42C idle at .93 vcore and 1.6 ghz -- i7 920 is a furnace!

[v8envy]

Originally posted by: Rubycon

If you're overclocking, then yes. A stock cooler should never be used (for long anyways) for an overclock. If you properly mount a stock cooler using the pre-applied TIM and run at stock speed you should NEVER approach TjMax regardless of what you're running. Unless your ROOM is like a furnace.

Then either I live in hell, or I'm an idiot repeatedly incapable of mounting a stock heatsink (in spite of many previous successful outcomes).

I removed the sink, cleaned the hard, nasty stock TIM off (which looked to have spread very evenly), used a thin smear of AS5, re-mounted. This was good for taking the idle from 42 to 40C. Load still shot up from 40 to motherboard thermal beeping in under 5 minutes. The motherboard only goes up to 90C for thermal monitoring, so I'd have to turn that feature off if I'm to run at 100C on the CPU 24/7.

In any event, I then did a perfectionist cable management job. Not a single wire hangs over the motherboard, and the SATA, sata power, and motherboard cables travel only a few inches to reach their destination. The PCIe cable was the only one too short to route behind the backplate, so it does travel from a drive cage to the video card in front of the board and in the path of airflow.

Unfortunately, 3 hours of case modding resulted in no apparent CPU temperature drop but was good for a possible 1-2C drop in chipset temperature.

Then I replaced a low speed 120mm side intake fan with a 3600 rpm 92mm Sunon server fan. You know the type, 57 CFM flow, 50db. No fan guard, and it's a freaking hurricane. Pointed it right into the heatsink. Only *this* finally did the trick for load temperatures -- they take about 10 minutes to hit 80C now. I doubt it'll hit 100C.

Of course a single one of those fans makes any machine sound like a server room, and I run a severe risk of having a cat sucked into the intake the very first time said feline tries to rub up against my machine. Whereupon the cat will be promptly and thoroughly shaved. With the shavings deposited on the insides of my machine.

Which is to say, I still don't have a satisfactory cooling solution using the stock heatsink. And I'd say what I've done goes miles beyond what a typical end user can be expected to do.

 

[ChaosDivine]

That is weird. Insufficient case airflow?
BTW : I'm using IC Diamond as my TIM if that makes any difference.

 

[v8envy]

Originally posted by: ChaosDivine
That is weird. Insufficient case airflow?
BTW : I'm using IC Diamond as my TIM if that makes any difference.

Obviously insufficient airflow with the original 120MM side fan, 2 80mm exhaust and 1 80mm intake fan (estimated air flow: 40 cfm). Barely sufficient now, with a high speed server side fan (estimated air flow: 70 cfm).

I'm on a mission to see just what it'll take to get the stock heatsink to keep the CPU under 70C. I'm going to hook up my kid's jumpy castle air pump (estimated airflow: 300 cfm) to this case and see what happens. After that I'm rigging up a 12x92mm server fan array (estimated total airflow: 600 cfm) just to see if that does it. For those of you going "that doesnt' sound like much..." 600 cfm is how much air a very 5 liter v8 gulps down at 4500 rpm. Or roughly enough air to generate ~200 horsepower.

In the very near future I'll be replacing the junk heatsink with a mild aftermarket one with the various case cooling solutions to show just how completely underengineered it is.

 

[Rubycon]

Originally posted by: v8envy

Then either I live in hell, or I'm an idiot repeatedly incapable of mounting a stock heatsink (in spite of many previous successful outcomes).

Perhaps there is a problem with the temperature reading then. Do you really believe it's reaching 90C? Does any part feel warm? If the HSF is cool and the socket is cool after 30 minutes of running at this temperature then that reading is probably bogus.

 

[WaitingForNehalem]

The stock cooler is fine if you don't overclock.

 

[v8envy]

Originally posted by: Rubycon

Perhaps there is a problem with the temperature reading then. Do you really believe it's reaching 90C? Does any part feel warm? If the HSF is cool and the socket is cool after 30 minutes of running at this temperature then that reading is probably bogus.

Entirely possible, but it'd be one annoying hardware bug if so. The idle readings are as expected (40C across all CPUs), part load temperatures are also as expected (mid 40s to high 40s), but only a full load (8x prime95, small fft) shows the heat building up. The behavior is exactly how I'd expect a part designed with a maximum of around 110W to act when trying to dissipate 130W. I have a nice software tool to see CPU multipliers and throttling states (it's very cool to see the automatic 22 multiplier with a single core load, and all the other low power multipliers dynamically) so I could also run the cpu to 100C and see if it throttles from C0 state to one of the throttled states. If so, I'd know the thermal reporting is either accurate or broken at the CPU level.

I'm not about to go touching a possibly 80C piece of metal though -- I'll try to find my IR thermometer and see if I can point it at a CPU heatsink fin. My garage is a bit of a mess, so I probably won't get the time to climb over all the cars, yard gear and boxes to get to my automotive toolboxes today.

 

[v8envy]

Update: curiosity got the better of me and I risked life and limb to get the laser thermometer. Unfortunately the side fan is installed on a window that interferes, so I had to run tests in the "case open meltdown mode" instead. At idle it reports 39-41C on different spots on the heatsink while software is claiming 42-44C on the cores. With the case open and resulting poor airflow most of the motherboard is at a balmy 36C, with the NB heatsink sitting at 49C. The passively cooled 8800GT reports 60C idle, and the thermometer agreed with a 56C reading from the heatpipe base.

Load testing was scary. I fired up prime95 and saw the temperatures quickly climb both in software and on the thermometer. With no hurricane blowing on the heatsink it took about a minute before the 90C alarm went off. I saw 82C on the thermometer right before the alarm went off and I just killed the test. I know everyone in this thread says I can just run this bad boy at water boiling point temperatures 24x7, but that still seems like a bad idea to me. I had an ATI video card start to artifact after 11 months of 85C operation. Could be a coincidence, but I'm still paranoid.

While hardly a scientific experiment with precision tools, I feel confident in saying the software temperature readings are likely within 5% of reality at idle and at load.

 

[God Mode]

Originally posted by: Rubycon
Why are people so critical about what a program tells them their cpu temp is? If it's stable all the way around why worry? Sure they're going to run warmer - you have FOUR cores in that little ah heck.

.

 

[Rubycon]

Be aware that if your IR thermometer does not have an adjustable emissivity its readings could be off considerably depending on what you're targeting.

That said it sounds like you have a hardware problem. Fair idle temp with a sharply increasing load temp approaching TJmax within a minute is usually indicative of insufficient pressure between the bottom of the HSF and the IHS. In rare cases the IHS may not be bonded well to the die requiring a higher than normal pressure only realized with third party (bolt through) coolers.

 

[v8envy]

My thermometer definitely a cheapie without any emissivity adjustments. I use it for relative readings, looking for engine hot spots. That sort of thing.

Which is why I "calibrated" by checking several heat sources including the video card. The aluminum reading at idle was very close to what the chip reported, and is also in line with what others are reporting. When I get some time I'll see about attaching a temperature probe to the heatsink though.

The behavior is very strange. I got an order of magnitude performance improvement (tens of seconds vs minutes) to overheat by upgrading thermal compound. That was completely unexpected, but very strongly supports your lack of contact theory. I will try rotating the sink just in case I got the absolutely worst possible concave/convex combination in its current orientation. I'm not hopeful, because the idle and load temperatures on all cores are within a degree or two.

On the other hand, the heatsink does function once I get case ambients below 30C or so, which goes against the lack of contact theory and more toward not being able to radiate heat fast enough theory.

 

[v8envy]

Well, I think I got it figured out. I'll ascribe my first results to poor mounting, air trapped under the heatsink, mounting posts coming loose or other user error. But the later performance (and difference vs. others users with the same thermal solution) is caused by my location. I'm a bit over a mile of altitude, with about 17.5% fewer air molecules available to whisk heat away from the heatsink.

I stumbled onto this by hooking up the jumpy castle air pump to my case. It did wonders for temperature -- idle at 38C, load never breaking 70C. But it wasn't just a tornado of high velocity 25C air blowing over the heatsink -- the Sunon fan was doing a better job of this already. This pump is designed to pressurize a fairly large rubber castle and keep hundreds of kg of kids bouncing around. The environment more closely matched what the stock thermal solution was designed for -- sea level.

So, the sink may be just adequate for a 130 watt CPU at sea level so long as in-case ambient temperatures are low enough. But with 17.5% fewer air molecules around to take heat away it just simply wasn't able to keep up at full load under "average" case conditions. The older heat sinks had a lot more wiggle room between CPU thermals and capacity, and still performed well even at altitude. This one not so much.