The temperature behavior (fast rise and then fairly flat) is exactly what I would expect in a properly functioning system. These thermal sensors are baked into the CPU cores themselves. When load is applied, they spike immediately as they are literally touching the transistors that are doing the work. The fact that the quick rise/fall levels off with only small changes means that this isn't a problem with saturating the radiator's ability to dissipate the thermal load.
3.0GHz at 1.225 is a very mild overclock and not a high enough voltage bump to result in a lot of extra heat. This would also explain the minor difference you observed when underclocking. I was running 4.11GHz at 1.37 volts on my 960 before the 970--and my load temps were about what you're running. One test you could run would be to go further--drop the vcore to around 1.1v, frequency to 2GHz or less, and maybe even disable all but 1 core. Basically, let's reduce the thermal load to the minimum possible and see what happens to idle temps. Does the radiator inlet feel warm when you touch it? It (and the water) shouldn't be much over ambient room temperature. Water has a tremendous specific heat capacity--meaning that it can absorb a lot of thermal energy with only small rises in its own temperature. I don't think we need an exact in-water temp reading, but if you could affix a thermal sensor to the outside of the radiator's inlet, it might be useful information. Speaking of the water--what mix of coolant are you using? Anything in the loop that isn't water actually will reduce its thermal performance. I now run a loop with pure distilled water and run some silver coils as a biocide. It's possible that the loop has a restriction or the pump is defective, resulting in insufficient water flow. This may even be a case where it came with debris/contamination from the factory or retailer. Along those lines, you could consider disassembling the water block and examining the pin matrix on the block's internal plate. If there was any small debris, it would probably collect in the very fine mesh on the block's surface, and that would reduce its effectiveness greatly. Be advised that disassembling the block will void its warranty. I don't know what, if any, role that the system plays in the DTS readings, but you could make sure that you're running the latest BIOS version for your board. At this point, unless the water is warm, we really can't make a definite determination of an instrumentation glitch vs. a real problem. My instinct is still to suspect a bad physical mating between the block and the processor's IHS. It could be something like a defective mount, warped motherboard, warped block, or even factory plastic film on the block (it's certainly easy enough to do)... Finally, I want to reiterate that your temps are not actually bad, they're just higher than I would expect for that cooling gear given the frequency and volts that you're feeding it. > -----Original Message----- > From: [email protected] [mailto:hardware- > [email protected]] On Behalf Of James Maki > Sent: Sunday, July 10, 2011 1:20 PM > To: [email protected] > Subject: Re: [H] Water Cooling and ambient temperature > > Greg, > > Again, thank you for the well thought out answers and analysis. > > I downloaded CoreTemp this evening and it is giving approximately the same > temps. > > Not sure on the fans. They are the stock fans that come with the Swiftech > kit. > > I backed off the oc frequency to 120 (a 2.5 GHz value) and backed off the > cpu voltage to 1.2V from the default 1.225 V. I made no other changes for > the original overclock in the BIOS. These changes resulted in little or no > change. As the CPU throttles itself back due to inactivity (currently > running at 1.8 GHz), I see no change in the temperatures. After changing to > the Arctic Silver 5, the temps are currently running 39 to 43 for the four > cores at idle. If I run the Prime95 torture test, the temp rises almost > immediately to 60 to 64 degrees C. (This is with no overclocking, running at > 2.5 GHz). After 3 minutes, 64 to 66 degrees C. After 10 minutes, the temps > are still 64 to 66 degrees. After 20 minutes, a slight change to 65 to 67 > degrees C (although the high and low fluctuate up and down over time). > > Raising the clock back to 150 (a 3.0 GHz cpu speed) resulted in the temps > raising to 71-74 degrees C. Halting Prime95 drops the temps to 42-45 almost > immediately. After 5 minutes, the temps are back to 38-43 degrees C. > Backing > the oc back to 120 (2.5 GHz) leaves the temps at 38-42 degrees C. > > What does it mean that the temps rise fast to a max and fall fast to a min? > There is little upward or downward trend between idle and 100% usage with > time. > > I could try the thermometer in the reservoir, although it would require some > dis-assembly and acrobatic maneuvers. :) > > I find it strange that the idle temps seem unchanged between the overclock > and the underclock situations. Also, the maximum temps reached are lower > this evening (which could be related to cooler ambient temperatures) after > changing the TMI, but the idle temps have actually raised even with the > underclocking/undervolting. > > I have been searching google about this problem/situation and see a wide > variety of temperature claims for the i7 920. Some claim almost ambient > temperatures at idle and 50-60 degrees C at 100%. Others seem closer to my > results with 35-45 degrees C at idle and 70-80 degrees at 100% usage. These > result are usually on air cooling. > > Thanks for the guidance. > > Jim > > > -----Original Message----- > > From: [email protected] [mailto:hardware- > > [email protected]] On Behalf Of Greg Sevart > > Sent: Saturday, July 09, 2011 10:02 PM > > To: [email protected] > > Subject: Re: [H] Water Cooling and ambient temperature > > > > It's certainly possible. I'm a fan of CoreTemp myself, but Realtemp is > also > > popular. Make sure that you're running the latest versions, too, since > they > > could be using faulty Tjmax data. The DTS (digital thermal sensor) on > these > > chips does not report the actual temp--they report the margin between > the > > current temp and the pre-defined Tjmax temperature. Therefore, if the > > utility is using an incorrect value for Tjmax, the reported temp will be > wrong. > > > > Replacing the TIM isn't going to make but a few C difference probably, and > > it'd be more pronounced at load. > > > > What fans are you using on the radiator? A lot of fans don't create much > > static pressure, which is required to move air through the restrictive > radiator. > > I'm using 3x Scythe SFF21F fans on mine that seem to do a decent job. They > > don't have the greatest static pressure spec either, but they're extremely > > reliable and are fairly quiet. I've had the same fans on the radiator for > 4 > > years... > > > > I'd also try backing everything to stock frequencies and volts. I think > you > > indicated a very mild overclock, but some boards can take great liberties > with > > vcore and such when left on automatic, so you could be dumping more > heat > > than you realize. These i7-9xx series chips are rated for a 130W TDP, but > it > > isn't that hard to get another 60, 80, 100, or even more watts with big > > overclocks and/or big voltage bumps--some of which may very well be > > inadvertent. > > > > When you apply a load like LinX/IBT or Prime95 small FFT, does the temp > > jump up quickly and increase only slightly (a few C), or does it jump > quickly > > and then continue to rise? Is there any way that you could get a > temperature > > reading of the water itself--like a temp probe on the radiator inlet? > > > > Do you have any of the power management processor features disabled in > > the system BIOS? Most overclocking guides say to disable most of them, > but > I > > refuse to. I personally don't consider it a stable overclock if I have to > disable > > features. Things like C1E, EIST, etc. can play a big role in idle temps. > > > > Finally, some chips just have bad thermal sensors and can be off by a good > > margin. They aren't engineered to be exact--they're engineered to be an > > input into the thermal management system. It could also be that my 970 is > > reading low (and the 32nm process shrink probably means that it's pulling > > less at idle, too). I just don't remember what my idle temps were on the > 930 > > and 960 that proceeded it. My load temps were close to yours, but that was > > with a big overclock (4.0-4.1GHz) with a healthy voltage bump. > > > > > -----Original Message----- > > > From: [email protected] [mailto:hardware- > > > [email protected]] On Behalf Of James Maki > > > Sent: Saturday, July 09, 2011 10:43 PM > > > To: [email protected] > > > Subject: Re: [H] Water Cooling and ambient temperature > > > > > > Could the temperature monitors I am using be wrong? > > > > > > > -----Original Message----- > > > > From: [email protected] [mailto:hardware- > > > > [email protected]] On Behalf Of James Maki > > > > Sent: Saturday, July 09, 2011 8:33 PM > > > > To: [email protected] > > > > Subject: Re: [H] Water Cooling and ambient temperature > > > > > > > > Well, I took the cpu block off. Cleaned it off (I may have applied > > > > too > > > much > > > > TIM the first time.) and reapplied less. I only have the Arctic > > > > Silver 5 > > > at this > > > > time. Reassembled and was seeing almost identical temps. Took the > > > > block > > > off > > > > again. Definitely did not have too much TMI this time. Cleaned off > > > > and reapplied again. Reassembled only to see similar temps again. > > > > The pump > > is > > > > working, so that is not the problem. Any other suggestions? Seems > > > > like a simple system that should just work. Can the block be > defective? > > > > > > > > Thanks for any insight. > > > > > > > > Jim > > >
