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

[GGuyZ]

most of the time PROCHOT# (TCC activation) temperature erroneous gets referenced as tjunction! coretemp uses this TCC activation temperature as a reference point to calculate temperatures from. TCC activation temperature can't be read by software, BUT there is a bit in a register (we discussed this a few posts earlier) which was implemented by intel. this bit is valid on mobile processors and defines if TCC activation temperature is near 85c or near 100c! i use the term "near" because in reality, this temperature is calibrated on a processor basis and differs from die to die. on one processor die it might be 87c while on an other processor die it might be 83c! coretemp, as i assume all other programs using DTS, uses this bit to decide if for a certain processor 85c or 100c has to be used in the calculations. although this is valid only for mobile processors, this seems to work also for most current BUT not all desktop processors too. we have seen more and more new processors coming out where this assumption is not correct anymore. thats the situations, where coretemp reads temps about 15c off!
so, to get coretemp back to more valid readings, the only thing to to is to change the reference temp from 85c to 100c instead on relying on the obviously wrong set bit in the register. that was done in coretemp 0.95 for e4300 and may be some other processors too.

You probably did not understand me correctly. First, where did you read that Tjunction is calibrated specifically for each processor. This does make sense, and while I'm aware that PROCHOT# and THERMTRIP# temps are calibrated in the making process, I'm uncertain the same applies to Tjunction. The reason I'm uncertain of this is because the MSR bit you were talking about(bit 30 of MSR 0xEE to be exact), have held constant information(either 85C or 100C as you mentioned) in the mobile processors.

Second, this bit may be valid in many desktop processors, but it has no official reference in Intel docs! It is possible that this bit information has been left by mistake, but it has no real use. Why am I saying this? Well, because in the intel dev forum it has been made clear that trying to derive the real temp from the pseudo Tjunction Temp is wrong. What I don't understand is why the CoreTemp program developer keeps following this mistake, while he knows this unfortunate fact.

well, in fact the new 15c higher readings might be the correct readings. i have a second rig with one of this e4300. unfortunately i have not yet pushed it into throttling. this is the only way to figure out which temp is right. for now i would assume the "new" 15c higher reading is correct. i can hardly believe, that my e4300 running on air keeps temps at 55c under full load! its far more realistic, that it should read 70c instead. will verify this as soon as i get my hands on this rig ...

throttling tests cannot help much in determining whether the CoreTemp reading is closer to reality when calculating 100c or 85c. In the previous case, throttling would have started at around 85C core temp. Now it will require 100C to begin throttling, but this has no indication on real temps.

unfortunately i don't have any information on smartguardian and so cant say anything here. you are right, this does not fit very well into this picture. different polling intervals might play a role here.

cadaveca, did the cpu throttle at 85c or did the bios shutdown the system as the bios is set to 85c too? may be smartguardian did not signal throttling correctly? there is a tool called rmclock, which i use to check throttling. may be you can give it a try http://cpu.rightmark.org/products/rmclock.shtml

I'm very interested in this as well.

[joebuffalo]

I am totally onboard. Who cares what the absolute temperature is. It requires us to guess what the 'tjunction' is set at. But we do know, without any guessing, what the 'delta to tjunction' is thanks to Core Temp 0.95. So that's what we should be looking at, yeah.

Riddle me this, though. How small can we let that delta get before we are pushing it? Does anybody have any recommendations for how close is too close for a load (100% TAT load) temperature?

[fgw]

I am totally onboard. Who cares what the absolute temperature is. It requires us to guess what the 'tjunction' is set at. But we do know, without any guessing, what the 'delta to tjunction' is thanks to Core Temp 0.95. So that's what we should be looking at, yeah.

i see we are talking the same language ...

Riddle me this, though. How small can we let that delta get before we are pushing it? Does anybody have any recommendations for how close is too close for a load (100% TAT load) temperature?

well, i will handle it this way:

• i don't want my processor to start throttling -> keeping dts readings greater 0 would be enough in this case! you might add some safety margin, lets say 10c or 15c, just in case. this would be fine as long as the processor is not overclocked or better said as long as vcore is at default setting.
  
• as soon as vcore is raised, this might be a little different: some time ago, i have read a story on thg about extreme overclocking and there was this link to the physics behind electromigration. don't want to comment on the thg story as i have the feeling they did not interpret it right, and also don't want to warm up old dishes again, but to make a long story short and if i understood the readings correctly: if vcore is increased, current flow increases and thus does current density. double the currentdensity requires 20c lower temps to NOT run into increased electromigration! currently i'm running a vcore of 1.50v and coretemp 0.94 (unfortunately 0.95 fails on my system) reads 50c to 55c under full load. this would translate to a dts reading in the range of 30c to 35c and seems sufficient to me.

[joebuffalo]

Wow, you must have some great cooling going on there.

What are you using to load your system?

I am running an e4300 at 2.8-GHz with a Ninja/Yate Loon cooler and I'm getting DTS readings of about 30 when loaded with TAT.

[fgw]

Wow, you must have some great cooling going on there.

What are you using to load your system?

I am running an e4300 at 2.8-GHz with a Ninja/Yate Loon cooler and I'm getting DTS readings of about 30 when loaded with TAT.

dont forget i'm on water!

i use orthos to load my system. small ffts to produce maximum heat and blend test to verify stability. currently running some blend tests at various vcore settings.

regarding your e4300: in a second rig i have an aw9d-max/e4300 running on air with a tt big type 120vx. the e4300 definitely tends to run hotter. have not tried coretemp 095 here as this version crashes my ab9quadgt/e6400 system. seems you are using coretemp 095 successfully. can you post screenshots or just the different readings (temp, dts, tjunction, cpuid, revision, processor) of coretemp 094 and 095? in coretemp 094 temps on the e4300 are going up to 70c. this would translate to dts readings of 15c! if i'm right coretemp 95 should read about 85c in the same situation.

also what voltage do use on the e4300?

[joebuffalo]

dont forget i'm on water!

i use orthos to load my system. small ffts to produce maximum heat and blend test to verify stability. currently running some blend tests at various vcore settings.

regarding your e4300: in a second rig i have an aw9d-max/e4300 running on air with a tt big type 120vx. the e4300 definitely tends to run hotter. have not tried coretemp 095 here as this version crashes my ab9quadgt/e6400 system. seems you are using coretemp 095 successfully. can you post screenshots or just the different readings (temp, dts, tjunction, cpuid, revision, processor) of coretemp 094 and 095? in coretemp 094 temps on the e4300 are going up to 70c. this would translate to dts readings of 15c! if i'm right coretemp 95 should read about 85c in the same situation.

also what voltage do use on the e4300?

haha...your cooling is listed right there in your sig...I was just too tired to nitice I guess.

These screenshots are at idle, 60% TAT load, and 100% TAT load.

I'm at stock voltage, 1.325-V

[GGuyZ]

Obviously, all of those who know the logic behind this mechanism will use the Delta temp and not the absolute temp.

But still, getting closer to 'the truth' behind this is intriguing to say the least. I'm especially curious, why did CoreTemp v0.95 decided to change L2 rev chips Tjunction temp to 100c? Where was this logic derived from? It doesn't make much sense because as we can see it seems that L2 chips are much much hotter. If it's a newer revision, going through the same manufactoring process, why is it significally hotter? I can understand ~5c. but not 15c.

[fgw]

... your screenshots ...

thanks for your screenshots. can't use tat on my board and coretemp 0.95 doesn't work either ...
from looking at your screenshots: thats what i have expected.

Obviously, all of those who know the logic behind this mechanism will use the Delta temp and not the absolute temp.

But still, getting closer to 'the truth' behind this is intriguing to say the least. I'm especially curious, why did CoreTemp v0.95 decided to change L2 rev chips Tjunction temp to 100c? Where was this logic derived from? It doesn't make much sense because as we can see it seems that L2 chips are much much hotter. If it's a newer revision, going through the same manufactoring process, why is it significally hotter? I can understand ~5c. but not 15c.

guess the assumption here is, tjunction could be either 85c or 100c and as 85c was obviously to low it must have been 100c then!
btw. from reading your post at the coretemp forum i have seen you also followed the thread at the intel software network forum...

To match the results displayed via TAT. Obviously TAT is assuming a Tjunction_max of 100.

the point, as we all know in the meantime, is tat was designed for mobile processors only! the use on desktop processors might produce results that look right but nobody knows if they are. i really doubt it.
from all what i have read on this so far:

• the decision if 85c or 100c are to be used, based on a single bit in a msr, is not valid for desktop processors and thus wrong.
• even on desktop processors the temp is not exactly 85c or 100c. its just some where near this point.
• this temp is calibrated on a per chip basis and thus different from die to die.
• there is no way to read this temperature by software
• the only reliable way is to use the dts value directly

although not that precise you can use a monitoring program that's reading the thermal diode. comparing this readings to coretemp readings while the processor is forced into throttling. this could give an idea about the value of tjunction. in fact its not tjunction, its the temp where thermal control circuit gets activated, but thats a different story. at least you can figure out if its near 85c or near 100c ...
i tried this on my e6400. throttling kicked in at coretemp readings near 85c. don't remember the correct value. might have been 86c or 87c, but it definitely was not 100c!