Real Temp - New temp program for Intel Core processors

[l!kw!dskw!d]

With the new version 2.9 out I take the opportunity to say THANK YOU for this sweet little tool!
As for myself, I think using RealTemp together with the linpack intelburntest made Core2-OCing both a lot more reliable AND less time consuming (or leaving more time for some more thorough testing).
So again, thank you, and keep up the good work mate !

[emoners]

With the new version 2.9 out I take the opportunity to say THANK YOU for this sweet little tool!
As for myself, I think using RealTemp together with the linpack intelburntest made Core2-OCing both a lot more reliable AND less time consuming (or leaving more time for some more thorough testing).
So again, thank you, and keep up the good work mate !

+1

[unclewebb]

Thanks guys. I'm pretty happy too with how RealTemp has shaped up.
I have plans for a few more useful features as well.

[Brian75]

I just finally finished putting my rig together and immediately installed this tool to be sure my temps are proper before I do any kind of OC'ing.

I've read through the documentation provided (particularly the calibration section) as well as through a myriad of pages in this thread. Based on what I've read, it seems normal for 45 nm Quad cores to have temperature readings that are off and more commonly where the Core 0 and Core 1 are close to each other, and Core 2 and Core 3 are close to each other, with a potential difference of 5-10 degrees Celsius between the first two cores against the second two cores.

In that case, is it far too unusual that these were what my temp readings were (w/o calibration):

Core 0: 30
Core 1: 33
Core 2: 10
Core 3: 14

I downloaded Core Temp as well just to see what it showed and it was in fact relatively the same.

Through my readings, I know this could be caused by a number of factors such as:

a) Bad sensors as Intel did not calibrate based on idle temps
b) IHS is not seated properly and/or contact points are not even

Sorry for the long-winded background, but my question is, has anyone seen variance this drastic? If so, would it be more probable that my IHS is not seated properly or bad sensors?

System Specifications:
Asus P5Q-E
Intel Q9450 Engineering Sample Stepping C1
Lapped True 120 Extreme w/ Push/Pull
Antec P180

Ambient Temperature: 21 degrees celsius

Thank you everyone for your insight, especially Unclewebb.

[burebista]

Try a CPU Cool Down Test to see if you have some stuck sensor(s).

[unclewebb]

Brian75: Nice timing. I just wrote yesterday how there is a lot of evidence to suggest that Intel is using different TJMax values on core0/core1 than what they use on core2/core3 in many Quads. 45nm and 65nm. It almost seems when they are building a Quad they reach into one bin and pull out a Dual Core with a near normal TJMax for core0/core1 and use a high TJMax Dual Core in the other side.

Your results are certainly at the extreme end of the normal curve and you likely have a difference in TJMax, maybe a stuck sensor or two and probably a fair amount of slope error as well. You've definitely come to the right place to try and make some sense out of your sensors.

Most other temperature monitoring software has decided it's easier to simply ignore these issues that Intel readily admits do exist.

Start by downloading the latest beta if you haven't already.
RealTemp 2.90RC1

At default MHz and core voltage, run the CPU Cool Down Test and post your results. This simple test quickly tells me a pile of information about your sensors. If you're allowed to post images then upload it to Imageshack and post a link to it here or else you can send it to me at:
Real_Temp@yahoo.ca

Also try running your CPU idle at ~2000 MHz and 1.10 volts. Manually adjust it to these settings in your bios if you can or else just make sure C1E and SpeedStep are enabled. Send a screen shot of CPU-Z and RealTemp so I can have a look. Include your room temperature and make sure your case is open for best results. Try to get your CPU as cool as possible is the goal so we can compare that to your air or water temperature. Be specific about what cooling you're using.

Processors like yours can provide me with a lot of information about just how crazy sensor data can get. Once I know what's going on, I'll come up with some recommendations for calibration. I should have lots of time tomorrow for this.

[jas420221]

Unclewebb, is there a quad esitmator for idle temps like RGE completed in your link on the last page?

[stasio]

Start by downloading the latest beta if you haven't already.
RealTemp 2.90RC1

RealTemp 2.90RC1 link didn't work

[burebista]

Try this link.

[Brian75]

Thanks for the super quick response. Also, though I've read through a myriad of pages in this thread, if I missed some procedure or recommendation previously on using Real Temp, I apologize.

Per your request UncleWebb, I've done the following:

1) Opened up case
2) Set case on it's side so that the TRUE is upright (I know this wasn't requested, but just wanted to be sure gravity wasn't playing a factor).
3) Bios settings:

vCore (manual): 1.10
FSB (manual): 266
CPU ratio (manual): 6.0
Speedstep and C1E: Enabled

Ambient temperature is appoximately: 21 degrees celsius
Cooling (Air): Thermalright Ultra Extreme 120 w/ 2 Scythe Fans Push/Pull

Screenshots of Real Temp and CPU-Z:

1) Pre-Cool Down Test - http://www.idealabs.net/RealTemp/temp1.jpg
2) Post-Cool Down Test - http://www.idealabs.net/RealTemp/temp3.jpg

Thanks again for all of your help!

[randomizer]

Wow, that's just... wow. Sticking sensors at 98.7% load?

[unclewebb]

jas420221: I ran a Q6600 G0 at 1600 MHz and 1.08 volts and I ran an E8400 C0 at the exact same settings on the same motherboard, etc. The difference at the wall was only 4 watts more for the Quad. Based on other testing, my best guess is that a difference in power consumption of 4 watts at idle is going to translate into a difference of only 1C higher for a Quad compared to rge's testing of an E8400.

That's the good thing about using low volts and low MHz. The difference between the majority of CPUs that Intel produces is very minimal. My 65nm E2160 with 1MB of cache at these same settings was 4 watts less than an E8400 so for a processor similar to that, I would use 1C less than rge's recommendations.

Brian75: At least you took some time and did your homework before showing up to class here at XS. I appreciate that and I also understand that not many people have time to read 110+ pages.

Can you try doing a second cool down test at your normal core voltage and MHz. If you overclock, that's fine. Run a test at those settings. That will help give me a better picture of your sensors. I need some data from higher up in the temperature curve. I think randomizer is impressed by your core0 sensor. That's typically the good one!

[Brian75]

Unclewebb, thanks for all of your insight thus far. I was quite impressed with Core 0 too...haha.

Anyway, I don't have any original voltages or clocks as I did not begin to overclock yet. My reasoning was that I wanted the temperature gauges to be somewhat accurate for me before I do so.

Do you suggest me go ahead and proceed with my attempts overclock then report back with the Cool Down Test then?

Naturally, this may take some time, but as I am eager to get things moving, I don't mind spending the time to get the data ASAP.

I know there's no way for you to really tell, but could this be any indication that the IHS is not seated properly? My original hunch is no, because that's the whole point of having idle temps in that they should be close to each other in the first place given ideal sensors.

Brian

[unclewebb]

It's time to get overclocking that Q9450.

This is going to sound a little crazy but your sensors aren't bad at all. Sure, they're going to need some manipulation but it's far from a lost cause. Once you start overclocking and create some heat in your cores, things are going to start looking a lot better. I see 3 cores right now (1,2,3) that are going to be very usable and core0 might even start to show some life once you get the MHz and core voltage up. Tell your Quad the holiday is over. It's time to see what she'll do.

I'd start at 8x400 with about 1.20 volts for the core and go from there. Use some conservative memory timings to keep that out of the equation. If you get it stable here then try another Cool Down Test. Depending on your board and memory, 8 x 450 might be quite doable without too much work.

I ran my E8400 for 3 hours a couple of weeks ago with it maxed out at 98C the whole time bouncing off the thermal throttle. I needed to run Prime 95 Small FFTs and I had to turn my CPU fan off to create and maintain that kind of heat. I like having TM2 enabled in the bios to keep the temps from going nuclear during extreme testing. You don't have to worry about that but it's still a good safety feature.

You have an excellent CPU cooler. The only problem you're going to have is getting your Quad to create enough heat to get your core temperatures up into a range where your sensors are reasonably accurate. You've got a long ways to go.

[Brian75]

Sounds like a plan to me. I was a bit hesitant originally as I did not want to damage my CPU, but my guess is that I'm far from having that happen anyway.

My goal is to get to 3.6 Ghz - essentially what you were saying, 8 x 450.

I'll get cranking on this over the weekend and will report results ASAP.

On a side note, do you think I should even reseat my heatsink to rule out variables? Though I prefer not to, because the process is quite arduous itself, if the end justifies the means, then I'm all for it.

Thanks again!

Brian

[jas420221]

jas420221: I ran a Q6600 G0 at 1600 MHz and 1.08 volts and I ran an E8400 C0 at the exact same settings on the same motherboard, etc. The difference at the wall was only 4 watts more for the Quad. Based on other testing, my best guess is that a difference in power consumption of 4 watts at idle is going to translate into a difference of only 1C higher for a Quad compared to rge's testing of an E8400.

That's the good thing about using low volts and low MHz. The difference between the majority of CPUs that Intel produces is very minimal. My 65nm E2160 with 1MB of cache at these same settings was 4 watts less than an E8400 so for a processor similar to that, I would use 1C less than rge's recommendations.

Would you believe the variance to be much different with a Q9450??

[unclewebb]

Would you believe the variance to be much different with a Q9450??

No.

I was expecting that there would be a bigger difference between 4 x 65nm cores in a Q6600 compared to 2 x 45nm cores in an E8400. The Quad also has an extra 2MB of cache (2 x 4MB vs 6MB)

There's a huge difference between these two when they are both running Prime Small FFTs at 3000 MHz but at low MHz and low voltage, the difference was only 4 watts, and that's at the plug. I think the official numbers were 136 watts vs 132 watts and the E2160 was at 128 watts.

With the CPU Cool Down Test and a Kill A Watt meter, it's very easy to see the relationship between power consumption and changes in core temperature. The difference in power consumption between a Q6600 and a Q9450 would be difficult to measure accurately because the difference would be so small. Temperature wise, at these settings, they should also be almost identical.

Brian75: I would not waste time re-seating your heatsink. The cool down test has already shown me that your CPU is behaving as it should. The difference in temperature between idle and small FFTs where you tested is in line with expectations. I don't see a problem with IHS to core contact either. Get it up to 8x450 and let's see some more numbers. You're probably going to have to use a split TJMax for your CPU. Something like 100C for the first two cores and ~110C for core2/core3. If you can and if you need to, for your next test, turn off one of your CPU fans as well to create some heat. The Thermalright is so efficient that it takes some heat to get your sensors up into a range where the data starts to become more accurate.

You'll likely need some slope error correction as well to get these 4 to line up reasonably well. That's why I need to see some more data. When we're all done, you're going to be surprised at the decent temperature data these crappy sensors are going to be able to provide you with.

[Brian75]

Unclewebb, that's a relief to hear that I don't have to re-seat the heat sink.

I'll get plunking along on this and if I'm lucky may be able to get to the results I desire in the next couple of days.

Brian

[HuffPCair]

@Unclewebb - have you thought of making it so you can put real temp into vista sidebar??

[emoners]

@uncle: that avatar suits u!

[unclewebb]

... have you thought of making it so you can put real temp into vista sidebar??

I've thought about it and I know there are lots of users that would like to have some RT information in their sidebar but I never use that feature so it hasn't been high priority. I'll add almost any feature, sooner or later, if I get nagged enough about it.

that avatar suits u!

That's how the icon looks when it hasn't been shrunken down too much. You can almost see the serial number on that thoroughly abused E8400. I Photoshoped the digits off of the Fluke onto the chip. That design was my daughter's idea so it's a keeper. She gets kind of bummed out though. She keeps asking, "Doesn't anybody like my icon? What are they saying about it?" She takes her icon about as personal as I take my program!

[[XC] 4X4N]

Just tried out the the new 2.9 test sensors. Awesome feature This program just keeps getting better, thanks for all you've put into it.

Tell your daughter the icon is great.

[Wotcher]

She takes her icon about as personal as I take my program!

Both the programme and the icon are world class, thanks for all the work you have put into it.

[jasjeet]

have i done this correctly?

[unclewebb]

jasjeet: It looks about right. It also looks like it has the issue that I've talked about recently where core0/core1 likely have a different TJMax than core2/core3. If you're interested, try setting TJMax to 100,100,105,105 and then adjust your Calibration factors to balance your idle temps again. The temps coming from my Q6600, which looks very similar to yours, look a lot better balanced and more realistic between idle and TJMax when set this way. Start and stop Prime95 Small FFTs with 4 cores loaded and watch how the temperatures look a lot more consistent from one core to the next.

Intel has said that TJMax is not a written in stone kind of number and there is some variation from one part to the next. They just haven't said how much variation there is. Quads consist of two separate Dual Cores and it sure looks like unbalanced TJMax is playing a factor in reported temperatures.