Starting From Scratch

[Dr. Blue]

I am just trying to understand most of the posts I have read over the last few months. Still confused, I choose to leave this basic question. Intel Q 9550 (EO) is the processor in question and Asus P5E VM HDMI is the MB....all settings are set on factory auto....when starting Real Temp Sensor Test......right off the back I have on Core Movement 16 12 8 7. Is that normal?? Other than that it completes the test and all seems fine with the exception of core 0.. it reads approx 6 degrees higher thru the entire test over the other 3 cores untill the last 3 phases of cool down then they are all very similar readings..all in all I was merely assuming that the Core Movement would have been a lil closer across all cores???

 

[Mussels]

intel 45nm's are known for wonky temp sensors.

 

[phanbuey]

yep... 45nm sensors = asking four drunks what they think the temperature in the bar is at that moment.

 

[exodusprime1337]

i second that, always had to read the boards temps on any intel quad i had in the 45nm e0 range, i wouldn't worry about it, just be smart and logical don't try to kill your proccy and you should be fine.

 

[unclewebb]

For a 45nm Quad, your results are typical.

When you have sensor movement of 16 on core 0 and only 8 and 7 on core 2 and core 3; that's a pretty good sign that those two end cores might have sticking temperature sensors at idle.

Post a RealTemp CPU Cool Down Test if you want me to have a better look at your sensors. That test gives me a lot of relevant data about your sensors without me having to ask you 101 questions.

Intel never designed these sensors to be used to report accurate core temperatures. It would have been nice if they did but these sensors were only designed to control thermal throttling and thermal shutdown and for that purpose they work great. Some users get lucky and can also use these sensors to report reasonably accurate core temperatures and some users aren't so lucky.

Even with wonky sensors, your computer will still run just fine. These CPUs have a lot of headroom so there's no reason to worry about your core temperature. Your Intel CPU is quite capable of looking after itself.

 

[phanbuey]

much more eloquent than my drunks in a bar example

 

[Dr. Blue]

Thanks for all your replies....Unclewebb: all is working fine now....I clocked it to 3.612 Ghz and all is fine, temps are as expected with this cooler I have for now....full load 59 57 58 58 on the cores at load.....and yes cores 2 and 3 seem to have quirky sensors.at idle anyways........cuz first thing in the morning cores 0 and 1 are sitting at 33 34 and 2 and 3 are at 39 39.and thats after being shut off all night.....its amazing what a difference in temps just by adding a 80mm fan on the back side if this Arctic Cooling Freezer 7 Pro .sucking off the heat dropped temps from in the 60's at load to what I mentioned above.thanks again to all of you!!

 

[unclewebb]

Very likely that 39C is the sticking point for the sensors on core 2 and core 3. That means that even is you move to the North Pole, they will still never display a temperature less than 39C.

Ignore these at idle and enjoy your CPU. At least they work well at full load which is better than what most of them do.

Speaking of drunks in a bar, I sometimes wonder what bar napkin has the details of a way to save 5 cents on each CPU if we decide to go with a crappy sensor.
Luckily Intel learned from this oversight and are using much better sensors on their Core i series of CPUs.

 

[Mussels]

the simple explanation uncle webb is probably sick of saying, is that intel only designed these sensors to be accurate at load.

The closer to TJmax they are, the more accurate they are

 

[unclewebb]

The closer to TJmax they are, the more accurate they are

Even that statement is only partially true because TJMax is not well defined. On some of the 45nm Quads, it's very likely that TJMax can vary by 10C from one core to the next. An error of only 10% where Intel "calibrates" these sensors is hardly worth patting Intel on the back for.

The simplest thing I've come up with is as long as your computer is not thermal throttling and it's running reliably then don't worry about your core temperature. That's a lot easier than trying to calibrate Intel's random number generators.

 

[Mussels]

Even that statement is only partially true because TJMax is not well defined. On some of the 45nm Quads, it's very likely that TJMax can vary by 10C from one core to the next. An error of only 10% where Intel "calibrates" these sensors is hardly worth patting Intel on the back for.

The simplest thing I've come up with is as long as your computer is not thermal throttling and it's running reliably then don't worry about your core temperature. That's a lot easier than trying to calibrate Intel's random number generators.

i've recently been testing that actually, and found quite a lot of oddities.


My method involves loading the CPU on a passive cooled heatsink at low clocks and volts, and using an infrared thermometer to check the base of the CPU (either back of the mobo, or just the base of the CPU heatsink) and comparing the temps.

EG, my xeon is rated at 72.4C by intel just like an E8400, however we noticed that the temps didnt match reality - my xeon has a TJmax of 95C, while the E8400 truly does have a TJmax o4 72.4C


(when idle goes below ambient, and load is 20C off of what the base of the heatsink is, you know its not right)

 

[somebody]

Mussels, I'm a little confused as to what your saying, are you saying at load your E8400 core reading was 20C below the reading of the heatsink using an infrared thermometer?

 

[Mussels]

Mussels, I'm a little confused as to what your saying, are you saying at load your E8400 core reading was 20C below the reading of the heatsink using an infrared thermometer?

I tested both CPU's with intels spec (72.4C) and the "assumed" correct temp i kept getting told online, of 95C


On the E8400, it was reading roughly 20C too high on the 95C (thus, we determined 72.4C was about right)

With the Xeon set at the same 72.4C TJmax in realtemp (if intels right on one CPU, its right on both, right? - wrong) it read 20C too low


My point is merely that as unclewebb has said, TJmax needs to be tested yourself, if you want accurate readings.

 

[burebista]

One observation: 72.4°C is not TJMax it's TCase.

 

[somebody]

Okay, as burebista points out there will be a difference between core and case.

The 72.4C temperature is a case temperature normally measured at the center of the IHS and called Tcase.

The core temperature is measured on the die and is called Tjunction.

There is a thermal resistance between core and case such that as the load increases it is normal for the difference in core (junction) and IHS (case) to increase. The junction will be hotter than the case but because the core temperature may be inaccurate this might not appear so.

Let's take an example. In our example at TDP the CPU has a maximum core temperature (Tjmax) of 100C and a maximum case temperature of 72.4C. TDP is stated as 65 Watts.

From the above we can work out the thermal resistance between core and case (IHS).

Thermal resistance Tjc = (Tjmax - Tcmax) / TDP

Tjc = (100 - 72.4) / 65

Tjc = 0.425 C/W


The difference at a cpu power of 20W would be 20 x 0.425 = 8.5C so if the case temperature was 40C then the core temperature would be 48.5C.

At a cpu power of 100W the difference would be 42.5C. Notice in this case we are using a 100W for a cpu that is rated for a TDP of 65W. This is the sort of thing you might see when overclocking.


It doesn't stop there. The case to air thermal resistance of the IHS is quite high therefore it would only take a few watts for the core temperture to reach over 100C. So we normally use a heatsink with Thermal Interface Material (TIM). The better the heatsink the lower the thermal resistance. Lets take an example of a heatsink and TIM offering a thermal resistance to air of 0.4C/W. By adding this to our cpu Tjc we now have a total thermal resistance of 0.4 + 0.425 = 0.825 C/W.

Using this in another example. At a local ambient temperature (the temperature of the air around the heatsink) of 22C and running the cpu at the TDP of 65W we can calculate the core temperature to be

(Total thermal resistance x power) + local ambient
(0.825 x 65) + 22 = 75.6C
So in this case the heatsink and thermal paste offers a working thermal solution.

Be aware that the local ambient temperature directly affects the core temperature regardless of power so if the cpu was in a case with poor cooling and the local air temperature around the heatsink increases by 10C to 32C then the core will likewise increase by 10C to 85.6C.

How about if we want to overclock where the cpu power increases to 100W? Well in that case we need to find a better cooling solution as the one we have would mean the core temperature being
(0.825 x 100) + 22 = 104.5C

Does this help?

 

[Mussels]

One observation: 72.4°C is not TJMax it's TCase.

except that its the only one that makes sense on the E8400 - anything else and the physical temperature doesnt match the real temperature even remotely



Its kinda the point i'm making here, Tjmax varies so much between chips (When in theory it 'shouldnt') that you really do need to test it

 

[burebista]

Now you see why I keep saying that the only value that make sense is distance to TJMax (whatever TJMax it is)?
Keep that distance to TJMax >20 and everything is fine. At least for me.

 

[Mussels]

Now you see why I keep saying that the only value that make sense is distance to TJMax (whatever TJMax it is)?
Keep that distance to TJMax >20 and everything is fine. At least for me.

fair point that.

 

[Dr. Blue]

well I was gonna post test results of the sensor test this morning but for some reason worker 2 on prime95 stopped......passed intel burn test and OCCT test .....now I gotta figure this out whats up in these settings im currently using.

 

[Dr. Blue]

here are the results at 3.4 Ghz::: if its even seeable.....lol

 

[Dr. Blue]

and this is at 3.655 Mhz

 

[unclewebb]

My method involves loading the CPU on a passive cooled heatsink at low clocks and volts, and using an infrared thermometer to check the base of the CPU (either back of the mobo, or just the base of the CPU heatsink) and comparing the temps.

An IR thermometer reading of the back of the mobo or base of the heatsink doesn't tell you too much about the actual core temperature at the hottest spot on the core. Even when I was testing without a heatsink installed, I still wasn't measuring the actual core temperature but at least I was getting closer.

When testing my E8400 I removed the heatsink, dropped the voltage down to about 1.10 volts and I dropped the MHz down to about 2000 MHz. At idle, 1600 MHz or 2000 MHz didn't make any significant difference. Based on rge's testing with a calibrated thermocouple and IR testing we came to the conclusion that the peak core temperature has to be approximately 5C higher than the measured IHS temperature. If you think about how heat dissipates, that seems reasonable to me.

I also used a hand held fan to stabilize the CPU temperature. An E8400 had no problems at all running without a heatsink as long as it had some air flowing over the IHS. My old 7000 rpm AMD Athlon heatsink fan finally came in handy for something. When using a TJMax of 95C, the reported core temperature and the measured IHS temperature were pretty much the same. Knowing that the hottest spot on the core has to be higher than the surface temperature of the IHS based on heat dissipation, it only made sense that TJMax must be higher so 100C sounds reasonable to me. Intel released that number at their first IDF conference and have never changed it.

I hope you can do some more testing mussels but I also hope you pull the heatsink off. As long as you have a fan handy, the CPU will be fine. Trying to keep a Q6600 B2 under control using this method isn't nearly as easy. Pretty much impossible actually.

The base of the heatsink or back of the motherboard is going to be significantly cooler than the hottest spot on the core.

Dr. Blue: Core 2 and core 3 look stuck to me. My best guess is that core 0 is reasonably accurate as is with a default TJMax=100C and core 2 and core 3 are likely closer to TJMax = 103C. They both get stuck so are useless at idle. TJMax on Core 1 looks to be significantly higher. Maybe 105C for that one. It also has slope error so if you correct for TJMax then you'd need to run a calibration formula and correct for problems at idle. I wouldn't bother. Just keep in mind that it has a few more issues than the other 3.

 

[Dr. Blue]

thanks unclewebb, I got it back up to 3.6125MHz and all is stable. I guess Im not really too concerned with the temps now that I seen the tests results and will prolly never run it that hard, at least its stable enough for every day usage!

 

[phanbuey]

you are being very conservative with that chip... run that sucker at 4.0Ghz you will be just fine. keep the volts at around 1.4 and you should be golden.

 

[Dr. Blue]

you are being very conservative with that chip... run that sucker at 4.0Ghz you will be just fine. keep the volts at around 1.4 and you should be golden.

LOL I thought I was pushing it LOL I better get a better CPU cooler first and speaking of that.Im more concerned with this mobo not too sure about the NB voltages...remember this is all new to Me lol