Separate names with a comma.
Discussion in 'Overclocking & Cooling' started by TheRapture, Mar 31, 2007.
Ahhhhh, you mean, the way we do things in AMERICA !
Unclewebb, you cannot use the IR gun with surfaces that are very shiny and reflect alot of light. I too did the same thing with my ninja that I ran it fanless, the heatsink got VERY hot but temp readings were like 27c from the aluminium metal plate that is very reflective. So you have to use tape for instance. Apply a piece of tape on top of the copper and then try again. My 27c whent to over 60 doing so. That is why I've used the IR gun using the hole in the middle of ninja heatsink to shoot at the very small little heatsink in the bottom that isn't reflective at all.
I'll do some more fanless testing soon..
SuperKeijo: Thanks for the info. Some tape sounds like a great idea. I'll put some on a different IC to make sure there is good heat transfer and that the IR gun has no problems reading it. I decided to take a break and learn how to use this new tool to produce some repeatable results.
Translation: The wife's home tonight and I didn't want to get caught with my new overpriced toy!
If your 100C TjMax theory proves true for my revision B2 E6400, I expect there will be one or two people looking to argue my results so I'm trying to cover all angles.
I did a quick fanless test.
As can be seen here neither the case fan nor cpu fan was spinning at all. Also note the piece of tape on the aluminium plate.
This was the reading from the plate itself:
It was clearly warmer so reading is false.
From the piece of tape I applied:
Much closer to the truth.
From the middle of heatsink:
These were with the cpu running at 3GHz, 1.3625v at idle.
So even here where the heat is transfere rather slowly from core to air, the coretemp 0.94 reading is still lower than heatsink (not lower than ambient anymore, which was around 25c)
I used some thin masking tape on the shiny copper surface where I was trying to obtain temperatures from yesterday. This worked great. The shiny surface was now dull and the measured temperatures were very consistent on this piece without any of the temperature jumping around problems I experienced yesterday.
I was able to get my core voltage down a little bit more. The Asus bios complains that there is a problem with the core voltage setting but it still lets you boot up fine at this level.
No one can argue with this consistent core temperature reading of a steady 21C over 13 minutes. I've never seen SpeedFan draw a graph with a resolution of +/- 0.2C.
The temperature outside of the open case is at 22C so a core temperature of 21C is possible due to rounding and equipment errors but it already looks suspicious.
With the computer and cpu fan running, I'm able to quickly remove the heatsink fan to get it out of my way and replace it with an 80mm Thermaltake that churns at 7000 rpm and delivers plenty of airflow. By holding this fan with my left hand near the Intel heatsink it was no problem maintaining a core temperature reading of 21C.
Temperature readings from the IR thermometer are instantaneous. It was no problem at all to briefly lower the fan, take a reading and then raise the fan back up to maintain the core temperature. The IR reading off of the copper cup in was never less than 27C anywhere within this cup. See picture on previous page for a visual.
It is impossible for the core temperature to be 6C less than the heatsink touching it.
The great theory that a C2D at this low MHz and voltage doesn't really put out a lot of heat is also dead wrong. Without any fan on the heatsink, the C2D puts out enough heat to increase the heatsink from 22C to 62C. That's a lot of heat energy even at idle.
I can't confirm that TjMax is 100C but this does confirm that it is definitely not 85C for my revision B2 E6400. All mobile Core based products, Solo, Duo, Duo 2, on the Intel website have a documented TjMax = 100C. At the moment I'm going to assume TjMax = 100C for my processor as well.
Hmm... so this means that they are running much hotter than people originally thought?
If your readings where taken with Tjunction 85 then yeah.
So this means that instead of using CoreTemp .94 like I am now.....which reports 85c TjMax....I should use .95?
I am not worried about temps too much, my rig is ultra solid...but I want to get correct readings if possible.
Thanks Unclewebb for those tests! It most certainly looks like the tjunction value is 100c for desktop c2ds aswell.
Maybe...the verdict is not in yet for sure...if this were true and all C2D's had a 100c TjMax, then my cpu would be running at 75c to 77c under ORTHOS....which seems WAY too hot for a well cooled cpu at 3.2ghz and 1.45v....at even stock volts and 2800mhz, I am showing 72c and 73c.....imagine what a stock cpu in an oem box would run at!!!!!!
We need to test more....and maybe SOMEONE at Intel, SOMEWHERE, has a spec listed....
That was my thought as well, which is why I asked.... hopefully we will get an actual answer someday
Time to add some more confusion to this thread.
With the heatsink fan off, at low MHz and low volts, the copper heatsink which is directly touching the 2 cores gradually heats up to an IR reported 62C. CoreTemp when using a TjMax of 85C also reports exactly 62C.
During this test the processor is basically idle and its temperature and the heatsink attached to it are given plenty of time to completely stabilize. Logically you would think that if you heated up a piece of metal to a set temperature that if another piece of metal was snugly attached to it, that it would also heat up to that same temperature which seems to be happening in this case.
At hotter temperatures, TjMax = 85C looks correct but at temperatures near 20C, it seems to be out by at least 6C.
The only explanation I can think of is the digital thermal sensor which is calibrated for high temperatures is not very accurate at lower temperatures. This is possible. Some sensors are calibrated and very accurate but only through a limited temperature range. The purpose of the DTS is to accurately and reliably throttle and shut down a C2D when it gets hot but there's no need for it to be accurate at lower temperatures and it might not be.
This would certainly explain why Intel has remained very quiet and hasn't released a way to convert DTS readings to an absolute core temperature. It might be impossible to accurately do this through the entire temperature range that a C2D can operate at. The only important thing that Intel is interested in is when the DTS nears zero so thermal throttling and shut down can take place.
Just some random thoughts. Could you try this test again SuperKeijo? I'm not sure if you want to use more core voltage or more MHz to get your core temp up to 60C but if you don't want to get up that high without a fan, I certainly understand.
I'm open to any thoughts or any suggestions for further testing.
Edit: I shut my computer completely off for a couple of hours today. When I got back home I checked for temperatures within the copper heatsink cone and the temperatures were reported consistently and exactly equal to ambient. There appears to be no problem with the accuracy of the data that has been gathered.
That last theory really shakes things up a bit when it comes to determining temperatures.
For the time being, should we just go back to using the single temp sensor that the mobo and its utilities report from for comparing temps? I know it means that we don't have a scale to compare to Intel's specs using those, it should give us a way to compare temps to each other. I mean, we used to have to rely on that in the old days, right?
I don't think you are thinking clearly hear. Between the cores and the spot you monitor the heatsink temp there is:
-about 3cm of solid copper!
Now I don't think you have a clear perspective on just how much heat you lose on this journey. From the IHS to copper you lose maybe 3c (compared to if they were soldered together), and the 3cm of copper loses maybe ~4-5c. And from actually inside the core to the outside of the IHS There's another few degrees loss. And the tape loses maybe 1-2c too. Thermal conductivity would have to be no less than INFINITE THE WHOLE WAY FROM CORES TO TOP OF COPPER. Needless to say it is far from that. It is purely a coincidence that the 2 values are the same here. In reality we lose a total of ~15c which sounds realistic in my book. Look at my values, there was only ~2-3c difference in the temp I recorded from bottom of heatsink compared to what coretemp 0.94 reported (coretemp being 3c lower!). It is just plain coincidence that you get the same reading on that and something to NOT draw any conclusions on.
Remove the heatsink, remove the IHS, then point the IR thermometer on the CORE ITSELF and tell me does coretemp 0.94 report the same as the thermometer.
Unclewebb, and if you look at the temps I was getting from the tape on top of the heatsink, It's only 1-2c different than what coretemp 0.94 reports, very close to what you're getting. 1-2c lower than you (compared to coretemp reading) because the ninja is such a tall son of a bitch so it loses a little bit more heat from where I was monitoring, but then again loses less heat from the bottom of the heatsink, because there isn't a big chunk of copper but just a small aluminium heatsink.
There is a plugin for MBM 5 now that reads direct from the Allendales temp sensor, it jives exactly with Core temp .94.
pic showing both in action
I still can't believe people are arguing about this, little common sense would suggest that since it is a core 2 duo chip, just minus the 2 meg of L2 cache that temps would be similar to regular core 2's.
It still makes the assumption that tjunction is 85c. This proves absolutely nothing. And unless you hadn't noticed we arent arguing on wether or not the tjuntion value is different on c2ds but wether it's 85 or 100.
You aren't reading what they are saying.. They are saying that the programs might be reading the values incorrectly by assuming the max is 85c when it could be 100c, thus our temps are off by 15c.
Sounds a little extreme. I wish I could say that it hasn't crossed my mind but......I think I'll leave the IHS on when I get to this step.
This kind of insanity goes back to my Pentium III days. I couldn't believe the temperatures back then either so I pulled off the heatsink and fan and used my thumb as my trusty temperature monitoring device. Let's just say that experiment didn't last too long. I think I saw over 100C in the bios before I pulled the plug. Surprisingly that P3 is still running today in my niece's computer.
Some of the crazy things I do are just plain crazy but some of the crazy things that I say are designed to get some feedback. So you think it's possible that the hottest part of the core could be 15C hotter than the copper heatsink touching it. I did not think that there would be this large a difference after both the core and the heatsink had stabilized in temperature.
In previous testing with one core running TAT at full load and the second core completely idle, the final temperature difference between the two cores was only about 2C. After switching the load from Core0 to Core1 and setting Core0 to idle, the temperature difference quickly swapped and the difference was once again about 2C. The two cores are not physically touching. The only thing touching both cores is the IHS and the heatsink on top so from that I came to the conclusion that the heat transfer was pretty efficient.
Thanks for your input. Before I dismember my C2D, I plan to try some more testing like above but at higher temperatures like 70C and 80C to see if any more interesting coincidences happen.
Decided to read up on Intel's thermal specs a bit, and according to this document, http://www.intel.com/design/processor/datashts/313278.htm The actual limit for tjunction on a 65 watt processor should be 61.4C, (intell calls it the Case Temp refered to as Tc) so where ever they are getting this 85 or 100C thing is likely the actual shutdown temp whereas Tc is probably the throttling temp. It doesn't sepecify the actual shut down temp in the spec, even states that it is not revealed through software, just the fact that it sends out a signal that shuts it off.
So it is very likely that we have all been running our proc's a bit hot and that they may actually be throttling a bit lol.
Something else of interest is the fact that show in this document, http://developer.intel.com/design/processor/designex/313685.htm for soldering a thermocouple directly to the processor case, that the soldering iron/plate should not exceed 155C (yeesh) or there may be damage to the proc. Also shows specs for actually machining out the groove to place the thermocouple in. Interesting stuff.
So basicly were still at a loss to explain the actual readings that were currntly seeing in .94 ot .95, probably going to take a programmer to figure it out for sure.
It has been suggested that the core temps could possibly be 15C higher than the Tcase. If this is true, we are talking about a huge jump in temps even within the chip itself, in which case it wouldn't be surprising to have the same jump from that much copper.
On the same line of thinking, if the temp that my mobo, mobo utility, or SpeedFan's TempX/CPU can be somehow 15C lower than a core temp as reported by .95, then very small amounts of material can make a large difference in temps. If you want to argue that .95 must be giving the wrong reading and that I should be looking at .94, that doesn't work out either: with C1E and EIST on and my E4300's FSB at stock, I am at 1.15V/1200 MHz and .94 shows 24 & 28 (SpeedFan's Temp1 is 28C, ambient is 25C).
No, actually 75c does not sound way too hot at all. 60c sounds way too cold for 2 cores running at over 3ghz! Thats a 50% overclock on your cpu! At those clock speeds and voltages the cpu is actually producing about 110-120w of heat which is about as much as a geforce 8800 GTS. The geforce 8800 reaches temps of over 80c easily with its big cooler (although relatively small rpm fan). It sounds alot only because you are used to comparing it to others' temps using a program that assumes the tjunction to be 85. I think Intel is in no hurry to tell us the truth about the tjunction because they don't mind people thinking their cpus run 15c cooler than they actually do!
Have you ever tried running the cpu at stock speeds and, say, 1.1v core voltage. You'll be amazed at how low the load temps are now. (even after adding 15c to what coretemp reports.)
Hey everyone. I have an e4300 batch Q645 and I am trying to see where my TJunction is supposed to be set. CPU-Z shows mine as revision 2.
I am having the same issue as everyone else in this thread: high temps w/ Core Temp 0.95 (idle 48 C) and low temps w/ Core Temp 0.94 (idle 33 C) at stock speeds. My ambient/room temp is approximately 25 C. I am using an Artic Cooler Pro 7 and Artic Silver 5. I have reset the processor twice and have received the same temps both times. The processor appears to be flat and so does the HSF.
My trouble is that I am trying to do the cold temp test that TheRapture recommend but I am having issue droping the vcore below stock settings of 1.32v.
I have a DFI Infinity P965-S. The bios allows me to raise vcore but it does not allow me to lower below the stock 1.32v.
1- Any ideas on how to lower stock voltage with my board or is it not possible?
2- Is my processor running too hot for stock settings?
Thoroughly confused on how hot this sucker is running
There definitely aren't too many, if any, C2D processors doing much throttling.
Based on a TjMax of 85C, you can run a C2D at a core temperature of 82C without any throttling. If the TjMax of my processor is actually 100C, which it more and more seems to be, then the Core 2 Duo, every Core 2 Duo, won't start throttling until the core temperature hits an absolute value of 98C.
The problem isn't that we're running these things too hot. The problem is that most users aren't running them hot enough because of fear of these big numbers.
The 61C number in the Intel specs is not a core temperature as reported by the DTS and can't be compared to the core temperature.
I am a programmer! That's not the problem. The problem is a lack of information from Intel and conflicting information from users on forums. The problem is that presently available temperature monitoring programs may not have been thoroughly researched. Intel TAT is a good example. Users accept the temperatures it reports based on the logo even though this program was never designed for the C2D desktop processors nor has TAT been updated since their release.
I'm very lucky to have an early revision B2 E6400 which everyone has accepted to have a TjMax = 85C without any further testing. It's looking more and more like it really is TjMax = 100C for my processor and all of the other C2D processors. The 85C TjMax number might turn out to be some sort of urban legend many years from now.
Coming from the P4 era, the big marketing sell feature was how cool these new C2Ds run and they do but changing the location of the temperature probe makes them appear to be running much hotter. It's also not just one probe like previously but Intel has strategically located multiple probes within each core so the DTS is always returning a number which represents the maximum temperature of any point on the core. It doesn't matter what type of app you're running or what part of the core you're stressing. If any area of the processor gets too hot then the C2D will first throttle and then ultimately shut itself down. You never have to worry about overheating a C2D. It won't let you.
Intel is certainly in no hurry to have a long line of uneducated consumers trying to RMA their processors based on a big absolute temperature number. I don't see it being in their best interest to clear things up. They have nothing to gain.
Back to testing.
quezwood: If your processor becomes unstable and CoreTemp 0.95 is reporting over 90C then it's time to get worried. Until then, enjoy your new toy and don't be afraid to overclock the hell out of it.
Would it be a risk to boot the cpu without any heatsink what so ever at 600mhz and 0.675v? I'd like to see what my thermometer says pointing directly at the ihs. Would there be any chances of me getting to windows? Or will it just overheat like mad and shut itself down in seconds?
SuperKeijo: This next step has crossed my mind once or twice.
I think the only way to do it is to put your computer into Stand By mode, pull off the heatsink, and then have it resume and come out of Stand By directly to the Desktop. If I do something this crazy I will have my handy 7000 rpm Thermaltake fan in one hand to keep the C2D somewhat cool with the other hand very close to the power button. If I can keep the cores at a safe temperature then I'd go for the gun and try to get some readings off that naked core!
Before you go crazy, I've got some more interesting results to post in a minute or two.
The TAT program uses a process called MeromMaxPowerVer0p3.exe when you click on the Start button beside Workload Level. This process is in charge of creating the extreme heat readings that you see when running TAT.
I decided to run one of these processes on one core and absolutely nothing on the other core to get an idea of the temperature difference between the two cores. Using Task Manager you can go in and change the Affinity of this running process which immediately shifts it from running on Core0 to running on Core1. Here are the results:
When full load is on Core0 it is approximately 7C hotter than Core1 and when you swap the load to Core1 then Core1 is about 5C hotter. The average difference is 6C.
This helps disprove one theory I had. Even though these two cores are joined with the IHS and have a big copper heatsink transferring heat between the two cores there can still be a temperature difference of 6C between the hottest point of one core compared to the hottest point of the other core. If you compared the hottest point of one core to the coolest point of the second core the temperature difference would be even greater.
In my previous test my IR thermometer was reporting a temperature of 62C for the copper heatsink while CoreTemp was also reporting this same temperature. It is quite possible that the hottest part of the cores could have been 15C higher than the temperature of the copper heatsink where I was measuring.
The other thing I noticed when switching loads between cores is that the reported core temperature switched immediately and within the 1 second interval that I had CoreTemp logging at.
Looks like TjMax=100C is viable for low temps as well as high temps and once again 85C just doesn't seem possible.
Sure seems that way, yes. I think I'll try the heatsinkless thing at some point I just have to get me some AS5 to replace the crap I have applied now and I'll be all set.
Hi, I had been googling about temperature utilities for my C2D since I was in difficulty about the various results which various utilities gave, and google led me to this thread.
I read the whole thread and wish to thank you for the explanations given here!
I've got an E4300 which I overclocked to 3.0Ghz using 333FSB, multiplier x9, 1.3375 vcore, DDR2-833 (from 800) and vdimm at 1.90. CPU cooler is Zalman 9500.
My ambient temperature is 25 deg.
Coretemp 0.95 reads both cores at idle at 35 degrees.
TAT reads, at idle, one core at 38 and the second core at 36.
Both speedfan and Coretemp 0.94 read my idle core temperatures at 20 degrees, which is not possible since that'd be 5 degrees below ambient.
Hence I think I can conclude that the Tjunction for my E4300 is 100 degrees not 85, and that in my case Coretemp 0.95 and TAT are giving the most accurate temperatures.
TAT always seems to give a slightly higher temperature.
Running TAT at 100% load gives me temperatures which alternate between 63 and 65 degrees.
Running Orthos small FFT's give me a rock stable temperature of 60 degrees for both cores.
Btw, should I be worrying that my temperatures are too high? I've heard of people having higher load temperatures, but then intel says the E4300 should not go above 61.4 degrees! I've read that you're saying people seem to be afraid of 'high numbers' but I've been on other forums and there they try to keep load temperatures in the low/mid-50's (on air) which is practically impossible for me, (unless these other people have been misinterpreting their temperatures...... lol)
60ºC at full load with tjunction 100 seems fine.
At games you'll probably get lower temps.
You might want to buy some extra FAN for your case...
I just got my abit quadgt and E6400 together last night.
It seems clear some mobos have conflicts with some software progs for monitoring.
For my vcore. Speedfan shows 1.57. Coretemp shows 1.325 and CPU-Z shows 1.213. Well, I have it set at 1.325 in bios, it has to be right. But how is someone less experienced supposed to tell which is accurate? : /
Right now at load, at stock speed, with stock heatsink on E6400, temps are about 60-65C. Does this seem accurate for stock heatsink? My tjunction is 85 according to coretemp 0.95
Also, what is this tjunction. I keep seeing peoples screenshots of tjunction of 100 in .95 coretemp. Mine has a tjunction of 85 for my E6400 B2.
My particular E6400 reports 100c in .05 Coretemp.....
So what the hell does that mean...... Why is my E6400 B2 stepping show as 85 tjunction in coretemp.
Well really... the way to test would be to have someone who had one that reported @ 85 and one that reported @ 100 to test both in their system with the same settings. Then we would finally know if some had different max temps or not.
I have a revision L2 chip....coretemp .94 shows 85c tjunction and .95 shows it as 100c tjunction.
A better name for what CoreTemp calls Tjunction would be TjMax. It refers to the maximum temperature that your processor will operate at. A degree or two beyond TjMax and it should shut down pretty quickly to save the processor from self destructing.
Speedfan is usually correct on most boards for core voltage. The voltage that CoreTemp shows only refers to the default voltage that your processor should run at. When voltage goes beyond approximately 1.40 volts, CPUz screws up and reports 1.213 volts which is a meaningless number.
Avoid setting core voltage to AUTO or default. When overclocking you can end up with a lot more voltage than you realize.
At the moment I'd recommend using CoreTemp 0.95 and in the Options menu set it to:
Show delta to Tjunction Temperature.
This reads the digital thermal sensor ( DTS ) in whatever core 2 duo you are using and accurately displays it without any further manipulation or guessing at what your TjMax really is. TjMax is not documented by Intel for desktop processors so at the moment an absolute temperature for your processor is just a guess.
The DTS is exact. It tells you exactly how many degrees you are away from the Maximum temperature that your processor will operate at.
I plan to do some more testing but it'll probably be a couple of more days before I can experiment some more with my IR thermometer, hopefully directly pointed at the core for a true core temperature reading.
I think the picture says it all. I finally put aside common sense and removed the heatsink so I could get a more accurate core temperature reading with my Fluke 62 IR thermometer.
I posted a nice long story over on the XS forum which is definitely interesting.
To summarize my findings, the last number reported by my IR thermometer before this processor started to throttle big time was 84.4C. Between 60C and 85C the temperatures reported by the IR thermometer were basically identical to CoreTemp based on a TjMax of 85C.
Is it possible that the cores were actually 15C hotter than the IHS that I was pointing my thermometer at? Possible I guess if the IHS wasn't touching the cores but the IHS is touching the cores. The conclusion I came to was that my TjMax is indeed 85C after all.
If this is true then the other conclusion I have to come to is that the DTS is NOT accurate when reporting idle temperatures. It was designed to trigger the Thermal Throttle and is calibrated for that purpose so if it isn't accurate at lower temperatures then that makes sense to me.
Does my low temp testing prove anything. Absolutely nothing if the DTS is not accurate at low idle temp levels.
All my testing confirms for me is that the DTS can not be used to display accurate absolute core temperatures throughout the temperature range that a Core 2 Duo can operate at. That would explain why Intel has never released any documentation to convert DTS values to an absolute temperature. It's simply not possible.
Read my post on the XS forum and previous posts on this forum and let me know if you have a different interpretation of the results I've found.
I know you totally disagreed with this previous post SuperKeijo but it now seems accurate to me.
You are a crazy man web =D
Thank you for testing this for everyone, guess 85 is right though the temps we are receiving may still be wrong.
For my processor a TjMax of 85C or very close to it seems likely but it doesn't really solve anything for everyone else. Without an IR thermometer to back it up, I wouldn't fully trust any software based temperature monitoring program.
Hopefully SuperKeijo can run the same test on his processor so we can compare numbers. With his ability to run his processor at a lower voltage he should have no problems controlling the temps of his C2D even without a heatsink attached. You do need some sort of auxillary hand held fan though.
I just wish I had one more hand today to get some better pictures and keep things from getting too out of control!
Unclewebb, nice stuff! I guess I have to get me some arctic silver first and then I can do the same test. Now I have nothing to replace the old thermal paste with so I guess I'll order some and do this test when I get the AS. I'll be back!