Sign in with
Sign up | Sign in
Your question

key to Wolfdale Temp reading mystery

Last response: in CPUs
Share
September 12, 2008 5:40:13 AM

Hi there! I was having major problems determining the *right* temperature of my then-new e8400.I used speedfan, everest, coretemp, pcwizard and also realtemp and basically they have all been showing DIFFERENT readings. Moreover, one core is always showing +10 to +25 degrees difference against the other core, which confused me even more. Then I started to do some web research and found a great guide about calibrating speedfan in order to get real temps (somwehere around here at THG forums). I collected some extensive amount of temp reading data, torturing the poor CPU at different frequencies and voltages. Then I realized that this dataset may be a great help in determining the *real* temp of your 45nm CPU. I assume that the thermal sensor readings of Core1 are wrong, quite obvious reason is that both other sensors (tCase and Core0) were reading close to each other, with Core1 temp 10-20 degrees celsius more, and also with high fluctuation of about 5 degs +/-. Please be aware of this fact when considering my data. These are the readings the mentioned programs actually showed me on the screen prior to any calibration. please, ASSUME THAT NO SINGLE NUMBER IN THE TABLE BELOW CORRESPONDS TO A REAL TEMP. The table is only to show relative differences in temps when increasing voltage, frequency and load. YOU CAN BASE YOUR TEMP CALIBRATION ON THESE DATA.
Short guide: please read the guide posted in this forum about calibrating speedfan (sorry dunno exact link now). You can find information about how to determine idle temp of your CPU when running at minimal speed and voltage. In my own case: the room temp was 22-23 degs. I'm using quite effective cooling (thermalright si-128se, well it's not the ultra extreme tower, but we can say it's a mid-to hi-end air cooler, plus artic cooling 5 silver compound. Considering this I could assume that my idle temp (case, not core) at 2000MHz@def voltage had to be around 25degs.Based on this number, which I'm pretty sure of, I set an offset for tcase (temp1) to -5degs. NOW WE COULD ASSUME THAT TEMP1 READING (TCASE) IN SPEEDFAN IS CORRECT. Then, after we have a look at core0 readiing in speedfan, we could also consider it to be ALMOST correct. The only flaw is that id reads LESS than temp1, and that's impossible. But the difference is about 2degs so we can say iit's witihin the tolerance relevant for our research (in fact it should read about 1-2 degs more than Temp1. THE READINGS OF CORE1 TEMP ARE COMPLETELY OFF AND I DID NOT CONSIDER THEM RELEVANT.
Now here is the table of raw readings.


Important findings:
1. Based on my observations, THERE IS NO ONE PROGRAM CURRENTLY AVAILABLE, THAT SHOWS ALL TEMPS OF WOLFDALE CORRECTLY. always bear this in mind! it's true.
2. Above written DOES ALSO APPLY TO BIOS READING! that means NOT EVEN BIOS READS THE TEMP CORRECTLY! (but it may do so, depending on the motherboard model and bios version - not my case, as you can see (MSI p35 Neo2FR with fintek chip and new BIOS version)
3. MANY OF 45nm CPUs HAVE FAULTY temp sensors. well, huh...it's not that simple though. Intel stated that these sensors (core temps) are designed in such way they TRIGGER THROTTLING AND EMERGENCY SHUTDOWN CORRECTLY, so THEY ARE NOT DESIGNED TO REPORT IDLE TEMPS. but I think that in my case it's quite obvious that Core1 sensor is faulty..This does apply only to core temps, the case temp sensor should read the temp correctly (you just have to set the proper offset, remember).
4. IDLE temp depends slightly on the core voltage set, and it depends very little on the actual frequency set. As you can see, diff in idle temp between min volt/speed (2000@1.15V) and max volt/speed (4000@1.45V) is only 5 degs.
5. LOAD TEMP (after approx. 20 minutes of Prime95) depends linearly on the CPU speed set, with a moderate dependency.
6. LOAD TEMP depends VERY HEAVILY on CORE VOLTAGE. as you may see, min frequency at min voltage was running some 10 degs cooler then min frequency at max voltage. This difference SCALES with cpu speed, so running 4000Mhz at min voltage (stability impossible though) compared to max voltage at same speed could result in gain of much more than 10 degs.

I'd be very happy if this data help you somehow. I've found it to be a crucial help for overclockers, cause it's quite obvious that knowing the real temp of CPU is important to determine your max safe overclock
(WARNING ! voltage of 1.45 is quite extreme for this type of CPU and I really don't recommend you to even set it, unless you have adequate cooling solution and you know exactly what you're doing. Max. specified voltage by intel is I think 1,35 or 1,3625. BELIEVE ME! 1.45 is really the max voltage wolfdale can sustain! I'VE SET MY VOLTAGE TO ABOUT 1.5 AND I HAVE UNRECOVERABLY DAMAGED MY CPU!!! today I'm gonna RMA it cause they should change piece for piece and I'll report faulty temp sensor to be a cause.

Explanation of the table: first three rows correspond to some random settings.then the next 7 rows demonstrate the relationship between frequency and temp (voltage was constant), and the last 4 rows (including the last row of the above seven) demonstrate the relationship between voltage and temp (frequency was constant). Then there are two vertical sections, one for idle and the other for load temps (after cca 20min of Prime 95). The grey columns show idle vs load difference.

Example of usage: You're preparing to do some overclocking. You just set up some program, let's say coretemp, to observe temps. Your current settings: 3500MHz @ 1,25V. At first, make sure you determine your lowest case temp properly (follow the instructions provided in the aforementioned guide.Now look at the table, and consider only the relative differences. Your lowest tcase is, let's say 27degs...at 2000MHz and min voltage.Now you have to add about 2 degs to determine current idle temp (2000-3500Mhz there's about 0-1degs diff, talbe shows 0, but I think it's safe to consider 1 deg..for the increased voltage let's add 1 deg, table shows 0, but the highest speed should account for some higher gain).well, so your temp should be about 29degs +/- some tolerance (I'll consider 1 degs tolerance). Now, your coretemp shows 38 degs for core0, and 67degs for core1. Based on this, the real temp of your core0 equals the one showed - 10 degs. Now you could assume that temp of your core0 is always *reading of coretemp - 10 degs*. The reading of second core is obviously wrong, so don't bother with it at all, possibly try to get another chip (but determine OC possibilities first ! :D )

This way you may very quickly calculate the approximate REAL temp of your CPU at ANY settings.Remember, you must consider many factors - room temp, your cooling solution efficiency, your case cooling efficiency, un/proper apllication of thermal compound and mounting of heatsink and everything that comes to your mind relevant to temp.


Hope this helps to solve some of the never-ending heap of 45nm CPU temp-related questions.I'll post something about my experiences RMA-ing my CPU in another topic. Maybe I'll get e8500 if it will be possible, for the higher multiplayer (should squeeze some 200-300Mhz more out of it). This guide relates ONLY TO 45 nm CPUs, it was written for dual core models, but you may also use some of the data for quad core models, however I recommend you to do YOUR OWN table for calculations. thanks for reading so much crap :D  Looking forward eagerly to your replies.bye
September 14, 2008 10:51:00 AM

man, this guide is quite good, but the form of it...add som spaces, colors etc. for better reading and orientation. the table is too big and doesnt fit in the article, try to cut it in more pieces. now you must use scrollbar located on the bottom of the article to move it right...its very unpractical. btw, nice trick with that "I have bad sensor and I have definitly not set 1.5V on on this chip" tactic with your RMA :)  good luck!
!