Real Temp - New temp program for Intel Core processors

[unclewebb]

I have an obviously stuck Core0 sensor. Yet real temp says its working fine?? What gives?

Come on now. Let's not beat up RealTemp because of Intel's crappy sensors!

Maybe your computer fell off the back of the truck during the move? Most of the sensors I've heard about are consistent. They'll always get stuck at the same fixed temperature. I did hear from one user that had one core where the sticking point changed from one value to another after initial break in. I don't think that is too common but it does happen.

When trying to use these sensors to report reasonably accurate idle temperatures, there can be a multitude of problems that over lap and screw up your results. The best way for me to understand your CPU better is if you run a quick test and send me the RealTempLog.txt file either by PM or send it to the e-mail address in the About... box of RealTemp.

I like to see 1 minute of idle followed by 2 minutes of Prime small FFTs followed by 1 minute of idle. You don't have to be exact with the amount of data you send me but generally about 5 minutes of data recorded with a log interval of 1 second really helps me see what's going on compared to just a single screen shot. The temperature curve is very consistent with small FFTs which is why I recommend it.

The 32 or 64 bit version is available here:
http://www.neowin.net/news/software/...30/prime95-256

When you send me your data include some details like CPU model, room temperature, MHz, core voltage, case open or closed, etc. The more info you send me the better chance I'll have figuring out your problem.

What does the RealTemp Sensor Test show now? Make sure you're computer is nice and idle when you run this test and show us a screen shot of what it says.



I need about 1.37 volts to keep my E8400 happy at 4 GHz so I should be getting more Movement than you during this test. If you have a 6C to 7C difference in your idle temps, this test should also show a significant difference.

If you're rock solid at 4.05 GHz with 1.28 volts then accurate idle core temperatures aren't that important. The sensors weren't designed for that purpose so be happy with that chip.

Fungus: More font options or a font picker is a good idea for the System Tray. The problem I found was that there is only 16x16 pixels available and you need room for two digits so each digit can only cover 8 pixels wide x 16 pixels tall. That's not a lot of room to work with and I picked through a pile of fonts and most of them looked like crap at that size. What screen resolution are you using and OS? I think at higher resolutions the system tray area might jump up to 32x32 pixels which might give us some more possibilities. Why not send me a screen shot of just your system tray area with about an extra half inch on each side so I can head to PhotoShop and count some pixels.

One option that RealTemp has is you can use the included RTFont. With a font designer program like Fony, you can try editing the RTFont and create something that you like.

1) Check your Font folder in the Control Panel to make sure RTFont is installed.
2) Start up Fony and select File -> Open Installed Font... from the menu or click on the little green open folder in the tool area.
3) Select RTFont from the menu.
4) Use Fony to edit each character.
5) When you're done, select the menu item File -> Install Font

This will install your new creation into your system. Run RealTemp and select the right most font option and it will use your new RTFont.

ClearType and anti-aliasing can give some weird results when working on small fonts so both options are turned off. I'll try using a font picker in the future for more system tray options. I spent so many hours trying to find some readable fonts for the tray that I thought I'd do users a favor and save them some hassle by limiting their font options. There's a reason why most programs don't give you too many options for system tray fonts.

[mpower1001]

When windows vista 64bit ultimate starts, real-temp won't start minimized, i did check the box in the settings tab, but every time windows starts it won't start minimized...Anyone

BTW: I'm using 2.75

[Fungus]

Oki doki, here is shot of my systray. 1920*1200 res

I'm not much of an artist, do dunno if I could make a font or not =]

maybe I can use reshacker and take the font from rivatuner... will try.

[unclewebb]

When windows vista 64bit ultimate starts, real-temp won't start minimized,

RealTemp has had a Start Minimized problem for a LONG time with Vista 64 and maybe even with Vista 32 as well. The very patient Burebista over at TechPowerUp helped me with testing recently and I think we've finally got this problem solved. I'll post an updated beta version of RealTemp within 6 hours so you can see if it's finally been fixed.

Fungus: That Fony program I recommended makes editing a pre-existing font simple. In your pic, are 55 and 69 from RivaTuner? It would be simple with Fony to modify the RTFont to look like that. I'll do it for you when I get the chance. Just send me some pics of as many different digits as possible. Small pics in bmp format are best.

Edit: Here's a 3 minute mod job on the standard RTFont. Uninstall the RTFont from your Font directory and try using this one instead. I'll shorten it a little if you like. The right most font option in RealTemp will use this one.
http://www.fileden.com/files/2008/3/...507/RTFont.zip

[Fungus]

That's thicker and readable alright. Could you do the same to font #2? (2nd from left) I like the height of that one better. =]

And yeah the other temps are from rivatuner.

[unclewebb]

rge and others should find the next paragraph very interesting. It comes from the Mobile Intel® Atom™ Processor N270 Single Core datasheet which also uses 45nm digital thermal sensors similar to, if not exactly the same as, what the E8000, E7000 and 45nm Quads use. Users striving for +/- 1°C accuracy with these sensors are dreaming.

5.3 Digital Thermal Sensor
The digital thermal sensor (DTS) accuracy is in the order of -5°C ~ +10°C around
90°C; it deteriorates to ±10°C at 50°C. The DTS temperature reading saturates at
some temperature below 50°C. Any DTS reading below 50°C should be considered to
indicate only a temperature below 50°C and not a specific temperature. External
thermal sensor with “BJT” model is required to read thermal diode temperature.

http://download.intel.com/design/pro...hts/320032.pdf

The amount of error at the calibration point is worse than expected but the ±10°C at 50°C spec is exactly the magnitude of slope error that users have been showing me and I'm seeing with my E8400. I guess I should be happy because I'm only out by about 10°C when the actual core temperature has dropped to 30°C.

It's certainly possible that the sensors used in the 45nm Dual and Quad Core chips are far better than this but I wouldn't exactly bank on that.

The reality is that Intel is probably surprised that we're getting any sort of accurate temps out of these sensors at all. I wonder why Intel didn't have the guts to share this kind of information at the IDF in August?

[unclewebb]

The Start Minimized option has been re-worked one more time:

http://www.fileden.com/files/2008/3/...alTempBeta.zip

It's always worked for me in XP but many Vista users, mostly x64 I think, have had troubles. If you've had problems and this fixes it then let me know. If it used to work but now it's broken then I guess I'll have to have another look.

It was working great for burebista in Vista x64 but I made a few changes since his last test session. If it's still broken then I guess I'm to blame.

Edit: I just added a bold font option to the first 3 fonts in RealTemp. I'll probably add this to the GUI in the future but at the moment just add this to your RealTemp.ini file:

Bold=1

It looks a lot more readable on 16:10 type flat panel screens. I think it looks better than RivaTuner!

[loonym]

As always, nicely done unc! So in a nutshell intel sensors are pretty much useless except as an amusement until the temps start approaching the point of throttling?

[Fungus]

Thanks uncleweb, it looks awesome now =]

[rge]

rge and others should find the next paragraph very interesting. It comes from the Mobile Intel® Atom™ Processor N270 Single Core datasheet which also uses 45nm digital thermal sensors similar to, if not exactly the same as, what the E8000, E7000 and 45nm Quads use. Users striving for +/- 1°C accuracy with these sensors are dreaming.

5.3 Digital Thermal Sensor
The digital thermal sensor (DTS) accuracy is in the order of -5°C ~ +10°C around
90°C; it deteriorates to ±10°C at 50°C. The DTS temperature reading saturates at
some temperature below 50°C. Any DTS reading below 50°C should be considered to
indicate only a temperature below 50°C and not a specific temperature. External
thermal sensor with “BJT” model is required to read thermal diode temperature.

http://download.intel.com/design/pro...hts/320032.pdf

The amount of error at the calibration point is worse than expected but the ±10°C at 50°C spec is exactly the magnitude of slope error that users have been showing me and I'm seeing with my E8400. I guess I should be happy because I'm only out by about 10°C when the actual core temperature has dropped to 30°C.

It's certainly possible that the sensors used in the 45nm Dual and Quad Core chips are far better than this but I wouldn't exactly bank on that.

The reality is that Intel is probably surprised that we're getting any sort of accurate temps out of these sensors at all. I wonder why Intel didn't have the guts to share this kind of information at the IDF in August?

That is interesting, and finally confirms from intel what national semiconductor had posted on their site. I was assuming national semiconductors figures were at least a little exaggerated...until reading that.

National semiconductor (one of the main suppliers for DTS for intel) claims that if using the 11 bit sensor with Beta correction trutherm enabled for 45nm (~30 cents more expense per sensor, but was $1 more last year) there is less than 3C of error up to 125C. If not using the 11 bit sensor with Trutherm beta correction enabled, they quote error as 5 to 30C on 45nm, and spread between sensors (cores) can be 5C or more.
http://www.national.com/analog/tempsensors/trutherm

Intel has used some of the newer sensors on some laptops...but with national semiconductor highlighting the must have beta correction which makes them accurate, I am wondering if even the laptops are accurate.

Maybe for Nehalem, intel is springing for the additional ~30cents per sensor (especially since cost diff. is now less) for the accurate ones for desktops.

But with that confirmation by intel...and knowing sensor on E8400 is 7 bit (?LM75) and not trutherm...now we know for sure about desktops.

[Brama]

But with that confirmation by intel...and knowing sensor on E8400 is 7 bit (?LM75) and not trutherm...now we know for sure about desktops.

Have I correctly understood that Intel buy the DTS from NS and put it inside the cpu package? Are you sure?

I thought the DTS sensors are part of the silicon die that Intel fabs produce.

Please clarify, thanks.

[unclewebb]

So in a nutshell Intel sensors are pretty much useless except as an amusement until the temps start approaching the point of throttling?

In a nutshell, I couldn't sell 45nm sensors at a flea market and keep a straight face if the person buying it was going to try and use it for accurate absolute temperatures. Psst, heh there buddy, wanna buy a sensor, cheap?

Intel has hinted more than once or twice that these sensors are good enough for thermal throttle and thermal shut down control but we shouldn't be using them to report absolute temperatures. I think the 65nm sensors had a lot tighter tolerances than that but since Intel doesn't like to talk too much about this subject, I guess we'll never know. Perhaps all of the 45nm processors sent back as RMA because of crappy sensors got the bean counters thinking that spending a few more pennies for Core i7 sensors wouldn't be such a bad idea. I think the Test Sensor button in RealTemp was able to help with that.

Thanks unclewebb, it looks awesome now

You're welcome. But, is it as good as RivaTuner? You better post another screen shot of your system tray so we can vote!

[rge]

Have I correctly understood that Intel buy the DTS from NS and put it inside the cpu package? Are you sure?

I thought the DTS sensors are part of the silicon die that Intel fabs produce.

Please clarify, thanks.

I would not claim to be sure about anything intel does.

According to intel slides and papers....several of the transistors on each core function as diodes and are located in potential hots spots, such as L1, DTLB, etc. The Digital sensor then scans each of these diodes on core and reports the highest on each core as core temp. The transistors functioning as diodes on cores, yes no doubt, they should be intel manufactured. The question is who makes the sensor that scans, filters, reads the diodes.

National semiconductor sells intel such sensors for 45nm according to their site, as to what intel does with them, is just an assumption....as always with intel it is reading and guesswork, then intel releasing just enough info...to make you redo everything

But if National's claim is even half true, that it has the only 45nm sensor with logic capable of filtering out the increased noise from the close proximity of the diodes on the 45nm and thus accurately reporting temps across the range to +/- 3C, then if intel is not outsourcing the sensor on desktops, maybe they should be Or maybe intel has caught up with national in sensor tech, and will implement their own accurate one in i7.

Edit found slide:

[ihjones]

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

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.

Interesting and thx unclewebb, So if I Read from the MSR at 0x19C and get 0x88410000 What exactly does this translate to? 2285961216 maybe?? if so then what is 85 96 12 ?? plz

[Fungus]

Ok, Vote on =]

[unclewebb]

Fungus: At least they're readable on your monitor now. I'm kind of biased but I think RealTemp looks better!
It's a nice addition with a minimal amount of extra programming. I like that.

So if I Read from the MSR at 0x19C and get 0x88410000 What exactly does this translate to?

Here's the "Reading Hex Digits 101" class. The first digits 0x just show that the next 8 digits are all in hexidecimal format. Each individual digit after 0x represents 4 bits of data. 8 digits x 4 bits/digit = 32 bits of info. The numbering scheme used by Intel in their documents is that the bits are counted starting at zero on the far right up to bit 31 on the far left.

The first 8 translates to 1000 in binary.

Bit Position
31 - 30 - 29 - 28
 1 -  0 -  0 -  0

Therefore, bit 31 is set and bits 30, 29 and 28 are all clear.

The docs say bits [22..16] is where the important temperature data is. The 4 zero digits in your example represents bits 15 to bit 0. Bits [22..16] contains 41.

Bit Position
 23 . 22 . 21 . 20 .. 19 . 18 . 17 . 16
( 0 .  1 .  0 .  0 ) ( 0 .  0 .  0 .  1 )    binary
         4                    1              hex

Bit 23 is ignored and not needed in this example.

Head to Google and type in
0x41 in decimal

and the answer will be 65 in decimal.

This means that your processor is 65 positions away from TjMax. If these sensors moved linearly with changes in core temperature then you could say that you are 65 degrees away from TjMax. Unfortunately these sensors, especially the 45nm sensors, will move either faster or slower than changes in core temperature so saying that you are a distance of 65 away from TjMax is more accurate.

Chapter 13.5 THERMAL MONITORING AND PROTECTION in Volume 3A of the Intel® 64 and IA-32 Architectures Software Developer’s Manual explains this MSR in great detail. Volume 3B explains lots of other information stored away in various other MSRs of Intel Core processors. In typical Intel fashion, some of the info is either vague or labeled as Reserved. That means you must guess or sign an NDA and give them a good reason why you need to know more of this top secret info.

I was thinking about adding a few more features to my MSR tool. Being able to enter in [22..16] and have the tool automatically read and display just that data would make life simpler. If you want to do some investigating of other MSRs then I'll think about doing an upgrade to this tool. Poking around with this tool is a good way to learn about Intel processors at a very low level.

[ihjones]

Thanks for the 101 lol I'll re think this one now that I have a better insight!! THANKS

So if I'm not mistaken one can use the MSR tool to find the closest and farthest from TjMax to calibrate RealTemp with??

Wait, I missed something! It would be nice if Intel published TjMax temps for an x9100 mobile or for that matter any processor

more reading......

[unclewebb]

Most of the 45nm mobile chips are TjMax=105C including the X9000 CPU.
http://processorfinder.intel.com/det...px?sSpec=SLAQJ
http://processorfinder.intel.com/det...px?sSpec=SLAZ3

Intel doesn't yet have any TjMax data listed for the Q9100 mobile CPU here:
http://processorfinder.intel.com/det...px?sSpec=SLB5G

But they do list it as TjMax=105C here:
http://ark.intel.com/cpu.aspx?groupID=37033

Now the only problem is that even with the correct TjMax, there is no easy way to calibrate a mobile processor. The amount of error in some 45nm sensors borders on the obscene so your reported temperatures may be little better than what a random number generator would display. For this reason and the fact that these sensors can bottom out and get stuck at temperatures as high as 50C, Intel doesn't recommend using these sensors to report absolute temperatures.

[ihjones]

Most of the 45nm mobile chips are TjMax=105C including the X9000 CPU.
http://processorfinder.intel.com/det...px?sSpec=SLAQJ
http://processorfinder.intel.com/det...px?sSpec=SLAZ3

Intel doesn't yet have any TjMax data listed for the Q9100 mobile CPU here:
http://processorfinder.intel.com/det...px?sSpec=SLB5G

But they do list it as TjMax=105C here:
http://ark.intel.com/cpu.aspx?groupID=37033

Now the only problem is that even with the correct TjMax, there is no easy way to calibrate a mobile processor. The amount of error in some 45nm sensors borders on the obscene so your reported temperatures may be little better than what a random number generator would display. For this reason and the fact that these sensors can bottom out and get stuck at temperatures as high as 50C, Intel doesn't recommend using these sensors to report absolute temperatures.

Cool, thanks for the great links So Intel doesn't even keep their web site up to date with the latest products, as I can't find my mobile x9100 (3.06 with a 1066 bus) there at all Meanwhile, I've set TjMax to 105 until Intel will come up with some more info... I noticed that Intel rarely quotes TjMax and referrers instead to Tcase (geometric top center) or Thermal Specification, the latter I presume is Tj (junction as in mobile processors) And here I'd been thinking there was also a sensor on each core..... So I have to ask if there is some other reliable source of temp monitoring to excite the thermal throttling mechanism that occurs at TjMax??

The effect of adjusting the TjMax in RealTemp lowers or as in my case raises the current temps by the same amount.. well that is obvious I guess since the temps will always be the same relative to TjMax lol

[rge]

Cool, thanks for the great links So Intel doesn't even keep their web site up to date with the latest products, as I can't find my mobile x9100 (3.06 with a 1066 bus) there at all Meanwhile, I've set TjMax to 105 until Intel will come up with some more info... I noticed that Intel rarely quotes TjMax and referrers instead to Tcase (geometric top center) or Thermal Specification, the latter I presume is Tj (junction as in mobile processors) And here I'd been thinking there was also a sensor on each core..... So I have to ask if there is some other reliable source of temp monitoring to excite the thermal throttling mechanism that occurs at TjMax??

The effect of adjusting the TjMax in RealTemp lowers or as in my case raises the current temps by the same amount.. well that is obvious I guess since the temps will always be the same relative to TjMax lol

X9100 specs is listed in manual, page 103, tabel 20, funny the manual has it, but not spec sheet, tjmax is 105.
http://download.intel.com/design/mob...s/32012001.pdf

There are several sensors on each core in potential hot spots...DTS logic scans the sensors and "highest temp" or lowest delta to tjmax is software readable for each core via MSR. (unless using TAT which accesses via peci and takes a running 3 second average of previous temps)

[unclewebb]

I've noticed that the Intel Processor Spec Finder has the wrong CPUID listed for the newer E0 desktop processors. The E0 are CPUID 0x1067A and the older C0 are 0x10676. Intel, at times, also interchanges the words TCase and TjMax in their specs. When reading their documentation it's always a matter of doing a little bit of reading between the lines combined with some guessing and further detective work.

[psychok9]

I saw the http://processorfinder.intel.com/det...px?sSpec=SLACR site (thank you for the info), Q6600 has 71°c max thermal specification (tCase).
Is the max TJ, for Intel "specification", 76°c (71+5°c)?

Q6600 Max Tj is 100°c only for comparation with G0 brother, or we have 100% info direcly from intel?
I read it's sure only for E8xxx cpu (dualcore).

How can the mobile CPUs be more resistant than the desktop ones?

Thank you, and sorry for my eng

[ihjones]

X9100 specs is listed in manual, page 103, tabel 20, funny the manual has it, but not spec sheet, tjmax is 105.
http://download.intel.com/design/mob...s/32012001.pdf

There are several sensors on each core in potential hot spots...DTS logic scans the sensors and "highest temp" or lowest delta to tjmax is software readable for each core via MSR. (unless using TAT which accesses via peci and takes a running 3 second average of previous temps)

That manual is really interesting Thank You

[unclewebb]

psychok9: The Intel thermal specs for desktop processors are usually TCase. The Intel specs for mobile processors are usually TjMax. The two are not the same but sometimes Intel makes a mistake and uses the TCase word when they should have used the TjMax word.

Intel has not released any TjMax information for 65nm desktop processors. I think Q6600 - G0 is TjMax=100°C based on IR thermometer testing. Measured temps are the same as an E8400 during testing and Intel stated TjMax=100°C for the E8400 so Q6600 is probably the same.

[rge]

I saw the http://processorfinder.intel.com/det...px?sSpec=SLACR site (thank you for the info), Q6600 has 71°c max thermal specification (tCase).
Is the max TJ, for Intel "specification", 76°c (71+5°c)?

Q6600 Max Tj is 100°c only for comparation with G0 brother, or we have 100% info direcly from intel?
I read it's sure only for E8xxx cpu (dualcore).

How can the mobile CPUs be more resistant than the desktop ones?

Thank you, and sorry for my eng

On link below, the first 6 (45nm) intel gave us specs for both Tjmax and Tcase max, so you can get an idea of the relationship. The rest (65nm) intel lists specs for Tcase max, we have IR readings, and from that can guess at Tjmax from same 5C difference between measured IR and listed tjmax on similar but known 45nm cpus. The diff between tcase and tjmax is more like 26-30C on known 45nm. But nothing on 65nm Tjmax from intel.
http://www.xtremesystems.org/forums/...postcount=2152

Tjmax Q6600 has to be 95C or greater, and is likely 100C +/- error/variability part to part, if same 5C gradient between IR and tjmax exists, that is seen on E8400.

There are a lot of potential reasons why tjmax would be 5C higher on mobile, limited space for effective cooling on mobile, lower wattage, lower voltage so maybe can withstand more heat and still make it to 3 year warranty combined with 100C making some throttle as cooling gets clogged and annoying intel with complaints, no IHS on mobile..?5C gradient across IHS....who knows...but intrinsic durability, yeah, like you say that is not likely.

Edit: Unclewebb beat me by 4 mins, and with much less wordy response