CPU cores have wldly different temps

[Sumotku]

Using an E8400 at stock on a gigabyte EP35-DS3R with a Thermalright Ultra120 Extreme I'm seeing a steady difference in core temps of 11 degrees. At idle they show 27/38 with ReatTemp and 37/48 with Everest. This is a new development, one of the first things I checked upon firing it up was the temps and the cores measured equally about two weeks ago. I updated from Bios F2 to F3 and no change. No other configuration changes. More specs: Seasonic 550HT PS/2 GB HyperX 1066 @ 2.1V/WD 640AAKS/MSI 7800GT. I took the time to pull the CPU cooler out, recheck it for flat (lapped to near perfection) and reverse it's orientation to the cpu with zero change to temp problem. My next step would be to buy a cheap dual core to test the MB with, but I'll wait until I hear your ideas.

 

[Viper GTS]

Just ignore it. Intel has essentially confirmed that the temp sensors in current CPUs are nearly useless, particularly at low temperatures.

http://intel.wingateweb.com/US.../SF08_TMTS001_100r.pdf

Look at page 6 and 8 in particular. Also, page 13 specifies the actual TJ value for your processor (100 C), the gap between realtemp & everest that you're seeing is due to them using different TJ values. Odds are unless you are using the brand new release of RealTemp both are incorrect.

Viper GTS

 

[Sumotku]

It'd be easier to ignore if it weren't a new development, for the first week they did not show this 11 degree avg difference. I'm using Realtemp 2.70 (latest). I've also tried disabling C1E and EIST with no change either. I'm wondering if it's more likely the CPU or the MB that's manifesting this new 11 degree variation.. and to clarify, it's the difference between cores -not programs- that is the concern.

 

[onedestinazn2]

this isnt a huge deal i get the same between my q6600's four cores. id only be concerned if your temps were high

 

[Viper GTS]

2.70 is not the latest, 2.75 is the version that reflects the intel confirmed TJ values.

Check out the very end of this thread:

http://www.xtremesystems.org/f...howthread.php?t=179044

Sounds like a stuck sensor, what does IntelBurnTest give you for load temps? They'll probably be much closer. Look at it this way: Is it really logical that two temperature sensors on the same physical die are 11C apart? Particularly when there's a layer of metal atop them?

Originally posted by: Sumotku
and to clarify, it's the difference between cores -not programs- that is the concern.

I know exactly what you're worried about, but you have got to quit trusting those numbers. Even Intel says they're garbage.

The sensor data put out by your CPU is obviously wrong, you might as well quit wasting your time trying to "fix" it.

Viper GTS

 

[Sumotku]

My thinking is that this is pointing to a different but related problem, why? Because they did not show this variation for the first week of the new build. If it were like this from the start I'd write it off as a minor nuissance. It's hard to see it that way now.

 

[StarvinMarvin22]

The sensors aren't designed to report you idle temps - don't worry about it

They will get stuck at large distance to TJ readings (low temps). I'd almost be sure that if you load your CPU up that the reported temps are closer together, right?

 

[Sumotku]

Running IntelBurnTest, both cores pegged at 100%, Real Temp 2.75 records a maximum of 43°/53°.

 

[error8]

Originally posted by: Sumotku
Running IntelBurnTest, both cores pegged at 100%, Real Temp 2.75 records a maximum of 43°/53°.

So, why are you so worried? With my chip at 4 ghz I'm getting 86 / 86 C on both cores with the IntelBurnTest. This is a reason to get worried, not 43 / 53.

 

[BonzaiDuck]

This may be "totally off the wall."

Did you lap down the convex ridge of the TRUE's base to a flat copper surface?

The other remarks here are also relevant. RealTemp has a test to indicate whether the sensors might be sticking. I can't remember if you mentioned that.

The sensors may be crap, and per Intel's own admission. But there was another angle on the issue: Look for a March, 08 Anandtech article about how sensors are read and interpreted -- it dealt with the E8500 Wolfdale.

My 680i motherboard is in the sunset of its lifecycle, but ASUS had forward vision toward making it Penryn-compatible when the processors were finally released. In September, 07, they posted a revised BIOS -- more likely to accommodate the G0 steppings of the Conroe. Later they went through two more BIOS revisions coming into March of this year. The first of them would recognize my E8400, but the idle temperatures as shown in BIOS monitor were way too high, or 50C degrees. The earlier BIOS showed them at 10C, which was ridiculous under a room-ambient between 20 and 30.

They finally posted another BIOS revision in first week of this month, August 2008. Suddenly, all the core values and tCase values seem spot-on, even if they didn't list a fix for this in their BIOS-revision "fix-list."

The Wolfdales, if I remember the Anandtech article correctly, implemented a different approach to the sensors and their interpretation against tJunction. So older BIOS revisions would have trouble with it.

When did you buy that Gigabyte (P35 chipset?) motherboard, which BIOS revision does it have, and when was that BIOS released?

Lap the TRUE anyway -- if you want to take the time. The trick to it: clamp two metal or wood blocks to the sides of the base so they meet the sandpaper level with the convex ridge of the base. Once you see a flat surface along the ridge, you can dump the clamp and blocks. The ridge on the TRUE's base accounts for less than a millimeter's difference between the base before lapping and after lapping.

 

[BonzaiDuck]

Here -- eyeball this:

Accurate Temperature monitoring?

I've been criticized here or there on these forums for missing some detail or other when I scan through articles looking for ideas. But I stumbled onto this while puzzling over the temperatures from my BIOS 1504 revision, and I was very quick to download version 1603 when it appeared a couple weeks ago.

I'm just willing to bet that Gigabyte and other board-makers -- with either nVidia, Intel or both types of chipsets in their model-lineup, were a tad behind Intel's own boards when it came to keeping up with CPU changes.

 

[Sumotku]

The TRUE was lapped on a 600.00 diamond embedded precision machined steel plate intended for flattening of much harder high carbon steel, it makes quick work of copper while leaving a mirror finish. The Gigabyte MB is rev 2.1, tried both F2 and F3 bios - same result. I used Real Temp's sensor test and got 5 and 6, right in the middle of average. The cores yield low temps at idle, 15° higher temps under load, none of them cause for concern. Now imagine your working along for a couple weeks and everythings looking great. Then one day your working and notice that the core temps are 11° (30%) different from each other, under any load, read with any program and the sensors are not stuck. It doesn't exactly give one confidence to see such an aberration, more a feeling of 'what's next?' The QX6700 in my sig has core variations as high as 5° but similiarly reflect activity in each to support it. This is quite a different thing.

 

[BonzaiDuck]

You know what you're doing. It's apparent.

I'm only guessing that something happened inside the processor with the sensors.

And I can only say that I was somewhere between puzzlement and Chicken-Little when I saw my own BIOS monitor values under the two different BIOS versions.

The $60 cheapo- Gigabyte boards I bought in January for two fam-damn-ily upgrades also had BIOS revisions labeled "F2" and "F3" -- and the F3 update was available in January -- or maybe it was February. But the Penryns hit the street just about that time. ASUS didn't have it right for my motherboard until six months later.

Maybe you've already poked around on the Gigabyte web-site for something more recent. If not, I'd look anyway before resigning to a conclusion of a defective sensor in the CPU.

Otherwise, nothing else to do, I think. You can save your pennies and watch the prices on the E8600. That's what I'm doing. If you lapped the IHS on the E8400 -- as I did and would have done -- somebody else might still want it, or it might eventually make a good "backup spare."

But as others said here, a bad sensor is no reason to chuck the CPU unless the imperfection nags you to distraction.

 

[BonzaiDuck]

Incidentally, the E21x0 processors are almost cheap as dirt. Those were the ones I used in the two upgrades I mentioned.

 

[Tullphan]

What's more accurate...Real Temp or Core Temp?
I've got both & Real Temp's showing about 4-5c than Core Temp.


Edit: Nevermind. I found this thread that tells me why.

 

[BonzaiDuck]

That's illuminating, Tullphan.

The prevailing skinny around here was that RealTemp is absolutely accurate. By the advice in that thread, I'd need to calibrate the program by addiing +5C. That means that my miraculous record-breaking load temperatures with the E8400 I was using weren't so "record-breaking."

 

[daw123]

I'm have the same issue with my QX9770 at stock speeds - two cores show 32C and the other two show 43C on HW Monitor.

Since its not running too hot, I'm just going to ignore it.

I will be going to liquid cooling any way when I overclock it to the 4.2GHz mark.

 

[BonzaiDuck]

This is beginning to look more and more like an Intel conspiracy. Note especially that nVidea had replaced the old nTune and nV Monitor software with "nVidia Tools," and they dropped the monitoring of TCASE temperature, which in turn means that thermal control of cooling devices exclude it. Yet, as far as I can see, TCASE varies reliably for what you'd expect for any given cooling solution, even if absolute values are biassed (a BIOS problem.)

 

[Sumotku]

A late note that may be of interest. I had occasion to swap the ill-tempered E8400 for one with balanced sensors. Only then did I discover that the whole time I was overclocking it I had an unlocked multiplier, specifically a .5 option in BIOS which disappeared with the new processor in place. "Careful what you wish for" comes to mind...