29 C for an idle temp on a 900 MHz TB...does that sound right?

[miniMUNCH]

My system temp is 22C right now. I'm running a Abit KT7-Raid and I have a Chrome Orb that I just put on yesterday. With the fan that shipped with the CPU I was run an idle temp of about 37C, maybe a little higher...it would go up to about 45C during Deus Ex. The heatsink/fan was a piece of crap.

Is this temperature normal...it seems too low to be true.

Edit: I'm just monitoring the temp using the VIA HW monitor and the little CPU thermistor.

 

[Mikewarrior2]

maybe its right, if your ambient case temp is around 0.

honestly, though... c-orbs/s-orbs do something that interferes with socket-thermistor readings. see here: Diode Versus Thermsitor Readings.

What is your ambient case temp?


Mike

 

[paruhd0x]

my 900MHz Tbird is ~50ºC...

 

[Losty]

mine is 30C idle
ambient is 17-20C

 

[rempe]

KT7 @ 1113
Idle 30c
mb 22c
fop38

 

[miniMUNCH]

My ambient temp is about 20-22C. When playing Deus Ex the temp climbs to 38-40C with the thermaltake on...

The temps I'm taking about here are default clock @ 900MHz...not OC'ed. I've been pushing the FSB up to about 110-113MHz...my memmory doesn't like it very much (146-150 2-2-2)...I have some Mosel Rev2 on the way though that will do 160 2-2-2.

I just want to know how much, if any, room I have to OC given my current temperatures. The 100 MHz OC only seems to bump the temps about 5C in games, less in idle.

 

[miniMUNCH]

Hey Mike,

I am unfamiliar with Celeron chips and chipsets but I have few question concerning your article.

What is the guaranteed accuracy of the internal diode on the Celeron...not that it is really important?

Does the Celeron socket have open paths to the interior of the socket so that air can flow to the thermistor contact area? I don't think "A" sockets have any open path ways?

Related to above - If the thermistor under the CPU core is not in the direct path of air flow (at least air exchange) how can air flow differences between the Orbs and Fop, etc. effect the reading of the thermistor becuse conduction heat tranfer rate >> convective (air) heat transfer rate.

Both the Orbs you used are designed for Athlon applications, not Celeron's...how certain are you that these slight design differences (in there really are any) didn't effect the outcome of your tests due to poor heat sink contact and other factors?

Has anyone else reproduced your results on other motherboard and CPU's...I'm particularly interested in TB tests in any exist.

 

[miniMUNCH]

Just one more question mike...when said you performed each test three times...did you set up each test case three times too or just once?

14C delta across...what?...less a millimeter of the chip core is really large!

 

[Mikewarrior2]

The thermistor is epoxied to the back. The test were run 3 times... IE system run for 1 hour, then shut down... allowed to cool off, then tested again.

The internal diode is calibrated for accuracy. Yes, the diode can vary +-Several C for any mb/cpu/slotket, but it is a constant "variation". TO calibrate it, we tested each cpu at 66mhz FSB, then at 100mhz FSB. Let's just say for example, the Temps were 34C and 41C. TO get the correction, you know that the CPU, when running the same program to load it, 100fsb will run 50% hotter compared to ambient case temp compared to the 66mhz FSB. So your diode is reading +1C high, so you manually enter the correction ot drop the reading 1C in MBM.

THe 14C delta is the change from internal diode reading to backside reading on the C-orb. The same thing occurs with socket-A cpus. YOu can't rely on backside temps as an indication of CPU core temp, but it is even worse with the ORB style heatsinks(arcticooler, S/C-orbs).

If anything, the C-orb/S-orb clip performs better on a P3/Celeron 2 than it does on a socket A. They excert more cpu pressure than is needed for p3s, so the cpu connection wasn't the problem, and is probably better than in most socket-a installs. The celeron2 core is slighly different size compared to a socket A setup, but not enough to drastically change the results.

There is a gap between socket-mb and socket-slotket in both slotkets and socket-A mb's. Whether this is the reason for the s-orb/c-orb malfing up socket-thermsitor readings(they exhibit the same, "low-thermistor-temp" in socket-A heatsink reviews) is unknown as of yet, but it is a problem that Thermaltake is getting sales based on "false-reviews".

NOW comes the real problem. You're asking me or other people for t-bird tests. You cannot perform t-bird tests. At best, they use a thermistor that contacts backside-cpu core edge. No internal diode readings, nothing remotely accurate. Socket-Thermistors as a whole are highly inaccurate and cannot measure temp changes properly.

Sometimes wtih compensation, if your cpu is a certain overclock range, will help approximate cpu core temp better, but the problem is the compensation is fixed, so it doesnt' help everybody.

 

[miniMUNCH]

Mike:

I see...so the thermistor measuring the 14C delta was mounted on back of plate that actually contacts the flip package...ya the temp reading would be way f'ed up.

You mentioned that the socket thermistors included on KT boards do not properly monitor temperature of the CPU core. I understand that the thermistor is contacting the core on the side opposite the main heat path but the thermistor is still contacting the heat source. How messed are the readings...3-5 degrees C or is it more? Do you have any links where I can read about this problem.

Thanks, by the way, for your info...good stuff!

One more question: if I were to buy a new cooling unit what would you suggest? A fop32? Also take into consideration I really don't want a fan that "howls"...how much louder in reality is the fop32 than the C-orb.

 

[Mikewarrior2]

If noise is your primary concern, look at the taisol CGK742092.

see my sig for more info on all of this.

As far as actually touching core, the kt7 thermistor does not. The cpu "core" is on the heatsink side of a ceramic PCB. It is not designed to radiate heat(the core is designed to do that). So while the heat coming off the ceramic pcb is significant, it isn't nearly what core temp is. IN fact, it is only a fraction of core temp(Ccore to Cback is at least .5 thermal resistance, then the kt7 is another 25-33% temp drop from directly behind the core).

Which brings up the kt7 UL bios compensation. A +10 to attempt to "better approximate" core temp. This still doesn't solve things for most people, but it helps. But the temperature change compression(results of measuring only a percentage of cpu core temp change) is still there.

Effects of temp compression and incorrect readings can be seen in any Socket-A heatsink roundup. Take the anandtech roundup, a 2C measured difference between c-orb and Alpha PAL6035. When actual difference would be around 20-25-30C in core temp.

On average, the kt7 can range anywhere from 0C off(rare) to 20+C off(also rare). THe median average that it is incorrect is usually 6-12C too low.


Mike