thumbsup:
Just got some sandpaper, will report back in a bit.
You were brave. Well, brave or weak. I just used a razor and my bare hands. :twisted:
When I got to 1000 grit, I saw a cloudy pattern on that suggested the waterblock might be ever so slightly concave, so I went down to 400 and worked my way back up. Second time through it came out great.
I'm currently bleeding my loop and waiting for the paste to settle in.
Ooh I see shiny things!
OK I just couldn't let this go, it bothered me that we had not definitively determined whether or not the benefits observed in delidding these Ivy Bridge CPU's came from replacing the Intel stock CPU TIM with something superior or if it came from reducing the gap between the IHS and the CPU (reducing the thickness of the CPU TIM).
After failing to nail the gap height correctly using metal shims and thick paper shims (I used 0.203 mm metal shim stock and then 32lb paper), I decided to try again with the paper shim method only I used a thinner paper (20lb).
In an uncompressed state the paper measures 0.09mm thick with my digital calipers.
I cut out a shim using the IHS as a template. Here it is before I put the IHS onto it:
I used NT-H1 for the CPU thermal paste, I also used NT-H1 for the HSF mounting TIM.
When I was done with my tests I removed the H100, cleaned up the TIM on the IHS surface and measured the thickness of the entire CPU package with IHS to see how close the paper shim came to the original stock thickness.
Original Stock Thickness (before delidding, contains the IHS adhesive and Intel stock CPU TIM):
And the measured thickness after replacing the stock CPU TIM with NT-H1 and using the 20lb paper shim under the IHS:
They are identical! :thumbsup: So we know the gap between the CPU and the IHS must be the same in both test conditions, meaning the differences (if any) in operating temperatures will be entirely due to thermal conductivity differences in the CPU TIM (Intel stock vs. NT-H1).
And the results?
^ Compare all the "a" cases to the "b" cases, the only functional differences in these two test conditions is the CPU TIM. Notice that the Intel stock CPU TIM outperforms the NT-H1 replacement TIM once the CPU-to-IHS gap is identical
(all the "b" cases are ~2-3°C warmer than the "a" cases)
And if we remove the paper shim and drop that IHS down onto the CPU (not perfectly zero of course, there is still some NT-H1 CPU TIM there after all) reducing the gap to as close to zero as possible then we get the "c" cases...and the temperatures show the expected fantastic drops we have all come to expect from delidding our Ivy Bridge chips.
Conclusion: The Intel stock CPU TIM is not the reason Ivy Bridge's run hot, and replacing the Intel stock CPU TIM is not the reason a delidded Ivy Bridge runs so much cooler - the benefits of delidding are entirely due to the resultant reduction in gap height between the CPU silicon die and the underside of the IHS.
I wonder if that was a mistake (stranger things have happened), or if the gap is there with a purpose (as we've seen, many of us see a very strange raising of temps over time after we've removed that gap).
Unfortunately we can't logic out our dilemma on that question because there are very reasonable rationale engineering reasons why the gap would be there on purpose, ranging from reduction to so called "pumping out" effects to reduction in thermal stress mismatch to simply building a buffer against variability in the IHS height (spec's cost money, loosen those specs and you can spend less money) and so on.
However we can test the question of "do temps rise over time because of the pumping-out effect" by way of using a TIM that purports to directly minimize and counteract such effects...and IC Diamond purports to accomplish exactly that.
So the test to perform is to use IC Diamond as a the replacement CPU TIM and see if the temperatures remains stable over time.
I was going to do something like this but after my first test with IC Diamond I removed my H100 and observed that the IC Diamond did the following to my H100 and IHS:
This is after rigorous cleaning of both surfaces. That ring you see there is from the IC Diamond (or the solvent they use, just guessing) which has stained/corroded the copper surfaces on my H100 and IHS.
To get rid of the stains I had to resort to relapping/polishing the surfaces. Naturally I'd prefer to not have to do that when testing TIMs.
I never worried about this for the convential HSF-IHS usage, although it was always "evident." Even for pulling off the HSF and reinstalling it, I never bother to "re-lap." I don't know if you would call the stain "corrosion," either.
My findings mirror yours, IDC. My gap must have been very small, because delidding and remounting the IHS with aftermarket paste resulted in a 2-3c rise in temps at high overclocks.
4.6ghz, ~1.304v, LinX loaded:
Unmolested CPU: ~98c
TIM replaced under lid: ~100-101c
Bare die: ~95c
Surprisingly, the IHS itself wasn't hurting temps as much as I suspected either. The paste has had time to settle in for about 18 hours after I lapped my WB and I'm only 3-4c below where I was at stock.
Awesome update IDC. While I don't think I'll ever de-lid myself, reading your very thorough work is very interesting and educating.
Do we get to look forward to that H100 mounted on the die?
Hmmm...De-lid myself could be read a couple of different ways