Real Temp - New temp program for Intel Core processors

Hehe, yea, I probably could.

Interesting thing last night webb. As per our PM's, it's almost uncanny that I did exactly what you mentioned b4 you replied. Great minds think alike I guess. Regardless, my temperatures are a much more believable 28,28,29,29 degC @ 26 Ambient. Considering that I am at a relatively low overclock using low voltage and under a high end single block water loop, that's about right in my book.

A few observations though:
1. the distance to TJMax numbers seem to update with a higher frequency than the Temp numbers. Is this normal? Actually the temp numbers barely move after the idle calibration. I did an activity test and results were 6,7,7,8. Interesting.

2. My Delta temp from idle to load on small fft's was only 6 deg. From 29 to 35. This is definitely an oddity for me (not impossible in any way mind you), It's just that normally I'm looking at 10-15 delta from idle to load. My idle temperatures went up substantially after the calibration, but the load temps are stellar. What are your thoughts on this?

What MHz and what does CPU-Z report for core voltage when running Prime95 small FFTs? I'll assume that you're using the multi core version of Prime and all 4 cores are maxed out in Task Manager. If you keep the volts down then the heat output from your CPU will also be down.

You need to do a fair comparison. With the same settings run small FFTs for the same length of time and write down your min/max temps with and without calibration. You can even run two instances of RealTemp for this test to compare calibrated to uncalibrated.

Prime95 v25.6 32 bit
Prime95 v25.6 64 bit

I've done a fair comparison, Prime defaults to 4 instances no matter what. All cores are maxed everytime I fire up Prime.

BIOS Settings were exactly the same from pre to post calibration.

In my Vdroop master (oops X48T) BIOS, Vcore is set to 1.225v for 3.2GHz. Everest CPUID reports 1.184v Idle/ 1.176v Load

My pre-calibration results are all documented over in the q6600 overclocking thread, I think you're subscribed to that one yes? My screenshots of post calibration Prime testing are at home, I'll post them here after work.

But in the meantime they can compared here

32,28,28,28 IDLE and 42, 40, 39,37 LOAD ----> Pre Calibration 30 DegC. Ambient. --> Roughly 10 Deg Delta

28,28,29,29 IDLE and 35, 36, 38,39 LOAD ----> Post Calibration 26 DegC Ambient. --> Roughly 8 Deg Delta for the most part but was as low as 6.

I was more or less wondering what you had observed for delta in your testing, not so much asking if your software was borked.

Good point about running two instances of RT though, I'll do that.

What are your thoughts about point #1 in my previous post?

Cheers,

Quote:

---

Originally Posted by unclewebb

I thought the TjMax conference was scheduled for Thursday at 1:40. I'll see if I can find a link to that. Every computer website will be rushing to break the news so I'm not too worried about not hearing about it.

Edit: Here's the official class:
TMTS001

---

Ah ok, I thought it was today and reported tomorrow. I guess we will all be watching in anticipation tomorrow.

By the way, was the data I sent you any good?

Keep up the good work and hopefully IDF is good kudos to you tomorrow
John

Uncleweb, I've got a small request. Could you compile a 64bit version of the exe? The 32bit exe works fine on the full version of 64bit WIndows OSs but it won't work on the 64bit skinny OSs like Windows PE because they'll only run native apps. I.E. 32bit PE will only run 32bit apps and 64bit PE will only run 64bit apps.

Not a big deal if you can't but I've been doing all of my tuning and testing in WinPE and have been without temp monitoring.

thanks

gtj: My old compiler doesn't support 64bit so native 64bit for RealTemp is a ways off in the future, if it happens.

JohnZS: Your data was good but your sensors are bad. :( They are very inconsistent from one core to the other. I'm sort of in a waiting mode until Intel releases some documentation for these sensors. Then I can have a look at some of the data I've been gathering over the years. T - 24 hours :)

Ol'Baditude: When testing you need to keep as many variables as consistent as possible. When you get home tonight test again with two versions of RealTemp running. What are you using to measure your air temperature near your computer? I pointed my IR gun at my digital house hold thermostat the other day when it was about 32C outside and there was a 2 degree difference in readings. The typical difference is 1C higher than the air temp near my computer. All these temp readings that we take as being 100% accurate because they are coming from a digital source seem to be far from that. Other users that have calibrated RealTemp on a Q6600 with air and water found that a good water set up has idle temps about 4C above their water temperature after it has idled for a while.

I think I understand point 1) in your previous post. If sensors read too low and you have to use a positive calibration factor then there will be some points where the Distance to TjMax reading will change by 1 but the reported temperature will not change. The DTS info coming from the chip might move from 20 to 60 or a range of 40 but if you calibrate it up to 25 at idle then the reported temperature is only going to move from 25 to 60. 40 different values from the digital thermal sensors have to be reduced or converted into 35 unique reported temperatures with my model. Write two columns of numbers and you'll understand the overlap that must occur.

Calibrating RealTemp is my band aid solution to less than perfect, non-linear sensors. Maybe when Intel comes clean tomorrow they will introduce some magical formula that everyone can use and agree on. If they reverse engineered RealTemp then I might be a little suspicious. :D Flattered but suspicious!

If there is a more accurate way to get temps out of these processors then I'm all for it but any new model has to be backed up with real world testing that I hope Intel has done plenty of.

I'm with you, unc. I wish we could attend IDF. I'm on pins and needles to know what Intel discloses.

Quote:

---

Originally Posted by unclewebb

gtj: My old compiler doesn't support 64bit so native 64bit for RealTemp is a ways off in the future, if it happens.

---

I could compile it for you. :)

Quote:

---

Originally Posted by gtj

I could compile it for you. :)

---

Then I'd have to give away my top secret code! I'll think about it, especially after tomorrow. :D

And what kind of advantage would be to compile RT in native 64bit form? The massive amounts of calculations when displaying FSB frequency? The awesome power required to change trey temperature numbers? Or the incredible number of calculations when ticking option to start RT at system startup?
I don't think so. It works under x64 and unclewebb should leave it there. There is loads of programs that are far more demanding, yet they simply refused to go native x64 way because there was absolutely no actual benefit from it. Apart from coding two separate programs with the exactly same content, wasting your time, online space and bandwidth. It's like pimping your car to go 500 km/h but you never ever excess 50 km/h anyway. Just makes absolutely no sense. It works, leave it at that.

Quote:

---

Originally Posted by unclewebb

gtj:

JohnZS[/B]: Your data was good but your sensors are bad. :( They are very inconsistent from one core to the other. I'm sort of in a waiting mode until Intel releases some documentation for these sensors. Then I can have a look at some of the data I've been gathering over the years. T - 24 hours :)
[B]

---

Oh no, :(
Perhaps I should recalibrate?
Thanks for your insight though Unclewebb.

RejZoR I'm not sure but I think gtj has some sort of fancy 64-bit environment which does not run 32bit executables at all, some sort of 64bit only PE Kernal but I'm not sure.....
Just vaguely remember reading something like that in a thread on here..
John

Quote:

---

Originally Posted by RejZoR

And what kind of advantage would be to compile RT in native 64bit form? The massive amounts of calculations when displaying FSB frequency? The awesome power required to change trey temperature numbers? Or the incredible number of calculations when ticking option to start RT at system startup?
I don't think so. It works under x64 and unclewebb should leave it there. There is loads of programs that are far more demanding, yet they simply refused to go native x64 way because there was absolutely no actual benefit from it. Apart from coding two separate programs with the exactly same content, wasting your time, online space and bandwidth. It's like pimping your car to go 500 km/h but you never ever excess 50 km/h anyway. Just makes absolutely no sense. It works, leave it at that.

---

Actually, if you read my original request, it states that the x64 version of Windows PE doesn't run x86 programs and vice versa. Part of what makes PE skinny is it doesn't have the WoW subsystem.

Quote:

---

Originally Posted by unclewebb

Ol'Baditude: When testing you need to keep as many variables as consistent as possible. When you get home tonight test again with two versions of RealTemp running. What are you using to measure your air temperature near your computer? I pointed my IR gun at my digital house hold thermostat the other day when it was about 32C outside and there was a 2 degree difference in readings. The typical difference is 1C higher than the air temp near my computer. All these temp readings that we take as being 100% accurate because they are coming from a digital source seem to be far from that. Other users that have calibrated RealTemp on a Q6600 with air and water found that a good water set up has idle temps about 4C above their water temperature after it has idled for a while.

I think I understand point 1) in your previous post. If sensors read too low and you have to use a positive calibration factor then there will be some points where the Distance to TjMax reading will change by 1 but the reported temperature will not change. The DTS info coming from the chip might move from 20 to 60 or a range of 40 but if you calibrate it up to 25 at idle then the reported temperature is only going to move from 25 to 60. 40 different values from the digital thermal sensors have to be reduced or converted into 35 unique reported temperatures with my model. Write two columns of numbers and you'll understand the overlap that must occur.

Calibrating RealTemp is my band aid solution to less than perfect, non-linear sensors. Maybe when Intel comes clean tomorrow they will introduce some magical formula that everyone can use and agree on. If they reverse engineered RealTemp then I might be a little suspicious. :D Flattered but suspicious!

If there is a more accurate way to get temps out of these processors then I'm all for it but any new model has to be backed up with real world testing that I hope Intel has done plenty of.

---

Will do webb. I'll just post back here when I'm done.
I use two analog thermometers. One is on my computer desk and one is on the wall on the other side of the room. They agree with each other and that is good enough for me. I do have an IR thermometer, I suppose I could point it at the back of my analog thermometer or something. lol.
Anyway, thx for the thorough answer.

Sorry John, didn't mean to insult your Quad. There's no reasonable explanation for the fact that the sensors on your center two cores that are only a few millimetres apart, are reading 14 to 15 degrees differently at idle. In the 65nm era, a good pair of sensors would report almost equal temperatures for two cores on the same die from idle all the way up if the load was well balanced.

Your Quad sensors are all over the place. I'm just waiting to see if Intel has a rabbit that they can pull out of their hat tomorrow to come up with a better way to get some temps out of these chips. My dream would be a second set of sensors that are 100% accurate from idle to TjMax that they forgot to mention them in their previous documentation. Probably not going to happen though.

Unc,

Nice job on the Info... Button. I like 2.74, but then again, I liked tham all. May I suggest replacing CPUID with Stepping? Is there space for 3 additional characters?

Hi Comp, the anticipation for tomorrow's presentation at the IDF is killing me! :D

At the moment, I don't think the Info... button of 2.74 will continue. Some of the reviews weren't great and I'm kind of biased against it. I like the old small button with maybe an i logo added to it. I might also try one of RejZoR's ideas about expanding that section to two lines and ditching the button all together. Some of the infrequently used info in there could be moved to the bloated Settings window.

I don't think I will change the CPUID title. CPU-Z is using the term Stepping to apply to only the last digit in this code.

The December 2007 manual titled:
Intel® Processor Identification and the CPUID Instruction

officially refers to this value in section 3.1.2 as the:
Processor Identification Signature

so perhaps I could shorten that to PIS ;)

When you look up a processor on the Intel site like this E8400:

they call it CPUID String:

I think RealTemp calling it CPUID is the most reasonable title for it.

Yes, nice idea. Two line so you can see everything at the same time. B4 you know it I won't even need CPUID anymore. Do it! Or the "i" would be a nice touch. But the info button has to go. lol.

Hey Unc.

Playing around with 2.73 and really like the ability to change the color schemes.:up: (the black background with white text is just sweet. fit's my color scheme and just works.) I think it gives a lot of flexibility to most users to find something that works with whatever color scheme/desktop background they're using.

I'd prefer to use mini mode almost exclusively for display mode if I have it up on the desktop (honestly though I just use the systray view mostly) but would like to be able to keep it in mini mode and minimize it (maybe be able to right ckick on it and choose minimize) so when I pop it back up, it's in mini mode. Maybe there's a way to do that and I just haven't figured it out..

either way, it just looks better every day..

First, put RT in mini mode. To minimize, double click on the core temperature system tray icons. To bring it back, double click on the system tray icons again and it will reappear in mini mode. You don't need right click functionality.

Thanks gbeans99. The new color schemes seem to be a hit. Easy to use and set up and lets users create something small that doesn't stick out on their desktop like previous gray versions did. Ol'Baditude's suggestion for minimizing and maximizing RealTemp works for me. I like software that is easy to use.

One trick I learned is that if you Tab to the Settings button in normal mode or left mouse click on it and then roll off of it without opening up the Settings window, you can double left click on RT and send it into Mini Mode and then hit the Enter key and get the Settings window to open up so you can adjust the Mini Mode color scheme and get a look at it while you're doing it.

Another tip is that if you are in Mini Mode and want to exit quickly you can just hit the ESC key on the keyboard.

A left mouse click on the RT icon in the top left corner during normal mode will bring up a menu where you can select the About box. A left mouse click on the TechPowerUp icon will bring you to the RT home page.

A double left mouse click on the small RT icon in the top bar will close down RealTemp. Lots of little features to make RealTemp easy to use.

Quote:

---

Originally Posted by unclewebb

Sorry John, didn't mean to insult your Quad. There's no reasonable explanation for the fact that the sensors on your center two cores that are only a few millimetres apart, are reading 14 to 15 degrees differently at idle. In the 65nm era, a good pair of sensors would report almost equal temperatures for two cores on the same die from idle all the way up if the load was well balanced.

Your Quad sensors are all over the place. I'm just waiting to see if Intel has a rabbit that they can pull out of their hat tomorrow to come up with a better way to get some temps out of these chips. My dream would be a second set of sensors that are 100% accurate from idle to TjMax that they forgot to mention them in their previous documentation. Probably not going to happen though.

---

That's ok mate you are only telling it how it is :D
Would remounting the HSF and recalibration help?
I guess the only thing which could help is IF there was a 2nd set of sensors (highly unlikely) or the middle cores had a different TjMAX value (more unlikely)....bah and to think this was a replacement from an RMA :mad:
John

JohnZS: You could try remounting the HSF but personally, I think the differences you are seeing are all related to sensor error. When you have one sensor reading too high and one too low and the combined error can be significant. I'm prepared today to be enlightened by Intel. I've given it my best shot to convert the sometimes screwy sensor data into some meaningful temperature numbers. Now I'm looking forward to how they suggest doing it.

I wrote a small utility program a while ago that lets you read information directly from the model specific registers within your CPU. If Intel announces some secret info hiding in there then you might be able to use this tool to discover what it is.

http://www.fileden.com/files/2008/3/3/1794507/MSR.zip

The digital thermal sensor data that all programs are using is in MSR 0x19C and is located in bits [22..16]. Avoid clicking on the Write MSR button unless you know what you're doing.

Whats the word on the Intel thing Uwebb?

Quote:

---

Originally Posted by unclewebb

JohnZS: You could try remounting the HSF but personally, I think the differences you are seeing are all related to sensor error. When you have one sensor reading too high and one too low and the combined error can be significant. I'm prepared today to be enlightened by Intel. I've given it my best shot to convert the sometimes screwy sensor data into some meaningful temperature numbers. Now I'm looking forward to how they suggest doing it.

I wrote a small utility program a while ago that lets you read information directly from the model specific registers within your CPU. If Intel announces some secret info hiding in there then you might be able to use this tool to discover what it is.

http://www.fileden.com/files/2008/3/3/1794507/MSR.zip

The digital thermal sensor data that all programs are using is in MSR 0x19C and is located in bits [22..16]. Avoid clicking on the Write MSR button unless you know what you're doing.

---

Thanks :up:
It's amazing just how much information is crammed all into the small space of 45nms
I am not quite sure if this is the best thing to do, but for the time being I am always amusing that the core reading the highest temperature is "telling the truth". Far better to be cautious....fingers crossed intel say that the 45nm quads have a slightly different register or something or maybe even 1 backup sensor!?!
FYI the previous QX9650 which I had RMA'd had terrible sensor problems, it was RMA'd in 2007 because core 3 kept Proc Hotting and causing the system to crash repeatedly. the sensors were all over the place...infact far worse than my current QX9650.
Now..lets see if Intel can fix this... or at least do something to make things easier for us all.
John