DFI Expert CPU temp sensor reporting incorrectly?

[Blaise]

I would make sure that my idle temps are under 140C (preferable 135C or less) and my peak load temps are under 150C (perferably under 145-148C). And, peak load is continuous prime95s for hours on end (in both cores if you have dual core processor).

Are you sure??? Remember, water boils at 100°C. I think you have added 100°C to each of the temperatures you listed in the above quote by accident.

[davidk21770]

:eek: Boy, I was having a bad typing. Remove the leading 1s .

I'll edit the original message (teach me to post when I'm tired).

[RichGar]

Hi

 

Tks for the answer.

The revision is AB0.

The temperature read by the sensor is 37°.

The temperature read by the Smart Guardian is 30°

The temperarture in the room is 30°

Then i can use the Stmart Guardian read as a room termometer.

I dont know why now its showing lower temp. I will test with full load with Prime95

 

Thank you again.

[davidk21770]

Hi

 

Tks for the answer.

The revision is AB0.

The temperature read by the sensor is 37°.

The temperature read by the Smart Guardian is 30°

The temperarture in the room is 30°

Then i can use the Stmart Guardian read as a room termometer.

I dont know why now its showing lower temp. I will test with full load with Prime95

 

Thank you again.

By the "sensor", I assume that you mean in the BIOS. You can't use that reading to calibrate SmartGuardian. The BIOS uses the CPU differently than windows and actually loads the CPU higher (simple program -- less idle cycles = more heat). SmartGuardian is reporting using the same sensor that the BIOS reports. If SmartGuardian is reporting 30C when the ambient temperature is 30C, your're reading low. There's no way that the inside of the case is cooler than the room and that the heatsink is so efficient that it's cooling to room temperature when you're normally clocked -- even at idle. I'd assume that you're off by at least 5C, maybe more. If you read back through the threads you should be able to find what people are getting with normally clocked CPUs and assume that you're in same ballpark (well, your room temp's a bit higher and you're using the stock heatsink, so you're going to be a bit higher).

[RichGar]

Tks again.

 

The sensor is one that i stick in the copper tube that is inside of the heatsink. The nearest place to the cpu i can reach. Then, the temperature is 7°C higher than the Smart Guardian. Definitly the smart guardian read is at least 7° lower than the real, may be more.

 

The last night i left running the prime95 during more than 8 hours.

The temperature in the sensor (the one i install) was 42°C.

The Smartguardian was 38°C. Only a difference of 4°C.

I can think, "if without load the smart guardian read 30° and with load it shows 38°, then the sensor i put in the heatsink must read 45° with load."

But no. Why? I dont know.

 

I will start overclocking with care.

May be with higher temperature, the smart guardian is more accurate.

Any idea?

 

Thank you again.

[davidk21770]

Sounds like your system is stable .

 

The SmartGuardian (MB) temp should be incorrect by a (roughly) constant. The idea is to figure out what the constant is and then know that it is off by that much (or calibrate smart guardian or your BIOS -- the 4/06 BIOS has a setting on the Health page to do this -- auto is wrong -- seems to be zero).

 

Heat flows. If you take a metal bar, heat it at one end, cool it at the other, and measure the temperature in the middle, the bigger the temperature difference between the hot end and the cold end, the bigger the temperature difference between the measurement and the hot end. And, the closer the probe is to the hot end, the closer to measuring the correct (hot end) temp it will be. So to calibrate, what we want is to take a mst as close to the CPU (the hot end) as possible and have the "cold" end (the heatsink) as close to the CPU temp as possible.

 

The sensor read by smart guardian is very close to the CPU -- much closer than your measurement probe. This is why when you put a sudden load on the CPU (start priming), the temperature reported by SmartGuardian will rise quickly, but your temperature probe will rise very slowly. Probably the reason that you got a better reading after priming for 8 hours than when you made a quick idle measurement is because the cold end (heatsink) rose in temperature to be closer to the hot end (CPU).

 

To get the hot and cold ends as close to the same temp as possible, I try to get everything as close to room temp as possible. I make the CPU as cool as possible (not generating much heat) and wait for the "cool" end (heatsink) to rise as far is it will in temperature (steady-state).

 

To make the CPU cool, I drop Vcore to 1.3v (or your lowest stable V), drop the multiplier to 4 (whatever the min in your BIOS is) and boot. Then I use nTune to drop the clock to 100 Mhz in steps of 50 MHz (I can't set it below 200 in the BIOS, and if you change it too quickly, your system will freeze). So, my 2.4GHz 4800+ (normally running at 2.8+ GHz ) is running at 400 MHz. Then I let it idle for a while so that everything comes to as even a temperature as possible -- when your probe hasn't changed for 5-10 minutes or so. I assume that my probe is probably close to the correct temp (within a degree) and make that my calibration; i.e., I assume SmartGuardian (BIOS) should be reading the probe temp plus maybe a degree.

 

EDIT: Note... use of nTune to change systems settings is strongly recommended-against in the forums here. The experience is that is can sometimes damage the MB. I've used it with impunity, but I've stopped using it just in case I've simply be lucky.

[Guest burningrave101]

On my previous AA0 board, idle/load temps were 24C/37C before +10C compensation. My idle/load temps on new AB0 board now reads 31C/42C with new simplified loop (CPU block, Rad, and Pump).

 

I think 31C/42C (idle/load) running 2.9Ghz @ 1.39V is not optimistic with my current setup. Before 24C/37C on AA0 board is obviously unrealistic but 31C/42C seems reasonable to me with new WC.

 

Anyone else have an AB0 board and done any actual temp measurements? I have an AB0 as well and i would be interested in knowing how far off on average these boards are in temp readings in comparison to the other revisions.

[soloz2]

where on the board does it say?

[davidk21770]

where on the board does it say?

It's on a label on the side of the 1st or 2nd PCI slot.

[RichGar]

Good idea.

 

Ill do it first.

 

Thanks.

[derekoldschool]

I'm currently prime testing my rig at 2.67 Ghz (297x9)

 

I just put the system together a couple of weeks ago, so I havent really had a chance to test it as much as I would like, but from my experience with overclocking my other computer (see BlackMamba in sig), I think my temps are accurate - or at least in the ball park.

 

Vcore = 1.62, CPU temps are 48-52C under load, and 28-30C idle ambient temp is 22-25C

 

Chipset voltage = 1.68v and it is currently 42C under load and usually goes down to 38-40C or so idle. I have replaced the stock dfi fan with a copper vantec Iceberq fan before i even installed the board in the case so I dont know how my temps would have been with the stock fan. I recently added one of those pci-slot style exhaust coolers under my vid card and this lowered my chipset temps by 3-4C

 

LTD voltage = 1.39v PWMIC temp is 49C under load. I have a 80mm fan perched in the cables pointed directly at the chip and I have put a small copper heatsink on the chip as well. Since I put the fan on it, its temp under load dropped by about 5C and my chipset temps went down by 2-3C

[d3M0n]

Just thought I'd let everyone in this thread that has said they have problems with the temp reporting on the CPU for the Expert motherboard... I didn't even see it till today when I popped on the DFI site to see if there were any updates, and VOILA!!

 

A new bios (4/6/06) is out for the Expert and GUESS WHAT?? It has a CPU temp calibration option!! Well... now you can have the correct CPU temp (according to your calculations) reported! There are some other fixes as well, but I'll let you read about that yourself... Here's a link

 

Hope everyone is happy with that... seems the best DFI could do under the circumstances... :cool: