Real Temp - New temp program for Intel Core processors

[randomizer]

Well 80C would make my E6600 look more normal, but per the thread that The Coolest posted the numbers in, people are still getting sub-ambient temps and rge's readings are still higher than the DTS is reporting. Bah!

[GAR]

is this accurate for Core i7???

[burebista]

I hope I'll have an answer tonight.

LE: Hehe, it looks like my friend brings me tonight earlier today.
This is how it looks on i7 920

[Monstru]

Hope it helps

[burebista]

Yeah man, thanks a lot for your work.

[Slay0r]

Hello again, Unclewebb.

I've just downloaded the 2.84 beta and at first glance I could tell the default TJmax on Q6600 (or at least, on G0's) has been put back up to 100°C from 95°C. In essence, was this intentional and due to some of your recent findings? Or a mistake?

Nice sensor test, by the way. As far as I can tell no stuck sensors over here.

[rge]

Unclewebb, awesome sensor test

That will make it easy to spot stuck ones...also will be great for temp testing at different temps/loads...if I get urge again.

My E8600 actually has accurate sensors.

[burebista]

WOW rge, perfect sensors.
Those are mine on E8400



Pretty decent for a 45 nm CPU.

Anyway, thanks uncle for this new and unique feature for testing Intel sensors.
Awesome job.

[unclewebb]

Slay0r: The original versions of RealTemp used TjMax=95C for the Q6600 G0. Based on rge's testing, which I fully agree with, 95C is too conservative. The actual core temperature is probably about 5C higher than the temperature I was measuring on top of the IHS with the IR thermometer. I originally didn't take into account that not having a heatsink installed changes the pressure between the cores and the IHS which changes the heat transfer. I also didn't take into account that the IHS heat spreader does an excellent job of spreading heat away from the cores. I think rge proved that point very well when his E7200 went nuclear.

It makes sense that the actual core temperature is going to have to be slightly warmer than the IHS temperature that I measured.

is this accurate for Core i7???

In theory, RealTemp should be reasonably accurate for Core i7. These new CPUs have TjMax information stored within them which RealTemp is able to read so that can't be argued about any more.

If you look at the screen shot above it's obvious that Core i7 still has slope error where the sensors move at slightly different rates compared to the actual core temperature. It's easy enough to correct for slope error with RealTemp.

rge showed us how to calibrate a Core 2 Duo here:
http://www.xtremesystems.org/forums/...postcount=2429

I believe a Core i7 Quad can turn off 3 of its 4 cores at idle and power consumption is about half as much at idle as a previous 45nm Dual Core. When calibrating a new Core i7, the gradient between your reported idle temps and your air or water temp is probably going to be a couple of degrees less than what rge found when testing his Dual Core. If someone wants to buy him a board and a Core i7 to do some testing with, I'm sure he won't complain.

rge and I also agree that Intel's explanation of slope error in their recent presentations is a little misleading. We've both found that the original RealTemp graph in the documentation is a much closer approximation to what these sensors actually do.

In order to come up with a more accurate calibration formula, I need a few people with some seriously messed up 45nm sensors to post a screen shot of the new Sensor Test. Two sensors on the same die moving at vastly different rates from idle to full load is what I'd like to see.

The Core i7 sensors are looking very similar to the previous 65nm temperature sensors which were a lot better than many of the Core 2 Duo 45nm sensors. My best guess is that even without any calibration, Core i7 reported temperatures from 70C to 100C are likely within a degree or two of the actual temperature.

[Slay0r]

Uncle, thanks for the explanation. I'll have to take your word for it since I can't do measurement myself, though what you say seems reasonable.

The sensors test kinda reminded me my pc needed a serious dust removal,

Q66 @ 420x8 1.280V, EIST/C1E on

before


after

[i43]

Thanks Unclewebb

just run a new RT2.84 sensor test on XP pro 32bit
Qx9650 /@ (stock) 3.0Ghz / everithing on (auto) Rampage Extrem (Bios v0601)
air cooled / Noctua NH-C12P / Open Air (table stand) |
No idle calibration in RealTemp - TjMax (Auto) - 95C for 45nm qx9650
C1E , SpeedStep -Disabled
CPU TM Func - Enabled / all other by intel spec. / Ambient 23C

seems ok .. shows that same difference between dies ~6-8C
what have read from my earlier tests, but this method seems much more comfortable to use .. no need to mess with bios and different clocks speeds or what so ever ! and run several tests etc ... nice work , well done

here my test screen:

[unclewebb]

Your welcome i43. Do you know what I like about this new test? One quick look at that screen shot tells me exactly what I need to know.

Core2 is stuck at 74 and Core3 is stuck at 68. Core0 and core1 are very well balanced like they should be and have a full range of motion from idle to TjMax. ie. no sticking issues.

If you followed rge's calibration procedure, you could get some very accurate core temperatures out of those two sensors. Have you tried running at low MHz and voltage to see how close your reported temps are to your air or water temp?

Intel may have said that TJMax=95C for your CPU but they said a lot of things about TJMax for 65nm that wasn't accurate at all. Even their recent correction doesn't appear to be accurate so I'm starting to question everything they've said about TJMax recently since the August IDF news release.

Edit: Looking at your ambient, I'd definitely be using TJMax=100C for your CPU regardless of Intel's fancy presentations.

[i43]

Thanks Unclewebb
here my low voltage run all same HW as my earlier post
except (adjustements followed by ' mr. rge ' instructions )
voltage set 1.1000v in bios (rather it show /jump/ 1.085 ~1.096 in everest /cpu-z but hope no deal with this )
C1E , SpeedStep - Enabled ( though ? i do see only C1E available in bios - i don't see SpeedStep (may be due to the lowest multi already set manual ? )
CPU TM Func - Enabled
Qx9650 /@ 2.0Ghz(6x333) / all other (auto) Rampage Extreme (Bios v0601)
Ambient - 23 C
Air cooled / Noctua NH-C12P / Open Air (table stand) |
(as mobo lay (flat) on the table, there shouldn't be any additional heat coming from VGA card or what so ever .. what may happen on normal desktop box/chassis)

still No manual idle calibration in RealTemp - leaved it for now .. TjMax (Auto) - 95C for 45nm qx9650
any way - these sensors never go below these values (25-25-21-27) since i first fired up this Yorkie
sure i will do some '' ghetto'' open window cooling '' later on a.s.a.p, just need some exact timing when left ''home alone '' and weather gets colder

about idle calibration from ' mr. rge ' post (for dual core)
he recommends to add current ambient /( for High end air (1fan).........................7C ) so consider quad at my scenario - then add more few degrees .. let say 9C .. OK i will tray later ... so do you think this is correct if i will calibrate al cores 34-34-34-34 C ?
I will try to calibrate for test purposes .. will see..

well, personally i am not much worried about these values/numbers
or have get used to live with them as they are..

--

not calibrated low volt screen shot attached
i shrink the attched screen .. but added also an external img. hosting links.


http://img521.imageshack.us/my.php?i...25244lchx8.png

[martymonster]

Here is my Q6600 G0 with a FSB of 375, room temp 26 Celcius

[unclewebb]

i43: If your ambient temperature is 23C then 7C to 9C above that would be 30C to 32C. Core 2 and core 3 are definitely stuck so you can't calibrate those sensors. The RealTemp calibration feature is designed to help with sensors that move at different rates. It can't do anything with sensors that don't move at all at idle.

If TJMax is actually 100C, that would boost your idle temps 5C from 25C to 30C which looks a lot more believable to me. There's not a lot of data from QX9650 owners so without testing with an IR thermometer or similar, it's impossible to say what TjMax really is.

Intel originally said the early E6x00 CPUs were TJMax=70C. When the Coolest and I complained, they bumped it up 10C and said they must have made a mistake. Now TJMax=80C for my E6400 but that's still 10C too low.

Intel documentation clearly shows how to carve up an IHS and properly mount a thermocouple to the center of the IHS. I know there has to be a lot more accurate information available than what they've decided to release to the enthusiast community so far. We need more documents from their engineering department, not their PR and fluff department.

[randomizer]

Ambient is probably in the mid 20's, but I don't have a thermometer on me to check.



At least you can see in plain view how stupidly hot this CPU runs. And to think I once tried 1.75V Note that C1E was enabled, EIST was disabled and the CPU speed was 2.7GHz at load. Maximum 0.8C spread between the sensors.

[rge]

There is no question unclewebbs E6400 tjmax is 90C based on 85C IHS at DTS=0. Randomizer, I wonder if your cpu runs that hot, or if it is proof that intel has used different tjmaxes. Would be interesting to IR that one. If your tjmax is lower, that would explain intel's 65nm tjmax releases (the one before unclewebb and coolest complained and the one after) and perhaps explain why some quads have significantly different temps even at high range.

[unclewebb]

The randomizer E6600 is definitely a curiosity. It seems to either have a different TJMax or someone didn't do a great job attaching the IHS to the cores. Have you tried using the OEM heatsink on it recently? If you ever get bored with it let me know and I'd probably buy it from you if the price was reasonable. I think rge and I could get the truth out of that CPU.

You better send it my way first though. rge seems to have a bad habit of venting the tops of his CPUs with a drill.

[kpo6969]

E8400 CO @ 3.6ghz

[randomizer]

Randomizer, I wonder if your cpu runs that hot, or if it is proof that intel has used different tjmaxes. Would be interesting to IR that one.

Intel said they kept the same Tj Max for the whole stepping. Given the backflip on the presentation and the inconsistencies with your testing, I think we'll run out of salt soon. A grain of sugar to sweeten the moment anyone?

Have you tried using the OEM heatsink on it recently?

I've been quite busy lately and haven't had a chance. I will do so shortly though. I'm not looking forward to ripping my rig apart just to change the heatsink but there's not much that can be done, unless the OEM heatsink can be used with a backplate. I haven't used it for so long I can't remember if it is even supposed to use a backplate.

If you ever get bored with it let me know and I'd probably buy it from you if the price was reasonable. I think rge and I could get the truth out of that CPU.

It started off being a pain, now it's actually interesting that it runs this hot. I only hope it isn't just a plain old heatsink mounting problem. If it turns out that the heatsink isn't the problem, I'll consider this. It's a shame you're from Canada, the 1AUD = 0.66USD exchange rate would be nice

[rge]

unclewebb, are you implying that a 20% cpu survival rate for temp testing is low?

randomizer, if intel did not change tjmaxes within same stepping, then that leaves us with +10C calibration (same effect as 10C higher tjmax) or intel still has not given us correct tjmax. What is interesting is intels statement, that G0 stepping tjmax was increased 10C, correlates exactly with our testing, just that our testing shows 90/100 versus intel claims (now) 80/90.

The q6600 B stepping 90 tjmax correlates exactly with Unclewebb's testing, but q6600 G0 90 tjmax is 10C low with exact same testing. Also, intel is going against their own claim of 10C increase in G0 tjmax on this quad, and given unclewebbs test illustrating the 10C increase...I think they just goofed again on this one.

And then unclewebbs M0 E6x00 which had same effective tjmax as his ?b/l stepping E6x00 of 90C. That would imply that his M0 stepping correlated with intel and was not calibrated higher, but his other one was calibrated 10C higher, or with n of 1 that may have been an intel error...need to test a few to be sure. And my E6850 was clearly 98-100C, ie again the 10C higher "calibration".

I wonder if intel would be willing to supply the approximate tjmax calibration per stepping or give a range. Because as is, without IR results for guidance, those intel numbers are not very useful given the sometimes there sometimes not 10C "calibration."

[unclewebb]

Thanks kpo6969. The more data the better. Each one of these tests tells a story.

I found the Core i7 920 test on the previous page very interesting. Core 0/core 1 mirror each other and core 2 / core 3 mirror each other. This makes it look more like two separate Dual Cores similar to the previous generation and not 4 unique cores.

Between 12.5% and 75.0% the difference between sets of cores is very consistent at ~2.7C. This doesn't fit the previous generation's slope error model. All 4 cores are moving on the same slope, they're just offset by a couple of degrees. Is that a slight difference in TJMax due to calibration error at the factory or is it just normal behavior with one side of the CPU running slightly cooler? Many 65nm Core 2 Duo based Quads looked similar to this.

I hope I see a few more posts with examples of Core 2 Duo slope error. rge and I noticed during testing that by 30 degrees to TJMax, both cores of a Dual Core generally lined up and there wasn't any significant difference from 70C to TJMax. That's different than the graphs that Intel presented at IDF. Their model implies that slope error goes from idle to TJMax.

The other thing I noticed was a huge drop in temperature when Prime 95 was stopped compared to what a Core 2 Duo does. It looks like most of the cores have been turned off which I think is what happens at idle with Core i7.

[randomizer]

I found the Core i7 920 test on the previous page very interesting. Core 0/core 1 mirror each other and core 2 / core 3 mirror each other. This makes it look more like two separate Dual Cores similar to the previous generation and not 4 unique cores.

It could be that the IHS pressure isn't quite uniform, more screenshots of Core i7 should give an indication on how common this is. We've all been fooled, Intel has released another cheesburger

[unclewebb]

I thought it might be a sign when 1 or 2 cores are dead that they'll start warehousing these chips and then sell them in 3 to 6 months as a special Core i7 Dual Core model. It's always been good business to sell everything that comes down the assembly line. Make my Core i7 a Dual. I hate heat and fan noise.

[randomizer]

Here's with the stock heatsink. PROCHOT in action people



A bit of extra variation there. I'll run the test again tomorrow just to make sure nothing was skewing it.