Real Temp - New temp program for Intel Core processors

[unclewebb]

concretefire: I'm not sure what you mean by question 1).

As for question 2), you can lap the IHS on your CPU but I wouldn't remove the IHS to do that. I'm pretty sure it is soldered on the Q6600 and you can end up damaging your CPU or end up with worse heat transfer if you do manage to remove it. The improvement in temperatures by lapping your CPU will depend on how square it is to begin with. If I was using an OEM cooler on a Quad, I'd start by replacing that.

I've just started reading over the new Core i7 documentation which is available here:
http://www.intel.com/design/corei7/documentation.htm

6.1.1 Thermal Specifications

"A single integer change in the PECI value corresponds to approximately 1 °C change in
processor temperature. Although each processors DTS is factory calibrated, the
accuracy of the DTS will vary from part to part and may also vary slightly with
temperature and voltage. In general, each integer change in PECI should equal a
temperature change between 0.9 °C and 1.1 °C."

At least with this new CPU generation, Intel is being a little more upfront about slope error from these sensors. That means the margin of error is plus or minus 10%.

[JohnZS]

JohnZS: Your sensors are still near the top of my list of bizarre sensors so thanks for posting that. I originally came up with the idea that some of these sensors move along different slopes. At the IDF, Intel jumped on that band wagon and agreed with that theory. With your sensors, slope error might be a minor issue but it doesn't seem to be the major issue. The slopes for all 4 cores are quite similar and none of your sensors are sticking.

Your data would lead one to believe that during calibration by Intel, TJMax may only be accurate to something like plus or minus 5C. The 45nm Atom sensors have Intel documented error amounts similar to that at TJMax but my contact says that the 45nm Core 2 Duo based sensors are not like that at all. After looking at your data, I guess I'd disagree. It's too bad that Intel has decided not to share the amount of error at TJMax with enthusiasts.

I recommend using Prime95 Small FFTs when testing but I tried using "Maximum Heat and Power Consumption" like you did and the results were very similar to Small FFTs. On Max Heat, the power consumption is very consistent at each level just like Small FFTs is. The maximum power consumption at the wall was a few watts less so the peak temperature was about 2C less when I tested but each step down in power was unique and very consistent. Linpack testing creates more heat but it doesn't have the consistency that Prime95 has.

I'm still looking at data at the moment. When testing, you'll get the best results if you make sure that there isn't a lot of junk running in the background. Vista has a lot of background baggage like SuperFetch you need to watch out for.

Thanks for the advice Unclewebb
I shall see if I can run a Small FFT's and kill most background processes. Yep I am starting to think that my sensors are just useless (might as well be generating random numbers), it's a shame as their randomness is really making me hesitant to see what speed I can push out of it.
I was thinking of playing it safe and asuming the sensor reading the highest temperature (Core1) is the only one telling the truth, however there is no telling how inaccurate it is...it could even be reading lower than what it should be!
bah
Thanks once again for your continued efforts
John

[unclewebb]

Don't give up yet JohnZS. Now that I can see and understand a little better what your sensors are actually doing, I might be able to come up with a better way to calibrate RealTemp to your unique CPU. As long as sensors are moving and not getting stuck there is hope. The Intel graph at the original IDF got me a little side tracked but I'm almost ready to take another stab at this.

[concretefire]

concretefire: I'm not sure what you mean by question 1)

Sorry. Look at my screenshot I posted (or any of them) At the top it says "Sensor Movement Test". I have gathered that part of it is to see if your sensors are actually working, or stuck......right? What I don't get is the numbers under the cores. Like for angelreaper's screenshot where you advised him one of his sensors was stuck. Core 0 = 7, Core 1 = 5,
Core 2 = 6 , and Core 3 = 0

I understand the numbers going from 100 down, (perfectly)

I just don't understand what the numbers under each core mean directly under the "Sensor Movement Test"
Thanks again for your help and great program.

[unclewebb]

Now I know what you're talking about!

The Sensor Movement Test is pretty simple. It reads the sensors to see what they are at when you start the test. Next it applies an equal load to each core for a few seconds and it checks the sensors again and compares them to what they were at before you started the test.

In theory, with an equal load applied to each core, you would expect that each sensor would change by an equal amount. A difference of 1 or 2 between cores is pretty typical. When you apply a load to a core and the sensor doesn't move at all then that is a pretty good sign that it is sticking at low temperatures.

In angelreaper's post, core3 failed both tests. It showed no movement during the first test and during the CPU Cool Down Test it doesn't change from the 50% level all the way down to the Idle level. If you don't have too many background tasks screwing up your test then you should expect to see lower temperatures / higher distance to TJMax at each step of the Cool Down Test.

With the original Sensor Movement Test, depending on MHz and core voltage, it was sometimes difficult to properly spot a stuck sensor. The Cool Down Test is a little more thorough.

[randomizer]

How extreme would your temps need to be before Intel accepted an RMA (after 2.5 years )? Idling at 37C delta to Tj Max with the stock cooler isn't exactly chilly but it's not playing with fire either.

[unclewebb]

At default MHz and core voltage can you make it hit the thermal throttle using Linpack? Maybe Intel will accept this as proof that something isn't quite right with your E6600.

[randomizer]

At default MHz and core voltage can you make it hit the thermal throttle using Linpack? Maybe Intel will accept this as proof that something isn't quite right with your E6600.

Not quite 7C delta to Tj Max is the highest it got when I let it keep going. Man it was noisy though.

[gymenii]

Testing E6850@4Ghz - not bad considering almost two years of constant abuse....

Works as advertised, Thanks

[unclewebb]

gymenii: The only problem with your old E6850 is that it runs so damn good that there isn't much need to spend a $1000 bucks to get a new Core i7. Your E6850 temperature sensors both move freely but show a little bit of slope error between cores. If you plotted your temperature curve on a piece of paper, you would see two lines at slightly different slopes until about 60C at which point both cores would line up and would continue on fairly equally until TJMax.

I'll send you a PM when 32nm arrives. Maybe you'll have a reason to upgrade then.

More screen shots of different versions of slope error on whatever type of CPU you have would be great. I'm getting more accurate, real world information out of these screen shots than anything I learned about slope error from the IDF presentations.

I finally found some Core i7 documentation publicly available on Intel's website yesterday. RealTemp is not yet working 100% correctly with Core i7 but I should have the problem fixed up and an updated version of RealTemp will be available later today.

[AKHandyman]

Where can I download the latest version? I see that there is 2.84, but I can't seem to find the link to get it ... am I missing something here?

[unclewebb]

I like to hide the good stuff for the insiders here at XS.
http://www.fileden.com/files/2008/3/...alTempBeta.zip

[AKHandyman]

You're the man ... thanks ... I just put my QX9770 in the Gigabyte EP45-UD3P board and it was auto-volting something fierce before I manually adjusted it. I wanted the most up-to-date version, and I knew where to come ...

Thanks again!

[unclewebb]

AKHandyman: Intel has recently stated that TJMax for your QX9770 is 85C but I'm not yet convinced about that. Testing at low MHz and low core voltage might be able to prove whether 85C is realistic or not. If your CPU is reporting core temperatures that seem too good to be true then let me know so we can run some tests on it.

[gymenii]

gymenii: The only problem with your old E6850 is that it runs so damn good that there isn't much need to spend a $1000 bucks to get a new Core i7.

I'll send you a PM when 32nm arrives. Maybe you'll have a reason to upgrade then.

Yep, plan to keep it for awhile. Maybe take a look at i7 next summer (or maybe not)

The addition of the Sensor Test (and accuracy) really sets RT far apart from the rest. Thanks for sticking with this

[AKHandyman]

AKHandyman: Intel has recently stated that TJMax for your QX9770 is 85C but I'm not yet convinced about that. Testing at low MHz and low core voltage might be able to prove whether 85C is realistic or not. If your CPU is reporting core temperatures that seem too good to be true then let me know so we can run some tests on it.

I am beginning to wonder ... after I loaded up the program, and ran the first test, the core temps were reporting ~10C LOWER! That though was just the difference from v2.6 to the current version 2.84 ... tell me what I have to do for these tests you'd like for me to run ...

[unclewebb]

AKHandyman: Between version 2.6 and 2.84, TJMax has gone from 95C to 85C for your processor so your core temperatures will be reported 10C less with the new version.

The first test you need to do is the new CPU Cool Down Test. At your normal MHz and core voltage, click on Sensor Test and then when that screen pops up click on Start and go get a coffee. The test will take about 10 minutes. There is an information window just above the Start button which will hopefully tell you what to do at each step if you're not sure.

This test uses Prime95 so if you don't have that installed on your computer then start by clicking on the Download Prime 95 button.

Make sure there isn't anything significant running in the background while this test is running. Give Vista a chance to settle down before starting this test.

When the test is done, post a screen shot or e-mail me a copy of your results. If you hold down ALT+PRINT SCREEN on your keyboard it will put a copy of the results into your Clipboard. Open up Paint or IrfanView and Paste and Save your results.

Send it to Real_Temp@yahoo.ca or head to www.imageshack.us and upload it there and then paste a link to it on XS so I can have a look.

Once we find out how good your sensors are then the next step is to run your computer at Ultra Idle. You can either go into the bios and manually set your core voltage to approximately 1.10 volts, and use default MHz (333MHz X 6.0) or you can simply enable C1E and SpeedStep in the bios. If your FSB is at the default speed of 333 MHz and your core voltage is set to AUTO, C1E should automatically drop your multiplier to 6.0 and should also drop your core voltage as well. Run CPU-Z to confirm your results at idle.

The goal is to get your CPU as cool as possible so your reported temperatures are as low as possible. No need to open your window for some fresh Alaska air; at least not yet.

It's a good idea to open your computer case though so there is no hot air being trapped. At idle, your reported core temperature should be about 8C or 9C above your room temperature depending on what type of cooling you're using. If this is off by 10C or 15C then either a sensor(s) is stuck or the Intel recommended TJMax you're using isn't correct. It's usually pretty obvious when something isn't quite right. I'm looking forward to your results. Let me know if there is anything you're not sure of.

[KURTZ]

this is my E8600 what about my sensors? are good or not? thx

[unclewebb]

KURTZ: Your results are not typical at all. Your sensors are not showing any signs of sticking which is good. The weird part is that in almost every case, the data from the sensors gets closer together as you get closer to TJMax. Your sensors are reporting the opposite and moving farther away from each other. I'm not too sure why.

What type of cooling are you using and how did you install it? As soon as you put a load to your CPU, core 1 starts running a little hotter. Nothing too major but it might have something to do with how you installed your heatsink or the paste or maybe your IHS or heatsink isn't quite square. A perfectionist like myself might pull the heatsink to have a look but if your computer is running great then I wouldn't worry too much about it.

[KURTZ]

thx Uncle for the reply

this is my system right now, i'm waiting some waterblocks so next week i put all my hw under the water ...



i'm under the intel's stock fan, and i put a little bit of Zalman STG-1 paste ... i do this test only cause i'm so curious ... :p


edit:



these are my friend's sensors ... he's under the water ... what do you think about?

[unclewebb]

Is it possible that you were a little cheap with the paste since you knew it was going to be coming apart next week?

It will be interesting to run the same test when you have a water block installed and see if anything changes. I've seen a lot of screen shots and yours just looked a little unusual.

Your friend has core0 sticking at 65 away from TJMax and core1 might be stuck as well at 66 from TJMax. The test is very conclusive for core0 but not conclusive for core1. You would need to test again with colder water to try to force the CPU to a cooler temperature and see if core1 can reach more than 66 away from TJMax. You could also redo the test with less MHz and core voltage when trying to get some lower CPU temperatures. Less core voltage is the most important one.

[KURTZ]

Is it possible that you were a little cheap with the paste since you knew it was going to be coming apart next week?

It will be interesting to run the same test when you have a water block installed and see if anything changes. I've seen a lot of screen shots and yours just looked a little unusual.

Dear Uncle you're definitely right! however next week i'll return here so we can compare these tests ...

[unclewebb]

Dear Uncle you're definitely right!

If RealTemp can be used to catch people not using their fair share of paste then this new sensor test feature will get two of these.


I look forward to your tests next week.

[msgclb]

The first thing I noticed is that you have to delete the configuration file when changing to a Core i7. I believe that I read that somewhere but I learned the hard way before I got Real Temp running.

I believe that I have a problem. I like to set Real Temp up as shown below. I don't want any temperature readings in the Taskbar, just in the Real Temp application. I'm wondering if the taskbar readings are from the Hyper-Threads. The next time I shutdown I'll disable HT and see what happens. I like to use Real Temp as shown in the second image.





I ran a Sensor Test using default BIOS options. The one thing that I noticed is my CPU Multiplier was all over the place during this test. I don't know what this says about my 920. The ambient temperature was about 26 degrees C.



Finally, to prove that there's a 21 multiplier using the P6T and a 920 here's what I started with before the Sensor Test.



Update:

You won't believe what I found was the problem. Well, maybe you would! When I went to shutdown I found that I had two instances of Real Temp that I couldn't get rid of. I had to kill them and when I did I got this.



I believe these were the two times I started Real Temp with old configuration files. Anyway that problem is fixed!

[unclewebb]

msgclb: RealTemp 2.84 is definitely not working properly with Core i7. Now that I have access to the Intel Core i7 documentation, I'm making progress. The fix is almost done and should be ready for testing before I go to bed tonight.