Unclewebb, i also have a E8400 with a very inaccurate temp reading. What can i do to get a custom profile from you? Hope u're willing to help me. Thanks in advance.
Printable View
Unclewebb, i also have a E8400 with a very inaccurate temp reading. What can i do to get a custom profile from you? Hope u're willing to help me. Thanks in advance.
The part i don't understand... is this suppose to be a bug in the line of the E8x00 series or they just changed the way it work and we need to wait for some fix on temps program's ?
Are we suppose to RMA our CPU or just live with it ?
As long as the DTS sensor is doing what it's supposed to do (trip at the TjMan), you can't RMA it. It's something you will probably just have to live with.
OK, am fine with that part, now the question is, did they design it like that ?
Sound like cheap work from intel for a 300$ part
We should design a program to read dts output and categorize it showing delta to Thermal junction max. Since the gradient is relatively small, it would be sensible to show a graph or quantitative progress bar that served as an overclocking thermometer.
None of the programs are working with ease, effectively, for the basic overclocker out there that is curious about his/her temperatures. Kind of like a thermostat in a vehicle.... (with all pertinent technical data in numbers displayed to the side as in the realtemp 2.2 or coretemp) you could see a meter like this:
-Temperature-
|||||||||low|||||||||||||med||||||||:oscar:||||||HIGH||||:explode2:
E8400:
--3.0g-----3.6g----4.0g----4.2g----4.3g---4.5g---4.6g--
speed: 4310 mhz
vcore: 1.42 volts
tj max: 94.5C
Core 0: 77C
Core 1: 78C
:ROTF:
I don't mean to be picky, but that's not what a thermostat does in a car :)Quote:
Originally Posted by jaredpace
i think u need 2 calibrate it on ++. i got e4300 and thats wat i use. i should do a occt run since i got almost the same setup as u. except im on 8x375 stock volts and my abmient temp is about 25cQuote:
Originally Posted by Unrealer
On a Q6600 B3, around how much is the estimated TjMax ?
So, 49º @ full?Quote:
Originally Posted by pitadotcom
http://i38.photobucket.com/albums/e1.../Dibujo-47.jpg
what does a thermostat in a car say?Quote:
Originally Posted by Wompa164
Its says OPEN or CLOSE. :rofl:Quote:
Originally Posted by jaredpace
hahah i dont get it?Quote:
Originally Posted by jas420221
I think you're confusing the thermostat w/ the coolant temperature gauge.Quote:
Originally Posted by jaredpace
The thermostat opens or closes depending on the temperature of the coolant.
The coolant temperature gauge will rise and fall w/ the temperature of the coolant.
Well said...He is confusing a thermostat and a thermometer.Quote:
Originally Posted by jason4207
One measures temperatures (thermometer). The other opens or closes/turns on or off at a pre determined temperature.
is it normal when 1 core 31°c and other 20-21°c in IDLE? in LOAD they're pretty same.
Unfortunately, it's very normal for many E8400 CPUs and very rare if you have an older 65nm Core CPU.Quote:
Originally Posted by TEDY
Unrealer: I had a look at your results and decided to give you some numbers to compare to.
I presently have my E2160 still in and I ran it at 1.424 volts (load) just like yours. I tried to run it at 3150 MHz but it wasn't 100% stable here which I guess is why it ended up in the Pentium bin. I dropped down to 3000 MHz and kept the voltage the same to maintain stability for testing.
With similar room temps of 18C to 19C I recorded a maximum temp of 54C. If you go back a couple of pages in this thread you will see that the RealTemp reported temps are almost identical to what the IR thermometer reports for this processor.
I was using a Tuniq Tower with the fan at about 1750 rpm in a closed case. Considering all the variables, your RealTemp readings for your E4300 look very reasonable and believable to me.
http://img402.imageshack.us/img402/9955/occtidlejb0.png
http://img258.imageshack.us/img258/2504/occtendpm2.png
At 50C to 54C, the OCCT reported temp which comes from SpeedFan was identical to what RealTemp was reporting. The idle calibration feature has no effect at this temp but added 3C to the idle temps. The second screen shot was at the exact moment that OCCT went to the Idle phase. RealTemp shows that the temp has just dropped while OCCT / SpeedFan is still reporting 53C which is what RealTemp was reporting a second before.
OCCT at 3150 MHz also peaked at 54C but it usually crashed within 5 minutes. A small change in MHz makes virtually no difference in temps as long as the core voltage is kept constant. I hope this testing helps you decide whether CoreTemp or RealTemp is more believable for your E4300.
Edit: Just a quick pic to show that OCCT / SpeedFan = RealTemp at full load since both programs use the same TjMax for the E2160.
http://img247.imageshack.us/img247/9...ealtempis5.png
IDLE temp:
E4300 @ 3.2Ghz (2.13 @ speedstep) 1.44v idle
Ambient temp: 17º
Big Typhoon @ 1000rpms
TAT: 37
SPEEDFAN (default): 21º
CoreTemp: 36º
RealTemp: 27º
http://i38.photobucket.com/albums/e1.../Dibujo-49.png
okay yes i meant thermometer. we need a cpu thermometer based on delta to tjmax.
Unclewebb:
I have the new C2Q 9300 which is listed as a M1 stepping, do you happin to Know what the Tj Max
core temp says that my idle is around 52 ( open air with TR Ultima 90)
my MB says 24
I have Vista and couldn't get real temp to work so i cant rebort its readings
Thanks
Right click on RealTemp and run it as Administrator and it should work fine in Vista.Quote:
Originally Posted by dart22
Unrealer: So what's your conclusion? Do you trust any of these programs to accurately report your core temperatures from idle to full load? If you are still unsure then I guess someday I'll have to buy an E4300 to solve this mystery once and for all. With the Idle Calibration at (++) and your cpu at low MHz and low volts, what does RealTemp report after is has been at idle for a minute or two compared to your room temp?
Your 27C idle temp looks pretty similar to my 26C idle temp above using the same core voltage, more or less.
Thanks this is what I wanted confirming, I am not too fussy if my idle temps are a few degrees out as long as under load temps are correct. All the documents people are posting seem to be saying a thermometer is required and they all have long technical explanations.Quote:
Originally Posted by unclewebb
Nice work!
Finally a program that reads the E8000s. :clap:
unclewebb, would be possible to make it minimize to TRAY?
Regards.
E4300 @ 1.8Ghz (1.2 @ speedstep) 1.23v idle/1.22 fullQuote:
Originally Posted by unclewebb
Ambient temp: 17.5º
Big Typhoon @ 1800rpms
IDLE
SPEEDFAN (default): 17º
CoreTemp: 32º
RealTemp: 24º
http://i38.photobucket.com/albums/e1...um2/idle-1.jpg
FULL
SPEEDFAN (default): 28º
CoreTemp: 42º
RealTemp: 33º
http://i38.photobucket.com/albums/e1...um2/full-1.jpg
Im really unsure :SQuote:
Originally Posted by unclewebb
Unclewebb:
Thanks i managed to make it work on vista, real temp is telling me that my Q9300 Tj max is 95 i still think its its not right because my cooler does not seem that hot to touch niether does the base of the heat sink!
Real Temp is giving me 46 idle and 68 load!
i used an IR meter and i could not get any temp over 33 Idle near the cpu!
Gigabyte's Easy Tune is giving me 28 idle 52 Load!
Also I just want to mention that the Q9300 has a different stepping (M1) than the Q9450 and the E8xxx (C1)
So whats your take on that ?
Determining TjMax is a guessing game as Intel does not document this value. The Q9300 is new so there isn't much user data to go by either. Have you tried running your Q at 6x333 with as little core voltage as possible to see what temp your cpu idles at?Quote:
Originally Posted by dart22
If you are convinced that TjMax=85C then you can still use RealTemp. Just go into the RealTemp.ini file and set this:
TjMax=-2
This reduces the TjMax that RealTemp uses by two notches of 5 degrees each so it will drop TjMax from 95C to 85C but if you're wrong then you will screw up your load temps.
The E2160 - M0 I tested was TjMax=85C so it is possible that the M1 Quad is also TjMax=85C. The Intel docs show that the CPUID is 0x10676 for the E8x00 processors and 0x10677 for the C1 and M1 45nm Desktop Quads so that's why I've assumed that TjMax=95C for all of them. When you run RealTemp and click the mouse in the TjMax box it will read and display the CPUID of your processor directly from the CPU.
Do you have any more info like your room temp and cooling being used, case open / closed, fan speeds, etc.?
I know that large after market air coolers don't get very warm to the touch with the E8x00 series even when the cores are fairly hot. If you want to do us all a favor then pull the heatsink off of your nice new Quad and point your IR thermometer at it. :D
I also found that my E8400 needs an Idle Calibration setting of (--) for the RealTemp reported temps to equal the IR reported temps near idle.
For Core 2/Core Duo/Core Solo (desktop/server) based systems, I've found out the TjMax officially and how to know it. I can't get Penryn info on it though, S.FAE says I have to sign NDA for it and uncle being a very scrupulous professor won't leak such details to me even if the sky fell down. :(
unclewebb: Have to thank you immensely man, I'm being honest, I've become frustrated reading the same things problematic->requested->questioned->regurgitated every time I login, and yet you're still there helping, repeating it away with good explanations. :clap:
I understand that not everyone has the time to read 500+ posts so I'm hanging in here with the tech support. I've been waiting and waiting this week to finally get a chance to install my Q6600. I should have time this weekend.
I wish I could provide an intel doc on it. Thanks for clearing up what you've done and giving the option to bump the tjmax. I will play with it on a later date because right now my cpu temps don't bother me they're nothing like my B3 Qx6700 chip I had at 3.6Ghz.Quote:
Originally Posted by unclewebb
Sorry if I missed it, but I can't make this program work on my Vista 64-bit PC. Each time I try to run it, it just "Driver not load". I read first post and it points me to nowhere. It just says about about IA32, but it's not even in my System folder.
Edit: Oops, I just read page 21 and found the answer I was looking for...thanks!
Quote:
Originally Posted by unclewebb
My Quad 6600 shows a 5 degrees difference (RealTemp shows 5 degrees cooler than OCCT's built-in), is that accurate?
I'll sit tight then since I ditched the e4300 and got myself a q6600 in the meantime (so far it looks like i was lucky too, it needs sod all volts to clock)Quote:
Originally Posted by unclewebb
Tjmax is shown as 95C on realtemp whereas it's at 100C everywhere else so i'm puzzled as to which to take seriously, even though 5C aint exactly the end of the world, but i'm pretty sure there's more into it.
Testing at low volts and clocks isn't as explanatory as it would be on a dual or single core, since 4 cores on the same chip are expected to be warmer than two or one obviously.
Have to thank you again unclewebb for keeping this updated and all and giving feedback when it's needed, top job :up:
I can confirm this does work on vista x64 (with SP1, not sure if that will make a diff.) Using e6850. Thanks, great program
I don't understand that statement unless your on chip DTS sensors are damaged or not working correctly. If TjMax=105C then CoreTemp should be reading this processor correctly and if TjMax=95C then RealTemp should be reading your temps correctly. If TjMax is some other value like 100C then it is easy enough to change TjMax in the RealTemp.ini file. There's no need to fear the temp these processors are running at. As long as your heatsink hasn't fallen off they will never get that hot, even when overclocked. If they do get too hot, they'll simply shut down long before you'll damage one.Quote:
Originally Posted by mcoffey
I don't know. Is OCCT accurate? Thanks for your info concerning getting RealTemp to work on your computer. I just updated the first post so new users will have less trouble getting this to work in Vista.Quote:
jcniest5: My Quad 6600 shows a 5 degrees difference (RealTemp shows 5 degrees cooler than OCCT's built-in), is that accurate?
unclewebb nice work man.
any chance of adding a thermometer based on delta to tjmax? :)
I understand that problem very well. ;)Quote:
Originally Posted by mcoffey
This may sound crazy coming from a guy that wrote a program that tries to determine absolute core temperatures but really, absolute temperatures are not that important. As long as your computer is stable at full Prime load, or whatever maximum load you typically run, then you'll be fine. Just make sure that you've left yourself some headroom which RealTemp reports as Distance to TjMax and your core voltage is within the Intel spec and your processor should live a long, healthy life.
Quote:
Originally Posted by unclewebb
welll unclewebb take some time for your self ! dont forget the fun of it all ;)
every thing else can wait really ! no one gonna die hahahahahaha
your a piller to your forum :up:
I've been following this thread since it began and appreciate all the work that's gone into this development.
There have been some great ideas posted here, and I just wanted to support one idea that I feel got passed over rather quickly.
Jaredpace suggested representing the temperature in the form of a coloured gauge rather than as an absolute temperature.
As I think most of us would agree, the actual core temperature is not that important in itself. System stability is our primary concern, and as long as the resulting temps under load are sufficiently below tjmax, then that's really all that's important.
It seems clear that Intel are reluctant to release the tjmax values for their processors, and whilst it's possible to perform many measurements and come up with proposals for the tjmax values, we can never be certain that these proposals are correct, especially considering that:
- the accuracy to which the sensors have been calibrated is not known
- it's not possible to position external sensors to measure temps at the same point
With the 'gauge' idea, it would only be necessary to agree on safe margins from tjmax (for colour-coding the gauge), which should be easier than trying to determine the tjmax values for each processor. Any such program could then report the distance from tjmax as an absolute value for information purposes.
I realise this would require a change in most peoples mindset, but I feel this would be a positive change.
Please understand that I'm not trying to detract from all the hard work and effort that's gone into this program so far. I know that other issues are also being addressed by this program. I'm personally familiar with the sticking sensors at low temps as my E8400 has 1 core permanently reporting 42C at stock (idle and load). I simply wanted to add my support to this 'gauge' idea as I believe it's the right direction for temperature-monitoring applications.
no, actual core temp is NOT interesting - and actually it only leads to more confusion. I could live with only knowing about Delta to TJ...and actually i think a graphical bar would be very nice showing the remainder to TJ, from green to red or something.
Great work "unclewebb" :up:
Here is my E8400 (from the first "batch" to Norway).
It clearly divert from your E8400 I think.
One sensor (Core1) obviosly is stuck. Core0 shows too low temp assuming a TJMax=95 (more realistic if offset +10=105).
This is with watercooling in a 19c room.
woot others agree a meter would be nice! it makes sense since distance to tjmax is where ALL programs share identical data...
if at around 40C rt/ 50C ct its in the green and at 80C rt/ 90C ct its up into the red, then accuracy of either program is not important as long as the user knows how much headroom he/she has remaining.
10C Until meltdown. a big colored bar will just make it easier for all of us to read. Similar to the ease of ordering a meal by saying a number.
edit: if i knew anything about coding software for computers, i would attempt it. :)
A colored graph based on Distance to TjMax is a good idea in theory but even here there's a problem. The core voltage you are running and how much you are overclocking directly effects the stability of your computer at a given temperature.
My E6400 with a TjMax=85C is a good example. It is Prime/Orthos stable at:
3200 MHz - DTS = 5
3400 MHz - DTS = 15
3600 MHz - DTS = 25
The amount of head room a person has to leave to maintain stability varies with how hard they are pushing their processor. For my E6400, even when it is overclocked by 50%, it remains Orthos stable right up to the thermal throttling point just before TjMax.
http://img135.imageshack.us/img135/1...kinghotgq3.png
Orthos and CoreTemp misreport MHz so ignore them. TAT shows the thermal throttle was Active (aka. rev limiter :D ) yet Orthos continued to run without any problems and CPU-Z correctly reports that the CPU is still running at full speed. You need higher temps than this before the thermal throttle will drop the multi down to 6.0 and real throttling begins.
When running this same CPU at 3600 MHz, I can't get anywhere near this temperature while running Orthos without randomly rebooting or losing stability.
A colored bar graph is not going to solve anything. How can I tell a user in the above example that is running their E6400 at 3200MHz and their DTS=25 that they are near the "red zone". That will only scare a person from pushing farther even though they still have a huge amount of temperature head room and are well within the Intel spec. Like wise, how can I tell a person at 3600MHz that DTS=25 is a nice safe number when they have trouble running Orthos for any length of time whenever the room temperature goes up by 1C.
People are getting far too hung up on absolute temperatures and even Distance to TjMax. As long as you leave yourself some head room to TjMax so your processor isn't throttling and you're Prime stable, then there isn't any need to even monitor core temperatures. If Intel thought that your processor was going to blow up at high temperatures near TjMax then obviously they would lower TjMax to avoid warranty claims. Instead, Intel has raised TjMax by 10C for the new 45nm dual cores which to me is a pretty good sign that core temperature just isn't that big of an issue.
Didn't they lower it again from 105 (Kentsfield) to 95 (Yorkfield) ? or are you referring to your 85 of ya E6400... or are the new C1 steppings agan 105 ? You got me puzzled here
With an E8400 at low temps, it's hard to assume anything.Quote:
Originally Posted by TL1000S
Brama's log file looked like this:
13:47:05 -6 25
13:47:30 -5 25
13:47:35 -4 27
13:47:40 -6 26
13:47:45 -6 26
13:47:50 -5 26
13:47:55 -6 25
13:48:00 -7 26
13:48:05 -6 26
which shows the sensor in Core0 reporting temperatures that are way too low regardless of TjMax.
Is your Core1 truly stuck or is it just reporting high idle temperatures like my E8400 does? Without a lot of detective work it's hard to say which one of your sensors has which problem but it's possible that neither of them can be used for accurate idle temperatures no matter what TjMax or software you use.
I was just comparing my dual core Conroe E6400 to my dual core E8400. TjMax has gone from 85C to 95C for these two processors.Quote:
Originally Posted by Leeghoofd
You bring up some good points, but I was thinking of a much simpler concept. The colours would in no way be an indication of stability (stability needs to be proven using recognised tools like Prime95 and Memtest), but purely represent distance from tjmax. In fact, your last paragraph pretty much sums up the idea behind this concept.Quote:
Originally Posted by unclewebb
We could choose arbitrary constant values from tjmax to determine colour. For example, 20C below tjmax could be the start of the red zone, and 45C below tjmax could be the start of the yellow zone. These values could even be configurable (.ini file). By also showing the actual distance from tjmax (a coloured gauge may not be precise enough for people to make informed decisions), each individual can choose whether to push their system further. For new overclockers asking 'what's a safe temp' or 'are my temps ok?', the colour-coding should make things much simpler by providing an indication as to what's considered safe. Those with more experience may well want to push things closer to the limit and can therefore configure the values.
This should also fit with Intel increasing/decreasing the tjmax values for their processors. If Intel were to raise tjmax, it should be an indication that the processor can run hotter. The colour zones should therefore still be relevant.
I just like this idea as a concept, although this is a very simplistic design. Whether it's wise to use a constant value like '20C from tjmax = red' for processors with varying tjmax values is debatable. In fact, there may be the same level of discussion around these values as there are about the real tjmax values for each processor family :rofl:.
I'm not proposing that you should consider changing the overall design of Real Temp. The results of your investigations are extremely useful, and I still think there will be a lot of demand for programs that attempt to report absolute temps for processors. In fact, I would expect a lot more resistance to a new concept like this as most people seem to have this need to know absolute values. I do see a lot of value in this concept though, and if my programming skills were not quite so limited I would like to attempt such a program myself.
According to your own tips/documentation here it should be stuck.Quote:
Originally Posted by unclewebb
(Run Test sensors indicates this). :)
I have numerous runs for stability tests at various combos VCore/Clocks, and the second core (1) shows *never* below 38 c at idle or low to mid load. (OCCT's graphs are good for documentation purposes).
It is only after some real hard stresstesting it will follow core0 from 38 to eg 55-60 (depending on VCore).
If that's not a sign of a stuck sensor there is no such thing as a stuck sensor. :D
uncle so to test sensor movement i should
set my proc to 333x6 lovest multi and vcore as lover i can 1,1v ?
to calibrate sensors -- - 0 + ++, right?
moderators: i suggest to sticky this thread.
My BIG problem with all this is simple...My new build is a 9650 in a 780eVGA. Right now I am trying at 1333 X 11...
Realtemp which is SUPPOSED to tell me the REAL temp is at 49-49-53-53
EVEREST is telling me 59-59-63-63
HWOmintor is telling me 58-58-62-62
I am in the middle of running PRIME...WHo do I believe? You start getting to mid 60's and I need to know is it 65 or 55!
You sir, definitely have a stuck sensor. Nice to see that RealTemp agrees with your own common sense. :DQuote:
Originally Posted by TL1000S
Doctahg: I think you missed my previous post. If you are Prime stable and have left some head room before TjMax, which you have plenty of, then absolute temperatures don't matter one bit. 50C, 55C, 60C, 65C and even 70C doesn't matter while running Prime as long as you are stable. You are within the Intel spec and you aren't going to hurt anything. With no documentation of TjMax by Intel and no user testing of TjMax for your new processor, absolute temperature is no more than just an interesting number no matter what software you are using to report it.
TEDY: When running Test Sensors just keep your MHz and core voltage at whatever level you normally run them at. This test puts an equal load to each of your cores for a short period of time and then does a before/after comparison of core temperatures. Properly operating sensors should move more or less equal amounts.
The post by TL1000S is a pretty classic example of a stuck sensor. One of his cores goes up 12C during this test while the the reported temperature for the other core doesn't move at all.
The results are effected by MHz, core voltage, 45nm or 65nm, dual or quad core, room temperature as well as what type of cooling you are using. I guess we need more users to post their results of this test with all of the above factors mentioned so other users will have something to compare to.
My E2160 reports 13/16 at 3000 MHz and 1.40 volts. The slight difference might be a sign of poor IHS to core contact, or a poor job of thermal paste installation on my part or it might be pretty normal. My E8400 moves far less during this test and was better balanced.
When trying to find an appropriate Idle Calibration you need to run your CPU as cool as possible and see how it compares to what I've found. When using a good air cooler at 6x266MHz and 1.10 volts, stable idle temp readings after warm up of about 4C above ambient seems normal. If your CPU is a long ways from that then try a different Idle Calibration factor.
~aoe~: It's possible to create a colored graph for DTS temperatures but as my testing shows, I don't think it really gives a user any usable information since stability varies with core temperature, voltage and MHz. Another thing I've noticed is that Intel might have adjusted TjMax from 85C to 95C since the original dual cores came out but I don't think that has changed the temperature where these processors will run stable when overclocking and over volting like some of us like to do around here. When well overclocked, both my E6400 and E8400 lose stability when the absolute core temp goes over 60C. The change in TjMax is only important if you are near default specs. Too many variables and no useful information means I probably won't implement this feature.
Uncle did you get my email?
It's on my things to do list!Quote:
Originally Posted by naTTan
Lol, take u're time, i am in no rush. Just worried the email didn't reach you :P
Thanks again for the help.
@unclewebb:
Your comment about absolute temperatures is something everybody should learn to accept as *facts*. :up:
I am so tired of telling "noobs" in norwegian forum about why they should not worry about more or less dubios software "reading" sensors but rather use som common sense.
The *only* way to measure temperatures are like you describe in the first (and later) posts about your session in the basement before designing/coding your genious little app. :D
one of the best "all-in-all" articles/posts in regards to CPU temps and calibration i just found here:
http://www.tomshardware.co.uk/forum/...perature-guide
But i have a problem, this guy gives tips how to calibrate using TCase using speedfan. Assuming that TCase is (depending on cooling solution) about 4 C higher than ambient.
Calibrating temps as he says i get pretty high temps - and here is the difference in what you said:
You said to calibrate using least possible Vcore (like 0.8) and he said "in bios leave on auto"....which i think is wrong. Since "Auto" would give each individual CPU different vcore, right?
So...what we need...really some "all-in-one" half-way accurate measurement of Difference Tcase --> Ambient at given Vcore. If we know its about 4 deg higher than ambient with high-end cooler (Ultra Extreme etc.) for a Q6600 at 0.8V this would already help a lot!
That above quoted article is very thourough...but as said i think he's making the mistake to calibrate with Vcore->Auto, but i can see being Tcase about 4deg higher than ambient with the least posssible Vcore like 0.8-1.0 and everything else in bios at default.
G.
well considering that all software programs reading the dts will report the same Distance to TJ max, (coretemp and realtemp) the colored graph could be the same at the point you choose to go from yellow to orange to red or whatever...Quote:
Originally Posted by ~aoe~
both programs even though they show 70C temps or 80C temps at load are going to show the same 25C distance to tj max.
jaredpace: I agree that all software that properly reads Distance to TjMax will report the same number for the same processor. The problem is that a reading of 25 on my TjMax=85C E6400 is completely different than that same 25 reading when I'm using my E8400 with a TjMax=95C. From that I see more confusion developing, not less.
Instead of more temperature information I think we actually need less. Once you're running stable there isn't much need to monitor core temperatures. The Intel Core processors do a great job of taking care of themselves.
unclewebb, i very, very much agree. We now hae half a dozen or more ways to measure temps...and i can see differences +/-10 deg which translates, very real, that i dont have the slightest clue what the temps actually ARE :)
In the future i think i will solely focus on DT->TJmax since i have so many different values depending on calibration and program used.
I also have a problem "understanding TCase, i was looking at that Intel Spec sheet http://download.intel.com/design/pro...s/31559205.pdf
in particular.
here, example 95W CPUs you see a chart giving Max TCase depending on the current Wattage/power draw of the CPU....so at max 95 Watts Tcase equals 72.
For the same CPU Tcase equals 58 deg at 50W...and so on.
So max. Tcase is not a fixed value but always in relation to the current Wattage, IF I UNDERSTAND THIS RIGHT.
So i have a problem understanding what Max TCase actually means...and also if it would apply for overclocked CPUS....say according to CPU wattage calculator i clock my CPU 3600 Mhz at 1.44 which equals 197W.
If i would expand this chart (assuming it would go on in a linear way!) i would come to max Tcase at 197 W equals 97 deg.
There must be a misunderstanding about the meaning of "max Tcase" since for the SAME CPU, how can max Tcase be LOWER with lower Wattage/Power Draw...so max TCase certainly canot indicate a safe max temp if its variable? Rather something like "expected" TCase temp at this and this wattage? <--- guessing...
i actually *do* think we have a pretty good dilemma here, something which deserves way more attention and people like unclewebb with programs like realtemp.
For example: If i use the new OCCT and use the internal temp measuring, which is the same temp as coretemp shows:
I do some Prime/OCCT testing and quickly get 70-71 temp on two of my cores. If i leave OCCT at default it would already complain "CPU too hot!"...which just doesnt sound right since with 70C/core i STILL have 30 degrees to go according to "Distance to TJMaxx".
So...either the CPU is "too hot"... (70/71 degrees)...or i still have *plenty* headroom with 30 degrees far from TJMax/Throttinling temp...so which one is right?
We know that the only "real" information we have is "Distance to TJMax", thats what the sensors say. Ergo, the current 70/71 core as coretemp/OCCT shows and warns "it's already too hot!" sounds like total nonsense to me.
We also can assume that Intel specced TJmax relatively relaxed and below extremely dangerous temp, Intel for sure doesnt have an interest in people RMAing one CPU after the other because the throtteling would happen already at very extreme temperature. It would defy the whole idea.
I am really thinking about writing a small app showing a bar or something according to DT->Tjmaxx and ignore whatever other programs say in regards to temps.
http://64.233.167.104/search?q=cache...nk&cd=20&gl=usQuote:
Originally Posted by flexy
read slides (espec. 10-17 and 11 and 14 in particular) its from a senior intel engineer, it is a partial explanation, not a full one though. There was another place that had more info, but I have too many links to look through, dont remember which one...I will post it if i can find it....assuming I saved that one.
Edit: What is interesting is if we could find the exact thermal resistance (intel knows it) you could calculate tjmax from their formula.
Tjmax C = Tcasemax C + (thermal resistance from tj to tcase C/W x TDP max W)
this is specified in many intel and other white papers, cant find the one I got attached pic from...yet.
Interestingly if take my E6850 and fill in numbers, you get
tjmax = 72C + (.38c/w*65W) = 96.7C, c/w figure based on intel specs for each, though using c-a instead of j-t, I believe since cooling solution is what drives the tcase figure, it may be relevant to use such, also adding up other quoted resistances gets you close to this figure...though it will vary from part to part, dont believe it is huge variance.
but add in the testing that is done with ? individual/batch corrections such as here http://www.yeongyang.com/report/RPG-...516-05)RPG.pdf ...and could be either still.
E6x50 and E6540 with 4mb L2 and CPUID ending in 6FB
tjmax = 72C + (.38c/w*65W = 96.7C
And For E4000 series with CPUID ending in 6FD or 6FB (CPUID read by coretemp)
tjmax = 73.3 + (.38x65) = 98C
For E4000, E6000 series with CPUID ending in 6F2 or 6F6
tjmax = 61.4 + (.38x65) = 86.1C
E6000 CPUID 6f6 and 4mb L2
tjmax = 60.1 + (.38x65) = 84.8C
E8400
tjmax = 72 + (.38x65) = 96.7C
Also given Go stepping q6600 are 95W TDP, cpuid 6FB, with tcase max 71,
and B stepping q6600 are 105W TDP, cpuid 6F7, with tcase max 62.2,
than B stepping tjmax is 11C lower than GO, which is why those with B stepping complain they run ridiculously hot, if subtract 11C from tjmax on B stepping, temps on B stepping would make more sense, ie 11C lower.
Unclewebb can you confirm your CPUID, I am betting you tested a 6F2 or 6F6 (if E4000 or E6000), and Gigabyte has 6FD or 6FB for his E4000. I can not be 100% sure of c/w, but I am pretty sure there is at least a ~12C difference in tjmax between those two cpuid, even though both e4000 series.
exactly the point. your software is a milestone and everyone is grateful for you making it for us. even if RT or CT is correct, and no-one including RGE knows the exact TJmax to find out for sure, the D to Tjmax is the only real relative value at hand. As for 6400/4xxx/8xxx having different thresholds, I'm sure that is all programmable into a self-adjusting graph once the program recognizes the cpu string. I applaud your program, yet am only expressing my concerns for what is needed next. People seem to agree.Quote:
Originally Posted by unclewebb
I think this is a definite way to show "less temperature information"
also if you have 25C distance to tjmax on an e6400 and the same 25C distance to tjmax on an e8400, are you saying that the headroom varies for either chip in that situation?
Is it because the higher 95C tjmax yields a greater range for temperature activity or just because it is produced at 45nm?
If the 6400 idles stock at 35C and the e8400 idles stock at 35C, I could see how there would be a difference. However this difference is programmable or adjustable with a ratio im sure.
if the e6400 at stock 35C temps can goto 85C tjmax, then it has a 50C temperature range to allow for overclocked load speeds. Likewise, the e8400 at stock 35C temps which has a 95C tjmax or 105C tjmax, either has 60C or 70C of temperature range headroom to achieve your overclocked load speed.
a data set per graph made specifically for each generation would show the same graphical output giving the user the same idea. It would definitely start an end to all the noobish "i see 58C here but 68C there which is right how much more can I oc??" bullcrap.
Quote:
Originally Posted by ~aoe~
The values are fully known around given by Intel Senior Engineers. They're just NDA material that's all. ;)Quote:
Originally Posted by jaredpace
I'll post as soon as I can get on the drive with data saved on it. Not been back to flat in 9 days yet all my gear is there and this is someone elses.
As for DTS feedbacks ignoring TjMax; then true, they have to be accurate in the first place to judge off them. They were fudged on the new 45nm lineup for too many around, hence why the mobile 45nm were all reluctantly delayed till C1 step (google) to just fix this error, but they worked perfectly fine on the vast majority of 65nm SKU lineups.
These are working stock E8400 DTS, just like the confirmed ES samples: http://www.xtremesystems.org/forums/...&postcount=138
Very close full load temps, ambient, speed, voltage, difference is only cooler: TT120 vs. Xigmatek S1283 there.
Keep in mind, from 65nm to 45nm, judging off just process node alone you are supposed to get 20-30% increase in circuit density and power density, 30% improvement in active power consumption, same for leakage power and just going off the surface area, you have a decrease in chip surface area to dissipate heat. Heat dissipation depends majorly on contact surface area and you now have 214/107 sq.mm of 130/65W compared to 286/143 sq.mm for 130/95/65W before. Electrical stability depends on heat density too, W/mm² and the new 65W@45nm have a heat density of 0.607 W/mm² compared to 0.455 W/mm² for 65W@65nm, hence why they have a lower voltage and heat threshold for stability and integrity and will require better cooling at the same wattage since you have less surface area yet more heat cramped into it. This is why you can't compare relative temperatures between both nodes.
rge I read Intel ignored and deleted your query but some of your questions were pushing the boundaries too much for them to reveal without NDA through a SFAE. If you want answers, contact your local distributor and go through them to contact with an S.FAE. The only details you really require are already known though. I have the details for a while now and how to retrieve them, just wait a few.
well if you know for sure, you should probably tell unclewebb. duh :)
Preliminary results are just in. TjMax=95C is looking correct for my GO Q6600. It's heating up a lot quicker than a dual core when I pull the heatsink off but I think I should be able to get some camera screen shots like in previous testing. Details to follow. :D
http://img206.imageshack.us/img206/6716/newquadwr5.png
I have a pile of pictures from 50C to 75C that look a lot like the following two where RealTemp is tracking the IR gun pretty closely.
http://img521.imageshack.us/img521/9018/temp654so6.png
http://img182.imageshack.us/img182/2895/temp65rr4.jpg
The picture in my previous post shows core0 and core1 tracking each other closely while core2 reports an idle temp 3C cooler. Perfect motivation for individual idle calibration for Quad cores. :D
I tried to move the IR thermometer a little left of center when testing to hopefully be over top of Core0 and 1 but this is far from an exact science. I'll be exchanging the camera for a small high speed fan tomorrow so I can do some more testing without the worry of controlling a fresh quad without a heatsink on it. The heatsink isn't attached at the moment so I can remove it, take some temps and then throw it back on to keep things from getting too out of control.
The difference in TjMax between RealTemp and CoreTemp is only 5C so it's not worth getting into an argument about test methods, theories etc. So far I don't see any evidence that warrants an increase in TjMax to 100C so I'll be leaving it at 95C. It will be interesting to see what my idle temps are with a HSF on it compared to room temperature.
The post above yours read:Quote:
Originally Posted by jaredpace
:)Quote:
Originally Posted by KTE
So communication with our contact went as following. It's non-verbatim, I have to protect confidentiality but it's very close to the exact words said. Hope I don't get in trouble for this. :(
Kinda makes things clear as crystal there. So, if you want to follow Intel like you do for MSR 0x19C, then check MSR 0xEE and you can go on from there combining the two.Quote:
KTE: I need a favor from you. Have you seen this thread? Common users don't have enough information on temps. and so we're guessing based on some unprofessional and limited testing and docs. You confirmed E8400 temps at 51C load earlier when you came in January but what about how we can work out temps? I know the 0x19C MSR but word has it that no one knows the correct TjMax and we're all guessing... why won't you release these details?
Contact: Hmm.. were'd you hear that? AFAIK all developers who inquired know.
KTE: I talk to most oc software developers regular, which ones are you talking about so I can confirm?
Contact: We released this info back in 2006 and since then at every stepping update but you know it's only allowed under NDA in the BIOS writers guide. Ask Rudolf who made the Lm-sensors driver for Linux, or Matthias who develops CPUTempWatch or Fiery at EVEREST, Franck at HWMonitor, or Arthur who makes CoreTemp just to name a few, they all should know the details as they use them in their software calcs.
KTE: I'll definitely ask them but why was this info never public?
Contact: Check.. or if that's not enough, then ;)
KTE: Hmm.. damn I searched everywhere but you knew what to search for and I didn't. These guys knew it all along :D
So.. let me get this straight: retrieve MSR 0xEE, bit:30 0=100 1=85 on earliest, for my Q6600 G0 its 0=100 1=85 and thats factory set TjMax?
Contact: Spot on Tye ;)
============2 days ago again============
KTE: Just seen you're on, quickly, some more temp. qs for you.
So you're saying devs know the info you gave me about MSR 0xEE to get TjMax for a Core 2 CPU plus even the latest TjMax updates for Penryn?
Contact: Yep just check their software and they'll have all the updates to them from official authorised documentation for our Penryn 45nm lineup.
Contact: Oh look :)
KTE: 0=105 1=95 Penryn.. hmm, that makes sense, that's why most of them were in sync I guess. Thanks [names contact], you've been big help many times now!
BTW what's the temperature Abit EQ/Gigabyte EasyTune and EVEREST "CPU" reports then? Sorry had to ask that its confusing me, it's lower than DTS...
Contact: Probably Tcase, there's an analogue Tcase sensor there as well as the DTS sensors in each core, which motherboards can read and EVEREST probably reads that temperature. There is nothing else that will give a CPU temperature value but you're best off asking Fiery at their forum
KTE: Ah ok. That sorta brushes things up. :)
Can I get quick peak access to the guide?
Contact: [names my uncle] can, just need to sign the NDA that's all :)
KTE: No thanks :D
KTE: Just reading link you gave... Fiery mentions any problems with the TjM/Temps and it's your fault :D
Contact: Hehe, well someones got to take the blame. Temperatures shouldn't matter too much under normal warrantied conditions though
7 software temp. values compared on my Q6600 G0:
http://img100.imageshack.us/img100/2186/tempstv6.th.png
Hope it helps, sorry for delay but I'm heavily backlogged on more than 180 PMs at just XS, many emails and many threads.
"Use of bit 30 in MSR 0EEh is *not* valid for desktop, workstation or server processors based on the Intel(R) Core(TM) microarchitecture. However, for mobile processors, this assumption *is* valid."
Intel does not have any MSR that exposes the trip point for the TCC in the Core microarchitecture.
Lexi S.
Intel(R) Software Network Support
----------------------------------
Someone at Intel is not being 100% honest. The original thread at Intel where this was posted has been removed but this quote comes from a page on XS:
http://www.xtremesystems.org/forums/.../t-136804.html
Lexi checked with the engineers and went back and forth discussing this issue for 3 pages and the above quotes come from his last post before locking the thread.
This has always been the problem. Software is reading this bit and assuming TjMax based on this bit but I agree with Lexi. S that this bit is not valid.
Time to play musical processors. The latest version of OCCT assumes TjMax=100C for my E2160 which in my testing is definitely wrong. I'll have to throw it in and check MSR 0xEE.
By using a high speed hand held fan I was able to do some more testing on my Q6600. With this method I have no problem maintaining a core temperature anywhere between 50C and 60C while taking IR readings. I had plenty of time for things to stabilize and at every point, RealTemp with TjMax=95C was exactly equal to the IR thermometer.
Edit: Just re-installed the E8400 and sure enough bit30 if MSR 0xEE is not set so Everest and CoreTemp and other programs are assuming that TjMax=105C. For better or worse, RealTemp is based on real IR thermometer testing and I will continue to disagree with using any information from MSR 0xEE to determine TjMax for the desktop core processors.
http://img359.imageshack.us/img359/7...msr0xeewc0.png
Thanks for the info, but several questions/problems. Btw...first let me preface this by saying I am not just being argumentative, and I think you, unclewebb, I and anyone else posting on this subject are simply after the truth, nothing more, nothing less, so I commend anyone's input, especially those with contacts. But with that in mind...Quote:
Originally Posted by KTE
1) If coretemp, everest, etc authors are all in "the know", then the tjmax of unclewebbs 1st cpu as 85 validates his method of measuring 85 with IR gun when DTS=0 and that the gradient from core to casing is less than 1-2c under those conditions,. Since intel white papers say resistance is even less for smaller dies like 45nm, ie E8400, and both coretemp and everest say tjmax is 105, how can you explain that under similar conditions when DTS=0, casing temps measure 95C, ie 10C lower? Either the resistance of E8400 casing would have to be much higher which contradicts intel papers, or resting power under similar conditions would have to be much higher for E8400...tj=tcase + theta(j-c)xTDP, given similar ambients, 10C difference is hard to explain.
2) If tjmax of my E8400 is 105, as coretemp and everest report, and the cpu temp is diode temp, then that is saying the gradient from core to diode of 15C running orthos is accurate on my cpu as well as on others which are similar. However when I load one core, the gradient from die to die is only 1C. Not only is that seemingly impossible to be 1C from die to die, but be 15C from die to between dies, but it directly contradicts the die to diode gradient in this paper in figure 5 http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf, and directly contradicts intels white papers stating this gradient has decreased with smaller die's, and directly contradicts statement on intel slide presentation by intel engineer that this gradient is smaller on smaller dies, and directly contradicts realtime measurements. Note that tdiode is still between cores and on opposite side of TIM1 and IHS as Tcase measured.
3) If that information on tjmax was released, why do the developers, such as coretemp author post on intel forums, asking for such information, and then a long argument ensues regarding he cant get NDA, because he is not a company....and there are recent arguments?
4) In one of your links above this statement is provided by a software person on intel forum
"It is true, some are calibrated to 100C, some are calibrated to 85C. We should call that value as Tjmax for Conroe/Woodcrest/Clovertown instead of Tjunction...
It is correct, the Tjmax value, whether it is 85C or 100C, can be read from MSR 0xEE. This (85C or 100C) should be documented in the processor spec."
We have all seen that and discussed before, it was NOT written by an intel engineer but one of software people, who repeatedly state they do not know the answers and always ask the engineers. Later on in that same thread, page 2, which is now suspiciously locked to viewing, an intel engineer who was introduced by intel as knowledgeable on the subject stated, "from intel specifically from one of our internal contacts for the processor documentation", that states,
"Note Tj is not a fixed value and the IA32_TEMPERATURE_TARGET[15:8] value can vary from part to part. Tj is also not software readable."
Also in direct response to Rudolphs question "Perhaps put 100 as TjMax and write to documentation that it is safe until you reach 100? (and also that this temp is artificial) Do you think users might benefit from MSR_TARGET_TEMP too?"
Intels answer "No."
That leads one to believe the first post was from a confused intel software person, who was dispensing information from mobile cpus, not deskstops. Either that or the second engineer who was introduced as someone who knows, was lying, which I doubt.
I would be happy if intel did let us calculate absolute temps, but for me it is not anywhere near crystal clear...yet.
The DTS reports an offset to "throttle", which is a "hardware set" value(resistance) within the pcb/core of a CPU (if I have understood some threads correctly).
This is a relative temp, not a "real" temp.
The author of realtemp have tried to do some "reverse engeneering" to find the actual temps. :up:
If we assume that all CPU's (chips) are equally processed on a wafer.
That there is no margin of errors in the "burning" process of the "hardware set" of DTS.
Then all should be fine.
We should now know that this not always is the case. "Stuck" sensors like many have reported for E8n00 indicates one (of probably many) errata.
The only way to monitor "real" temp is to hardwire a calibrated sensor into the case (to measure casetemp). This has long been documented in Intel white papers.
To do IR to measure is difficult/impractical for OC/ordinary use (as shown by "unclewebb") as it means you have to run your system without any HS/cooling. :)
My conclusion:
Just give a damn in "temperatures", but go for safe stable 24/7 use.
Or go for "broke" setting personal records as we do here on xs.org.
"Temps" are just numbers. Sometime they may be "real" (accurate) temperatures, but most of the time they are not.
I think. :)
I have an E8400 installed at the moment and it is the whole reason that I decided to write RealTemp. The reported temperatures for it just didn't seem right.
The heatsink is off but I've got a small high speed fan pointing towards the CPU.
http://img394.imageshack.us/img394/4091/testbedse9.jpg
Some users might think that without a heatsink attached that core temperatures would be out of control but far from it.
http://img394.imageshack.us/img394/9...400idlesn8.png
They were dead stable for over 15 minutes which was long enough for SpeedFan to draw a complete graph with only brief moments of the temp going up by 1C and then dropping down again.
After 15 minutes of this it was time to get out the IR thermometer. There was no hurry so I had a chance to scan around until I found the highest temperature which was 38.8C.
http://img395.imageshack.us/img395/6454/temp388czl2.jpg
CoreTemp is using TjMax=105C and SpeedFan is using TjMax=100C.
RealTemp is using TjMax=95C combined with its exclusive Idle Calibration feature.
Have a good look at those 3 programs and you decide which one seems to be displaying the most honest core temperature.
To believe TjMax=105C, you have to go completely against Intel's own in house testing of their processors at idle. Bit 30 of MSR 0xEE is not valid for this E8400.
If you would like to check your own processor you can use CPU-Z and do a Register Dump or you can head to openlibsys.org and download OlsMsrEditor 1.1a. Both programs will let you read MSR 0xEE.
@Unclewebb:
Excellent documentaion of your work/research. :up:
It would be interesting to do the same exercise provided I got me an IR-meter.
So far I have concluded that my E8400 must be totally wrong regarding DTS/TjMax settings.
There is no way Core0 can be 11 degrees @ 1.33VCore (idle) as reported by Realtemp assuming TjMAx=95.
The DTS for Core1 we can conclude is "stuck"/malfunctioning (see previous posts).
I think you will have to test quite a few E8n00, Q6n50, E4n00, E21n0 before concluding anything regarding a "pattern" between observed temp using IR-meter related to a certain TjMax offset.
Ideally there should be a pattern as there should not be significant deviation between CPU's on a wafer/different wafers using the same production technology.
But clearly we regulary see "batches" of "hot" vs "cold" versions of a given CPU (ie Q6600 B3 vs Q6600 G0).
Intel explains that as "the higher thermal solution performance of the Intel® Core™ processors" for explaining the higher tcase revisions of same processor.Quote:
Originally Posted by TL1000S
For example, E4x00 comes in two flavors,
CPUID 6F6 or 6F2 with tcase 60C
and the higher thermal version CPUID 6FD or 6FB with tcase 72C
Q6600 G0 stepping has tcase 72C
Q6600 B stepping has tcase 61C.
Since Tjmax=tcase +theta(core-case)xTDP, it is likely that the B stepping Q6600 and others with tcases in 60 have tjmaxs lower than those with tcases in 72 range.
Someone with E4x00 with tcase of 72 needs to check IR reading to compare with IR reading of Unclewebbs as he has the 60C tcase version. Same for GO versus B stepping quads. Given theta(core-tcase) is likely not that different, it would suggest a 10+ difference in tjmax, hence cause the B stepping quads to possibly read temps artificially too high. But wont know for sure til someone IR's one.
TL1000S: My personal opinion of your processor is that the DTS on one of your cores is stuck and the other is suffering from the "reporting way too low" problem that Brama had a few pages ago. If that is the case then it's impossible for RealTemp or any other program to give you accurate idle temperatures without a thorough IR thermometer calibration. I still believe that TjMax=95C across the entire E8x00 series but I'd need a big box of processors from different batches and my IR gun to definitively prove it. For some reason, I don't think Intel would be in any mood to share after seeing my Test Sensors feature.
rge: The funny thing about my E2160 is that it has the higher Thermal Specification of 73.2C but all of my testing leads me to believe that TjMax is only 85C. My original E6400 had a TjMax=85C and a Thermal Specification = 61.4°C. I lost faith in trying to figure out TjMax based on Thermal Specification after this.
RealTemp does assume TjMax=85C for the B3 Q6600 vs 95C for the G0 Q6600. They were ready to hang me on one forum for this theory but no one has stepped up, yet, to prove me wrong. A user on XS complained about this but after he did some testing, he wasn't able to show me anything to contradict my original TjMax assumption. A B3 Quad contains two B2 dual cores which have a TjMax=85C so I've assumed that the B3 would be the same. If someone wants to loan me one I'll be happy to do a thorough test. :D
TL1000S: Does a (++) Idle Calibration get your idle temps close to believable? This feature is not exact but I thought for you processor it might be pretty close on the one core that isn't stuck. Show us a RealTemp screen shot vs your water temp if you can.
I'll explain in other ways to the individuals of main relevance here and we will continue discussing and investigating with honesty. I think I've retreived what was needed but as the inidviduals concerned know I'm restricted quite heavily now, apologies. :(
You have the 6FD 2160 that is 73 Tcase http://processorfinder.intel.com/det...px?sSpec=SLA8Z, and not 6F2 that is 61 tcase http://processorfinder.intel.com/det...px?sSpec=SLA3H ?Quote:
Originally Posted by unclewebb
Edit, never mind just found your E2160, it is the 6FD with tcase 73. That is pretty screwy. Both have same TDP, and if Tjmax is same, then theta (core to tcase) is different between the two.
plugging in tjmax of 85 for both, 6F2 (L2 with tcase 61) theta is .38C/W which is typical, the 6FD (MO with tcase 73) is 0.18 C/W. If both have tjmax of 85, I wonder if one uses solder and one a die attach adhesive. But the theta of one is twice the other which is certainly plausible given all the white papers describing the different approaches for die attach adhesives.
Now, I really want someone to send you their B stepping quad, or get gigabyte to measure tcase of the e4300 version with other tcase. maybe intel is trying out in mass scale different bond adhesives, ie solder or other.
Yes, ++ get idle temps close to believeable. :)Quote:
Originally Posted by unclewebb
I have noe screenshots/photos available as of writing this, but idle temp is about the same as watertemp using (++) Idle Calibration.
I shall check again.
Keep up your excellent work! :up:
TL1000S: My graph in post #1 of this thread shows what I've been finding. The DTS data changes at a different rate than the change in core temperature once you move farther and farther away from TjMax. I don't have a cooler attached to my Q6600 yet but I've got a pretty good feeling that it too will show this sort of behavior.
The (--) and (++) Idle Calibration were designed with my E6400 and E8400 in mind as well as what I've gathered from other users over the last year or so. I'll admit that trying to correct for a sensor that does not perform 100% linearly is a bit of a guessing game but I think correcting down low where the actual problem is, is a far better solution than software that has been adjusting TjMax upward to try to correct for "too low" idle temperatures. If TjMax is incorrect then your reported temperatures will be mostly wrong across the entire temperature range except for a single point near idle.
rge: Once I sell off one or two processors I would definitely get an early E4300 to solve that mystery and if I trip over a B3 Q6600 I wouldn't be able to resist adding it to my collection. :D I'm very confident that TjMax=95C is correct for my G0 and it will be interesting to see idle temps in my 11C basement after it sits all night.
As an aside, I did a comparison of how much power my computer uses at idle with 3 different processors. Everything was equal for each test like core voltage and MHz.
E2160 - 115 watts
E8400 - 119 watts
Q6600 - 125 watts
The 45nm E8400 may be a more efficient design but the extra cache uses more power and creates more heat than a dual core with only 1M of cache. The difference is only 4 watts though which any good air cooler would have no trouble dissipating at idle. This helps confirm my theory that processors set to a fixed low voltage and low MHz at idle should idle at very similar temperatures across the dual core line. Sky high idle temps that some software is reporting for the E8400 just can't be right.
The Quad is a little different and uses more power and creates more heat but even so I don't expect that it's idle temperature is going to be way higher than a dual core in this test. It has twice the surface area compared to a dual core so better heat transfer will also help keep idle temps down. Maybe 6C over ambient at idle is reasonable for a Quad at 1600 MHz and 1.10 volts.
On B3 Q6600 the TjMax would be around 85c right ?
This is what RealTemp assumes but I believe other programs like CoreTemp are assuming 100C. If anyone out there has a B3 I would appreciate it if you could run OlsMsrEditor and do a RDMSR of 0xEE and post your findings. I would also appreciate it if you could run some low speed, low voltage tests and post your results.Quote:
Originally Posted by Demo
http://img369.imageshack.us/img369/7...msr0xeebr3.png
On my B2 E6400 with TjMax=85C, this first digit is a C so bit 30 is set while on my Q6600 it is an 8 so bit 30 is not set. I'm not convinced that this bit has meaning across the entire desktop line but I'm also not convinced that it doesn't have any meaning. I'm just doing my research and will continue to do so until everyone has a program they can trust. I might be at this for a while. Reports for different CPUs is always appreciated.
My Q6600 has the heatsink on it and it is pretty obvious that the sensor for core2 is moving at a different rate than the other ones. The difference between cores was usually 3 during previous idle testing with the heatsink sitting loosely on top. Now with the lower core temperature the delta sometimes jumps back and forth between 4 and 6 compared to the other sensors. Even CoreTemp is reporting it at or below room temperature. These sensors were never designed for reporting idle temperatures and this is just another example of this. This isn't just a 45nm or dual core issue.
With a (++) Idle Calibration on this core the temp looks very believable but the other cores only need a (+) calibration so individual calibration for the Quads has just moved up to the top of the things to do list. I'll try to do some more testing tomorrow morning from the basement. :D
Here is my contribution and greetings from Sweden.Quote:
Originally Posted by unclewebb
So how do I calibrate for a quad when cores 1,2 seem to be ok, but 3,4 are at ambient?
http://img379.imageshack.us/img379/9...1600vo2.th.jpg
Give me a day or two and I'll have an update of RealTemp ready for you. My Q6600 is similar with core0 and core1 being identical, core3 is close to the first two and then core2 is always significantly less. I plan to go with individual core calibration which will help things out at idle but I think with my quad there is always going to be a degree or two of difference. This could be a sign of two cores operating at a slightly different temperature but after seeing all the E8x00 sensor issues, this looks like more of the same. Since Intel doesn't design these sensors for reporting idle temperatures, they probably didn't do anything to balance two different dual cores when they were building the quads.Quote:
Originally Posted by lowtcell
In preliminary testing with TjMax=95C, cores 0,1,3 need a (+) and core2 needs a (++). With this correction, idle temps look reasonable and load temps at around 60C are close to exact.
Thanks ChrisZ for the info.
Thanks, I'll wait for the update :up: At full load, cores 3,4 are always 2c bellow 1,2, no matter what clocks or Vcore I'm at.Quote:
Originally Posted by unclewebb
I discovered that my EVGA 780i CPU Core voltage goes down to 0.50v. I decided not to tempt fate to test my idle temperature so I set my Vcore to 1.125v which gave me 1.08v. My ambient temperate was 22c.
Attachment 75959
This is what I got putting a load on my Q6600 at 3600 MHz (400 x 9).
Attachment 75960
msgclb: Your Q6600 looks similar to the one I have. The idle temp sags on core2.
I'm presently testing individual idle calibration and it seems to get my Quad a little better lined up at idle. I get almost exact load temps (~50C to 60C) on core0 and core1 compared to an IR thermometer with TjMax=95C but the idle temps are too low so setting a calibration of 1 for three of the cores and a calibration of 2 for core2 brings the idle temps up to believable.
Here is a pic of the low MHz / low core voltage test that I recommend for setting your calibration factor.
Here how my Q6600 looks at a room temp of about 18C to 19C.
http://img186.imageshack.us/img186/3...temp225wu4.png
I need to do some more full load testing before the next release.
Can someone run OlsMsrEditor and read 0xEE on an E4300? I'm very curious about that one.
Thanks.Quote:
Originally Posted by unclewebb
my crappy B3 X3210
http://img220.imageshack.us/img220/5539/realtempnr8.png
here is an e4300:Quote:
Originally Posted by unclewebb
Attachment 76036
MSR 0xEE is the same for both cores:
MSR 0x000000EE edx=0xA8000000 eax=0xC17D4700
there's a version 2.25 already? what's the difference from previous versions?
There is 2.24 as far as im aware of.
Thanks by the way Unclewebb, i was wondering why my core 1 was at 65C through everest at 1.42v underload, now its way more believable at 55C core1, 45C core2, thanks so much man.
I have one question, on both of my cores they have VERY high movement. usually core 1 has 13(almost always) and core 2 has 9 to 11 usually. This is with full idle like the thing stated, is there anything wrong? Is it good? it said it should be 0-10, Is it because of my higher voltage? Thanks for the program, cant wait for the next release!
Furiøùs: When I was originally writing the Test Sensors feature I was using a 45nm E8400 at about 4050 MHz and not too much core voltage. With additional overclocking and core voltage you will get bigger changes in temperature during this test. My Q6600 goes up about 10 degrees per core during this test at 3305 MHz and 1.40 volts. At the more typical 3600 MHz and 1.50 volts or so, it's going to be jumping more than that.
I think a difference between cores in a dual core of about 3 is pretty normal but I'm not sure what causes that. In theory, if both cores are given the same work load you would expect that both cores would heat up an equal amount but that isn't always the case. It could have something to do with how you applied the thermal paste or how Intel soldered the IHS to the cores or how you screwed down your heat sink or if both surfaces are perfectly flat and square to each other or who knows. The test was basically designed to find sensors that are getting stuck which is too common in the first batch of E8x00 processors. Uneven idle temps and zero movement in this test is a very good sign of a stuck sensor that won't provide accurate idle temps.
fgw: Thanks for that post. The first C in eax shows that bit 30 of MSR 0xEE is set in your E4300. It's also set in my E6400 and E2160 which both have a proven TjMax=85C. More evidence that the E4300 probably doesn't have a TjMax=100C but I still need to test one to confirm this.
emoners: Version 2.25 is beta at the moment and I'm working on individual idle calibration for each core which seems to be a necessity for many Quad core processors. In the pic I posted above you will see that even though the data coming directly from the DTS sensors ( Distance to TjMax ) is inconsistent, the reported core temps are much closer. I'll update the first picture in this thread when it is ready to be downloaded.
ken: Have you ever tried running your B3 at low MHz and low core voltage like 1600 MHz and 1.10 volts to see how your reported core temperatures compare to your room temperature? If someone with a B3 could do that it might help prove what TjMax really is for the B3. Either that or maybe I could trade my G0 straight across for a B3. Any takers? :D
For comparison, my G0 Q6600 at 3305 MHz and 1.416 volts idles at a calibrated 30C in a closed case with a Tuniq Tower on low speed with a room temperature of about 18C.
I dont have anything to measure my room temp and I dont think my mobo (P5B-D) can go as low as 1.1v but I will give a try later today and report back :) Rite now my X3210 (375x8 with 1.384v, cooled by TRUE) idle at 27/27/24/24.Quote:
Originally Posted by unclewebb
p/s: any chance you could change the color(or size) of the core temperatures to something different than the rest so its easier/quicker to see?
Well, on the other hand mine on idle are, ummm, crazy is the word?Quote:
Originally Posted by lowtcell
This is how it looks after hours on idle, CPU is water cooled, I've changed the goop from Arctic Silver 5 to Arctic Ceramique because I couldn't believe the deltas... alas results remain consistent and unchanged. Have a good laugh, I'm not kidding ;-) Ambient is 21 C.
Here's my XEON Quad-core X3210 G0
Assuming Tjmax is the same as the Q6600 and the cores idle 6*C above ambient w/ my cooling (TRU120E & 115CFM Panaflo fan, MX-2)...It looks like cores0/1 are good at (0) calibration, and cores2/3 need (+).
The good old P5B-D is my test bed. You can set core voltage in the bios to 1.10 volts and that gets me about 1.08 volts at idle. A large font like TAT uses for core temps is an excellent idea. I'll keep that in mind.Quote:
Originally Posted by ken
On my Quad, the sensors for core0 and core1 look pretty decent but I'm not too sure about core2 and core3. Maybe dual cores with good sensors go in one side of a Quad and the other side gets a dual core with the not so good sensors.
At full Prime load (small FFTs) on all 4 cores, core0 and core1 both report 62C while core2 and core3 both report 57C or 5C less. Most of the dual cores I've seen are very accurate at 60C. If the sensors are accurate for all 4 cores at full load then the only thing I can think of is that core2 and core3 are not receiving the same core voltage as core0 and core1 and this is causing them to run cooler at full load. How about some Quad screen shots with Prime small FFTs running on all 4 cores? Do core2 and core3 always run cooler?
jason4207: Your processor is definitely a candidate for individual idle core calibration. What temps do you get at full load?
chrisZ: Thanks for that screen shot. That makes TjMax=95C look very believable for the 45nm Quad processors. The next RealTemp should make your 4 idle temps a lot more comparable. How about some Prime load temps now?
Thanks for clearing that up Unclewebb, Realtemp rocks. told a few of my buddies about it too, they agree also. Im pushing alot of vCore at 4.1GHz, so yea. Probably why its higher. thanks man
At full load the 2 dies are different by 5*C, so they are tracking together pretty well. Maybe the 2 dies have different TjMax's? :shrug:Quote:
Originally Posted by unclewebb
I would say it's a mounting issue or die-IHS contact, but the die in question is actually colder...you'd think it'd be hotter if it was getting bad contact.
What do you think?