Intel TAT / CoreTemp / IDCC all different temperatures....

[joebuffalo]

imho, that's a HUGE DTS.

A very wise (and conservative) man wrote this Core 2 Temperature Guide:

http://forumz.tomshardware.com/hardw...ict221745.html

and he seems to think that 65-C is AOK for the Tj_max = 85 systems. That's a DTS of 20. Right now my DTS is ~30 with my e4300 at 2.7-Ghz (stock 1.325-V) topped with a Scythe Ninja / 1300 RPM Yate Loon fan and AS5.

So I'd say a DTS of 30 leaves a LOT of headroom. A DTS of ~20 seems reasonable based on the opinions in the thread above.

I'm still just not sure that we should be running L2's at the same DTS as the older Core 2 CPUs. Should we run the L2's x-degrees hotter just because Tj_max went up x-degrees. (That's why I am keeping my DTS at ~30 for now)

[TheMafioso]

BUMP
r we reachin on any conclusion ???

[fgw]

A very wise (and conservative) man wrote this Core 2 Temperature Guide:

http://forumz.tomshardware.com/hardw...ict221745.html

and he seems to think that 65-C is AOK for the Tj_max = 85 systems. That's a DTS of 20. Right now my DTS is ~30 with my e4300 at 2.7-Ghz (stock 1.325-V) topped with a Scythe Ninja / 1300 RPM Yate Loon fan and AS5.

i would not rely too much on this post over at the thg forums. there are a lot of misinterpretations in it!
lets start with the things you can confirm by looking at intel datasheets:

Datasheet
http://www.intel.com/design/processo...hts/313278.htm

Thermal and Mechanical Design Guidelines
http://www.intel.com/design/processo...nex/313685.htm

yes, there are more than one sensors measuring temperatures

• one thermal diode located on the processor die
• two digital temperature sensors within each core on c2d
• four digital temperature sensors within each core on quad

intel defines tcase at the temperature measured on top of the ihs and at the center of the ihs.

afik, bios or software is not displaying tcase! the temperature displayed there is simply the processor temperature measured via the thermal diode. the thermal diode as implemented by intel is delivering a voltage proportional to processor temperature. the software using this signal has to calibrate this readings and convert it to a temperature. as different software, or bios, is calibrating this readings differently, you get different temperatures but by no way tcase as described above.

the digital temperature sensors on the other hand, are located directly within the cores near the hottest places. this sensor has a smaller footprint and thus can be located easier within the core. the reading, as opposed to the thermal diode, is not a voltage! digital temperature sensors convert the temperature directly into a binary value (dts reading) and store it in a register within the processor thus eliminating the need of calibrating this sensor readings.

the problem here is, this dts reading is not absolute. its a relative temperature to the thermal control circuit activation temperature (prochot#). as this thermal control circuit activation temperature is calibrated during manufacturing on a per part basis and is not read by software it is impossible to convert dts readings into an absolute temperature.

on mobile processors tjunction as a reference temperature instead of thermal control circuit activation temperature. also, there were just two possible tjunction values used on mobile processors in the past (don't know intel plans on future mobile processors). on this mobile processors, there is a bit in a register which can be read to check if tjunction is at 85c or if it is at 100c! absolute temperature can now simply calculated by substracting dts reading from tjunction. thats how coretemp, and i assume all other programs using digital temperature sensors readings, calculate absolute temperatures.

using this method on desktop or server processors might result in a temperature reading that look right, but this does not mean it is correct.
there is only one correct temperature reading on intel core technology processors: relative temperature (dts reading) as introduced incoretemp 0.95

everything else, besides mobile processors as discussed above, is speculation!

and so is most of the following:

• Tcase + 15c = Tjunction its more like: dts reading + 15c = thermal control cicuit activation temperature
• Thermal Case Temperatures of 60c is hot, 55c is warm, and 50c is safe. well, this might be true, but there is no way to measure tcase besides a temp probe on the ihs! do you use one? i don't.
• The difference between Tcase (BIOS, Motherboard Utilities and SpeedFan: CPU or Temp X) and Tjunction (TAT and SpeedFan: Core 0 / Core 1) is Tcase + 15c = Tjunction
• Vcore should not exceed ~ 1.5v.
• Tjunction is always ~ 15c higher than Tcase.
• Tcase is always higher than Ambient. as any temp as long as you are on air or water ...
• Tcase Idle should be ~ 1 to 15c higher than Ambient.
• Tjunction Idle should be ~ 15 to 30c higher than Ambient.
• Tcase Load should not exceed ~ 55c with TAT @ 100% Load. again, how is it measured?
• Tjunction Load should not exceed ~ 70c with TAT @ 100% Load.
• Idle to Load Delta should not exceed ~ 25c.
• Tjunction Results are Hottest Core Idle and Load.
• Vcore will typically sag ~ .025 volts under Load. depends on motherboard
• Any hardware and / or software may misreport Tcase and / or Tjunction temps.
• 965 chipsets may misreport Tcase and Tjunction temperatures with +/-15c offsets. the chipset is not involved at all - all sensors are within the cpu. well, calibration of thermal diode is done in bios
• If TAT will not run, then Orthos Priority 9 Small FFT’s simulates 88% of TAT ~ 5c lower.
• C2D’s manufactured with concave Integrated Heat Spreaders may report high Deltas and temps. true
• An improperly seated CPU cooler is the leading cause of abnormally high temperatures. also true, but again - no surprise here

added comments to some of them although almost all are either questionable or obvious ...

[joebuffalo]

I agree with a lot of what you said there.

Although I think the mobo does come into play when talking about Tcase (keep in mind he is not defining Tcase the same way Intel is...is is talking about the thermal diode output) because the mobo is where the thermal diode reading is converted into a temperature reading.

I didn't post his link because I think that thread is correct. In fact I disagree with a lot of it...that's how I wound up here on the DTS bandwagon. My point was just that he recommends a DTS of 20 (doesn't actually SAY that, but you and I know that is the jist of it) and I think he is pretty conservative (maybe it's because I've read his opinions on many overclocks) when it comes to temperatures.

I think your 30-degree DTS is VERY safe. Too safe for a bunch of overclockers...imho.

[fgw]

I agree with a lot of what you said there.

Although I think the mobo does come into play when talking about Tcase (keep in mind he is not defining Tcase the same way Intel is...is is talking about the thermal diode output) because the mobo is where the thermal diode reading is converted into a temperature reading.

I didn't post his link because I think that thread is correct. In fact I disagree with a lot of it...that's how I wound up here on the DTS bandwagon. My point was just that he recommends a DTS of 20 (doesn't actually SAY that, but you and I know that is the jist of it) and I think he is pretty conservative (maybe it's because I've read his opinions on many overclocks) when it comes to temperatures.

I think your 30-degree DTS is VERY safe. Too safe for a bunch of overclockers...imho.

you are right! but i mentioned this 30c by two reasons:

• electromigration: to maintain the same reliability, temperature has to be decreased by 20c when current density gets doubled. i know, increasing vcore from 1.28v up to 1.50v does not result in doubling the current density, but the scale is logarithmic, so in my case temps might be decreased by 17c or may be 15c!
• simply the fact, that my watercooling system is able to maintain this sort of safety margin ...

[unclewebb]

there is only one correct temperature reading on intel core technology processors: relative temperature (dts reading) as introduced incoretemp 0.95 everything else, besides mobile processors as discussed above, is speculation!

I agree with the above statement 100% but.......

I still think that CoreTemp 0.95 which uses either a fixed 85C or 100C Tjunction value in its calculation results in a pretty accurate absolute temperature for the desktop processors.

I decided to come up with a simple test so users can determine if the absolute temperatures reported by CoreTemp are valid or not.

On my E6400, revision B2, CoreTemp 0.94, CoreTemp 0.95 and SpeedFan 4.32 all report the same temps. For this test I used SpeedFan for its graphing capabilities.

The goal is to run your processor as cool as possible. Start by opening your case. Next lower the FSB to 200 MHz, lock the cpu multiplier in the bios at 6 and reduce the core voltage. Setting core voltage to 1.200 volts or less is adequate. With my P5B Deluxe I was able to get the core down to 1.104 volts as reported by CPUz.

You can do this with any Core2Duo so this test is consistent no matter what processor you have.

I believe that when you run an air cooled processor at reduced speed and voltage in an open case, the core temperature will start to approach the ambient temperature. I used SpeedFan to keep the cpu fan at its maximum.

With a room temperature of 18C as measured with a consumer grade digital house thermometer, I ended up with a reported core temperature of 17C during this test.



In theory with air cooling you shouldn't be able to get lower than ambient temperature but if you keep in mind that both values are integer values and are at best only accurate to +/- 0.5 degrees and that my $20 Radio Shack thermometer is far from perfect then I think my results confirm that:

An air cooled C2D at 1200 MHz and 1.1 volts will put out so little heat at idle that the core temperature and the ambient temperature will be very close to equal.

For my E6400, the CoreTemp / SpeedFan assumed Tjunction value of 85C, seems to produce a very accurate absolute temperature value.

This test can be used by anyone with any Core2Duo processor to at least give them an idea if their temperature monitoring software is in the ball park or if it's totally out to lunch.

[joebuffalo]

Fair enough. But riddle me this, enclewebb:

What good is the absolute temperature? How do you know what is a safe absolute temperature? That's the beauty of the DTS, it tells you the 'Degrees To oh Shoot'

[biatch0]

Hey, just want to get some input from you guys about my CPU temps. I was previously using the stock Intel cooler that came with my QX6700... but after a bunch of temp readings that were confusing, I took the safe route and assumed the higher temp readings were the accurate ones. Below are screenshots from TAT/CoreTemp/Everest from the time I was using the stock HSF:




And here's a screenshot of those after I installed the new ASUS Silent Square Pro:



The temps I'm getting are still a little on the high side which worries me... anyone with an aircooled QX6600/6700 with temp screenshots?

[unclewebb]

What good is the absolute temperature? How do you know what is a safe absolute temperature? That's the beauty of the DTS, it tells you the 'Degrees To oh Shoot'

Users seem to like knowing the absolute temperature of their cpu, even if that number isn't 100% accurate.

Using my testing method you can at least determine if your Tjunction is about 85C or about 100C and which version of CoreTemp is reporting your absolute temperature closest to reality.

Going by DTS is great in theory but already there are discussions of what is a good DTS number to shoot for and a 101 qualifiers like whether you are overclocked or over volted and by how much, etc., etc. I've already discovered that if you are at default MHz and voltage that you can run reliably at a higher core temperature than if you're fully overclocked which makes sense.

People have never been able to agree on a safe absolute core temperature any more than they will ever be able to agree on a safe DTS value.

My E6400 with an approximate Tjunction of 85C showed Orthos stability started to decline when DTS went below 15C and declined further when it got to 10C. I have absolutely no idea whether those results will apply to an E4300 which has a Tjunction of 100C. It's possible that a well overclocked C2D looses stability when its absolute core temperature exceeds 70C to 75C regardless of what Tjunction Intel has rated it at.

Hopefully some E4300 owners can shed some more light on this topic by doing their own testing.

[unclewebb]

biatch0: You didn't mention whether those are idle temps or if you are running something like Orthos at full load.

If those are idle temps then I think you should look at re-mounting your cooler.

If those temps are while running Orthos then temps in the low 60C range with all 4 cores loaded at 3000 MHz looks pretty good to me. The Quads put out more heat and are going to run hotter at the same load and same frequency compared to a Core2Duo.

[fgw]

In theory with air cooling you shouldn't be able to get lower than ambient temperature.

thats true! its also true for watercooled systems, except its not ambient temp but water temp that can't be beaten!

would not mind if temps are off by a few degrees, but if the processor runs at stock, or even below that and you get core temps 10c or more below ambient than tjunction in coretemp is wrong and thus your absolute core temps!

yes, your simple test can quickly verify if coretemp or whatever software used, gets the reference temp (tjunction in coretemp) correctly!

[Fr3ak]

I somehow have the feeling that CoreTemp 0.95 does not work "correct" for all E4300. I built 2 E4300 rigs last week using the Infinity and the idle temp at stock volts/stock clock was 42-46C on both rigs. So I am having the feeling thats not the real temp for one of the best air coolers and a C2D at stock. With 1.34V real I had 80C according to coretemp at 3150Mhz after 19hours of Orthos. Those temps dont seem to be right to me. 15C less than that sounds more like it. Unfortunately I have only remote access to those rigs, so I cant test much from here :/

[joebuffalo]

I somehow have the feeling that CoreTemp 0.95 does not work "correct" for all E4300. I built 2 E4300 rigs last week using the Infinity and the idle temp at stock volts/stock clock was 42-46C on both rigs. So I am having the feeling thats not the real temp for one of the best air coolers and a C2D at stock. With 1.34V real I had 80C according to coretemp at 3150Mhz after 19hours of Orthos. Those temps dont seem to be right to me. 15C less than that sounds more like it. Unfortunately I have only remote access to those rigs, so I cant test much from here :/

Have you read this thread?

[Fr3ak]

Most of it.
From what I understood, 0.94 reads 15C to low for the E4300, so it was changed in 0.95 which reads a 15C higher temp.

Now what if the 15C issue is not working for all E4300 batches?

[unclewebb]

I somehow have the feeling that CoreTemp 0.95 does not work "correct" for all E4300.

That's what I've found as well. They seem to come with either a TjMax of 85C or 100C and no software so far can tell the difference.

If it guesses wrong your reported temps will be wrong by 15C. Lucky a smart guy figured out a possible way to determine which one is right.

The only thing I've found is it might be related to the manufacturing date code. I found an early E4300 with date code L640 which seemed to have a 100C TjMax and one with a date code of L644 which had a TjMax of 85C. Between week 40 and week 44 of 2006 is anyone's guess.

[joebuffalo]

Nobody knows what the Tjmax for Core2 Duo desktop processors is. As such, it is impossible to determine the absolute core temperature of the processor.

I think your method, unclewebb, although logical, is flawed. You are still *assuming* many things to make you calculation.

You are also assuming that there is no thermal gradient within the processor. You assume that Tjunction and Tcase are equal.

You assume that the air inside your case is the same temperature as the air outside the case.

You are assuming that Tjmax is either 85 or 100.

I see what you are shooting for. But you method is based on a LOT of assumptions that I don't see the point. Why do you need to know the absolute temp, based on assumptions, when you already know the relative temp with certainty?

CoreTemp 0.95 works PERFECTLY with any Core2 processor because it measures, and can be configured to report, the DTS...which is all you need be concerned with.

[fgw]

joebuffalo,

second that!

[SLi_dog]

CoreTemp 0.95 works PERFECTLY with any Core2 processor because it measures, and can be configured to report, the DTS...which is all you need be concerned with.

You mean like this?




It would be seriously great if it is.

It just seems like it's based on the same Tjunction value set by the software, which is the cause of this argument, but it just reads in reverse

[fgw]

You mean like this?




It would be seriously great if it is.

It just seems like it's based on the same Tjunction value set by the software, which is the cause of this argument, but it just reads in reverse

jumping in here as i requested this feature (show delta to tjunction temp) over at the coretemp development forum.

anyway, YES this is the feature!

all temp monitoring software which is using dts (digital thermal sensor) is reading the temperature from a register within the processor. this register holds the plain dts value showing the temperature offset to

• tjunction temperature for mobile processors
• thermal control circuit (tcc) activation temperature for desktop and server processors

absolute core temperatures are simply calculated by the following formula:
core temperature = tjunction - dts or core temperature = tcc - dts

it doesn't really matter if you call it tjunction or tcc as in both cases this temperature is just an assumption made by the writer of the monitoring software. the value used in the formula counts, not the name ...

accuracy depends on the correct chosen value for tjunction or tcc which is obvious by looking at the above formula. as intel does NOT make this values available to public nor is this value readable by software, software developers have to guess what value should be used for tjunction.
thats the difference in coretemp 94 and coretemp95! both versions of coretemp simply use different values for tjunction, either 85c or 100c. thats why readings in coretemp might be off by 15c on certain processors.

using the delta to tjunction temp instead, displays plain dts readings without conversion to an absolute temperature value thus being completely independent from probably wrong selection of tjunction value in monitoring software. using this feature in coretemp makes you completely nonindependent from tjunction and gives correct values on ALL processors supporting dts.

this delta to tjunction temp just says how many degrees is the processor away from tcc activation (throttling) and thus looks reverse to regular coretemp readings: large dts means cool processor, small dts means hot processor!

the last question here: what is a save dts value?
well as with regular coretemp readings this depends on the environment your processor is running in. my very personal rule is as follows:

• running on high vcore, which to me is everything above 1.5v, i would try to keep dts above 30c. (this translates to about 55c in coretemp94)
• on lower vcore, a dts value above 15c should be fine and easy achievable. (70c in coretemp94)

this assumptions take into account that higher vcore results in higher current and thus in higher current density which besides temperature is the most responsible parameter for reliability or failures due to electromigration. reducing operating temperatures maintains reliability even if current density (vcore) is raised.
the general rule is roughly: decrease temps by at least 20c for every current density increase of 100%! yes, there is some safety margin in my personal rules.

again, this is my very personal opinion gathered from reading through a lot of documents on this issues ...

[unclewebb]

joebuffalo: You've brought up some valid points. I'm going to do some more testing and hopefully in a day or two I might have some more information for you.

I was just testing RightMark CPU Clock Utility. Version 2.25 is now capable of monitoring core temperatures of the Core2Duo.

http://www.softpedia.com/progDownloa...oad-18249.html

This utility seems to track very closely the absolute temperatures reported by Tat and CoreTemp from ambient to over 80C. It also shows you when TM1 and TM2 throttling become activated and by what percent.

The interesting thing is that it is the first utility I've seen for the C2D that displays core temperatures with a resolution of 0.2 degrees.
ie. 40.0, 40.2, 40.4, 40.6, 40.8, 41.0

Edit: Upon further testing I've discovered that this program averages the previous 5 temperature values which gives the look of better temperature resolution but it is still using the same Intel documented DTS integer value and guessing at a Tjunction value like every other program.

[SLi_dog]

this delta to tjunction temp just says how many degrees is the processor away from tcc activation (throttling)

I guess that's my question though, is it reporting the actual DTS sensor and its approach to throttling or is it just calculating the readings in an alternate method based on the application determined Tjunction?

I really have a limited idea of how these sensors work and/or I may not be configuring it right however I'm wondering if it would simply read a different "delta to Tjunction" value if I changed to a CPU that CoreTemp believes has an 85C Tjunction value?

Cheers

[joebuffalo]

DTS is read *directly* from a CPU register. No calculations are involved. No assumptions are made. It is the ONLY temperature reading that is reported directly from the CPU.

EVERY software tool that says it can report an absolute core temperature reads the DTS (they all read the exact same DTS value), assumes a tjmax, and calculates an absolute temperature.

[SLi_dog]

DTS is read *directly* from a CPU register. No calculations are involved. No assumptions are made. It is the ONLY temperature reading that is reported directly from the CPU.

So.......if it was somehow possible to change the CoreTemp Tjunction value from 100C to 85C, the Delta to Tjunction wouldn't end up being 15C lower as it's purely reporting the DTS value?

If not, how is it different to CoreTemp reporting based on an assumed Tjunction value?


Just trying to understand this, thanks for the patience

[fgw]

So.......if it was somehow possible to change the CoreTemp Tjunction value from 100C to 85C, the Delta to Tjunction wouldn't end up being 15C lower as it's purely reporting the DTS value?

correct!

If not, how is it different to CoreTemp reporting based on an assumed Tjunction value?

the point here is, you simply don't know if the assumed tjunction value which is used in temperature calculation is correct or not.

e.g. if you have a dts (Delta to Tjunction) of 30c you get a core temperature displayed in coretemp of either 55c or 70c depending on tjunction value assumed by coretemp for your processor. is it now 55c or 70c your cores are running at?
on the other hand, if you are looking at plain dts (Delta to Tjunction) you can see you are still 30c away from throttling and this holds true regardless what tjunction value was selected by coretemp as no calculations and assumptions are involved.

btw. this is the same for all other programs using dts to read core temperatures. afik, coretemp is the only program giving you the ability to display dts (Delta to Tjunction) directly and thus is the preferred choice!

[SLi_dog]

correct!


the point here is, you simply don't know if the assumed tjunction value which is used in temperature calculation is correct or not.

e.g. if you have a dts (Delta to Tjunction) of 30c you get a core temperature displayed in coretemp of either 55c or 70c depending on tjunction value assumed by coretemp for your processor. is it now 55c or 70c your cores are running at?
on the other hand, if you are looking at plain dts (Delta to Tjunction) you can see you are still 30c away from throttling and this holds true regardless what tjunction value was selected by coretemp as no calculations and assumptions are involved.

btw. this is the same for all other programs using dts to read core temperatures. afik, coretemp is the only program giving you the ability to display dts (Delta to Tjunction) directly and thus is the preferred choice!

But don't you find it strange that the "Delta to Tjunction" throttle point with the quad core co-incides exactly with the software written Tjunction value of 100C?

I guess I'm still not convinced but I'll bow to your knowledge guys. Thanks for the help and patience