Real Temp - New temp program for Intel Core processors

[unclewebb]

@uncle: the latest version is the 2.79.8 right?

That's correct. Intel didn't release enough information at the August IDF for any software developer to accurately translate DTS data into meaningful temperature numbers so I've decided to slow down development. Too busy overclocking!

Dua|ist: I've been reading about this on the E8500 forum but there's a problem with a lot of the data that's been gathered. Here's a link to the X-Bit article. It's easier to read in English!

http://www.xbitlabs.com/articles/cpu...duo-e8600.html

Intel has documented that their 45nm sensors used in the Atom chips have plus or minus 10C of error at 50C. Slope error implies that as a CPU cools down, the amount of error would increase even further. If review sites or individual users do nothing to try and compensate for this then their reported temperatures or conclusions are meaningless.

X-Bit has decided to ignore the "slope error" and error at TjMax that these sensors have.

Their power consumption numbers are interesting but not consistent. It doesn't make sense to me that at idle and full load at default MHz that power consumption is identical for C0 vs E0 but when overclocked there is a 9 watt difference. I would need to see some more test data with more processors before I'd conclude anything. I'd use Prime95 small FFTs for consistent power usage. I have a wattage meter on my computer and I know it can float around a watt or two.

If the 9 watt difference is 100% accurate then that might translate into a 1C difference at full load but probably not into the differences that some users are seeing. Most of the differences that I've seen posted are more related to sensor error than anything else.

If you have a similar C0 and E0 then why not run them at the same FSB frequency and use the same multiplier. Use RealTemp and calibrate them at low MHz / low voltage so they both report about 5C above room temperature. Set them back to fully overclocked at the same MHz and core voltage and then compare Prime95 small FFTs full load temperatures. That would be interesting and a little more scientific than the X-Bit testing.

[Dua|ist]

Those E0 And C0 that you compared did they have same VID?

Nope, but they were pretty close. The C0 8400 had VID = 1.225v while current E0 8500's VID is 1.25v

unclewebb: Thanx for the link. I see your point. But first of all we're talking about temperatures at load around 70C (whichever Tjmax they used, this is much above the 50C, so no "slope error" here, right?).
And then if we assume their measurements are correct, -9W actually seems not that small difference, and it resulted not in -1C but in +2C according to their testing, which is what confuses me most.

That C0 is not mine anymore but I might ask to play with it a little more. Problem is, this 8500 E0 has both sensors stuck at anything lower than 36C, so no clue how'd I calibrate it. But right above those 36C both sensors show same temperatures. The C0 8400 (calibrated) would heat up to ~70C under Linpack at 4.16GHz @ 1.400v (CPU-Z under load), the current E0 8500 heats up to ~73C at 4.3GHz @ 1.344v (CPU-Z again). I must admit I'm a little afraid pushing 1.400v through this one given the current temps, and thus can't set the E0 to same clocks & volts as C0.

[unclewebb]

But first of all we're talking about temperatures at load around 70C (whichever Tjmax they used, this is much above the 50C, so no "slope error" here, right?).

That's what I used to believe but that's not what Intel said at IDF. Their version is that there is some error at TjMax and slope error goes from one end of the temperature graph to the other which makes sense based on how temperature sensors work.

They called the amount of error at TjMax 'X' which is somewhat useless information but does confirm that there is some error and TjMax is not a fixed, written in stone number. Perhaps X means about plus or minus 5C. You might think that plus or minus 5C is way too large an assumption but that is actually a tighter tolerance than what Intel lists their 45nm Atom sensors at when tested at 90C.

If you put a handful of processors in a 70C oven and ignored the slope error for the moment, they could indicate anywhere from 65C to 75C based on the data coming from the on chip digital thermal sensors. If you take into account slope error then the temperature range might even be a little wider than that.

What I'm trying to say is that we're all putting too much faith in the quality of data coming from these sensors. The 45nm sensors are far worse than I ever imagined. How can you possibly compare temperatures when there is such a huge amount of error in these readings?

And then if we assume their measurements are correct, -9W actually seems not that small difference, and it resulted not in -1C but in +2C according to their testing, which is what confuses me most.

That just confirms that uncalibrated temperature readings are worthless. If power consumption goes down then the core temperature will also go down. If it doesn't, then either the core temperature reading is not accurate or the power consumption number is not accurate or maybe both.

I did some testing today while running Prime95 small FFTs and an 8 watt reduction in power consumption at the plug was resulting in a 2C drop in core temperatures. That's with a Tuniq Tower on high at 4000 MHz and a core voltage of 1.40 volts. I tried to keep everything as equal as possible and used the Clock Modulation feature to switch back and forth between a duty cycle of 87.5% and 100%. This reduced power consumption by about 8 watts while still maintaining the same CPU frequency and voltage while running Prime95 small FFTs. That might not be 100% scientific but it's probably pretty close. The CPU temperature was in the low 50C range while testing.

I've always found that if a Core CPU is running too hot then Prime95 small FFTs will fail or LinX will fail or your computer will re-boot. Too hot depends on how hard you are pushing things. The harder you overclock and over volt, the cooler you'll have to run to maintain stability. If you are overclocking then your temperatures will be nowhere near the thermal throttling or thermal shut down point so there's no need to worry about what temperature you're at. It's just a number and for most people that are unable to calibrate due to sticking sensors or simply don't bother, it's not even a very accurate number. Overclock as high as you want and let stability be your guide.

[burebista]

It's just a number and for most people that are unable to calibrate due to sticking sensors or simply don't bother, it's not even a very accurate number. Overclock as high as you want and let stability be your guide.

Word bro.

[davidk21770]

...
I did some testing today while running Prime95 small FFTs and an 8 watt reduction in power consumption at the plug was resulting in a 2C drop in core temperatures. That's with a Tuniq Tower on high at 4000 MHz and a core voltage of 1.40 volts. I tried to keep everything as equal as possible and used the Clock Modulation feature to switch back and forth between a duty cycle of 87.5% and 100%. This reduced power consumption by about 8 watts while still maintaining the same CPU frequency and voltage while running Prime95 small FFTs. That might not be 100% scientific but it's probably pretty close. The CPU temperature was in the low 50C range while testing.
...

Why doesn't someone with water cooling and a water coolant temperature reading, drop vcore and clock as low as possible and then run tests with heated water to see how the sensors compare to the water temp. If everything's as low as possible, there shouldn't be much difference between the water temp and the cpu temp.

[WoodButcher]

david k, been there, done that. http://www.xtremesystems.org/forums/...25#post3165025
rge and others did the same, you have to read the novella!

[unclewebb]

davidk21770: The RealTemp calibration method is similar to your idea. My theory is to calibrate by running a Dual Core at 6x333 for 45nm or 6x266 for 65nm and drop the core voltage down to about 1.10 volts. If you are using an after market heatsink and turn your fan up to high and open your case, then your reported core temperature should be about 5C above your room temperature near your computer. If you are water cooled then you might get a degree better than that and if you have a quad then you can add a degree or two to that number. When water cooled you would compare to your water temperature instead of your air temperature.

This is not an exact science but if your sensors are not sticking then your reported temperatures should be very close to those numbers. You need to make sure you don't have a lot of background processes running during this test so your CPU can idle down to its lowest temperature. Task Manager should only be showing 0% to 1% CPU Usage.

If you use the RealTemp calibration feature then I think your reported temperatures will be pretty close to your core temperature from idle to TjMax but there's no easy way to 100% accurately measure the core temperature and prove this one way or the other. This type of calibration will also help reduce the amount of error due to TjMax not being an exact number.

Now I just need to convince people that calibrating isn't an option when you have a 45nm CPU. When surfing from forum to forum, it's very rare to see anyone using this feature. With 65nm, if TjMax is correct, then you'll probably be pretty close even without a calibration but 45nm CPUs are all over the place and you need to do something before trying to compare temperatures of them.

[randomizer]

If you are using an after market heatsink and turn your fan up to high and open your case, then your reported core temperature should be about 5C above your room temperature near your computer.

What about in my case? I'm around 25C higher than ambient and that's on a 65nm chip. Maybe I should quit while I'm ahead.

EDIT: Ladies and gentlemen, we have hit the jackpot!

http://intel.wingateweb.com/taiwan08...PWS002_100.pdf

It looks like they simply extended the previous document.

One of the biggest surprises that I noticed as I went through is TjMax for a B2 E6x00 is only 70C! That will mean my temps are being reported 20C too high, which also means I am idling about 5C above ambient. Also of interest is that the Q6600 is reportedly 90C and not the 95C+ you had all been getting from testing. What is Intel trying to pull here?

[drminer]

randomizer,

The Intel link you provided no longer points to anything. Could you post the PDF somewhere? Thanks!

[rge]

In the IDF catalogue, that is about the only presentation that is now unavailable for download of the 80 or so listed . Probably too many people clicking on it...I hope someone saved it on their computer and can repost. I just woke up sure it was going to be there...apparently it was for a very short time.

@randomizer, hopefully you saved it, if not can you scrub your hard drive and pull it back up

[unclewebb]

Randomizer: In your post on the previous page it shows a minimum temperature of 26C on core0. If it usually doesn't go below 45C then that looks like you have some sticking sensors or a lot of stuff running in the background. Why not post 5 minutes of TaskManager CPU Usage beside your next RealTemp screen shot?

http://www.xtremesystems.org/forums/...postcount=2348

I noticed the other day on my E8400 that one sensor finally went below its sticking point when resuming from Stand By. That only lasted about 2 seconds though and now it's back to being stuck at the same point.

One of the biggest surprises that I noticed as I went through is TjMax for a B2 E6x00 is only 70C!

There are a lot of things I'm willing to believe in life but that isn't one of them. Hopefully the IDF document reappears somewhere.

Also of interest is that the Q6600 is reportedly 90C

That was my guess for the Q6600 - B3 so maybe it's a good document after all.

[rge]

it is back...

[Dua|ist]

Thanks, rge.
Looks like Intel ruined all theories once again by telling us what we wanted to know. Maybe it's really time to ignore the "accurate" sensors and keep pushing things till they start throttling?..

[burebista]

I uploaded it here.

[emoners]

Thanks, rge.
Looks like Intel ruined all theories once again by telling us what we wanted to know. Maybe it's really time to ignore the "accurate" sensors and keep pushing things till they start throttling?..

or at least some thing would smell like smoke

[rge]

LMAO.....intel had stated each cpu is individually calibrated so that throttling will not occur below tcasemax, meaning each tjmax is adjusted up or down ? 5 ?10C etc. and also sensors have error even at tjmax.

My IR gun gives exact same temp in past when I had access to very accurate surface probe temp (from wifes workplace) so it is accurate.

Pic 1 shows E6850 corrected tjmax (in red) and temps that would be displayed. So tcase is ~10C hotter than core temp....mmmmmm ok
Clearly using tjmax of 80 on my E6850 would give at least 10C too low temps.

Pic 2 shows E8400 corrected tjmax (red) and temps that would be displayed. So tcase is 5C cooler than core temp...plausible, but now no longer provable.

i7 supposedly has reasonably accurate sensors. But for 45nm and 65nm I am adding 2-4C (gradient at idle UC plus increased grad from no heatsink) to what I get with IR gun at DTS=0 and calling it a day...clearly using tjmax's provided by intel can give ridiculous errors from combined "individual calibration" and "sensor error".

[unclewebb]

The IR measured difference between my E6400 - B2 and my E8400 - C0 is 10C. According to Intel's new document, the difference is actually 30C.

Edit: Intel calls these values Target TJ which don't seem to have too much to do with actual TjMax.

After the August announcement I thought that for a lot of users, using the Intel Target TJ as TjMax might not necessarily result in more accurate core temperatures. After this announcement, I can state as a fact that using the Intel specified "Target TJ" as TjMax for your E6x00 B2 processor will result in reported temperatures that aren't even close to accurate. These new facts go against any real world testing that's been done by rge or myself and makes both of us wonder how accurate the August news release really was.

Maybe this means I need to run out now and buy a Core i7. Naah, they can keep them.

Edit: I re-read the document and they seem to be calling this number Target TJ. My opinion is that if users use Target TJ as TjMax then they will end up with some very unbelievable core temperatures. My E6400 is a good example:



In my 11C basement, it would be reported as running at minus 6C. Not that bad for an air cooled processor! I definitely recommend not using Target TJ as your TjMax value. Using Target TJ as TjMax for the 45nm processors might be questionable as well. Some 45nm processors will likely be reported a couple of degrees on the high side but if you do a RealTemp calibration then even if TjMax is off by a couple of degrees, your reported temperatures from 25C to 65C will still be reasonably accurate. This new info isn't going to help us out any.

Page 13 of the new Intel IDF bible says:
"The values listed for TJ Target are not specifications."

[mikaelc]

Hello there,

It looks like they didn't disclose anything. Just a point of reference for further research. A target Tj (desired?) which is individually calibrated for each unit. You can stick about that point, but merely can set it as a fixed value in RealTemp or any other monitoring software...

And... I don't see any information about PDC 2000 series, so I'm still stuck with my L2 stepping. Seems I'll have definitely downclock and undervolt, put the HS off and measure the Tcase with a multimeter...

And as for multimeters, guys... I don't have an IR gun and I will have to use my multimeter for temperature measurements. But I am afraid of doing that. AFAIK the temperature probes have something in common with conductivity. Do you think it's safe to place such a probe on an IHS?

[rge]

Suckers buying TECs and chillers...results with just PA 120.3 w/3 nexus fans at 1000 rpms. ambients 26-27C.

pic1
E6850 3.9ghz 1.475 vcore bios, 1.44 load vcore,
full orthos load at 40C...only 10C above intake ambients even with 8800 ultra in the loop...ignore speedfan cpu temp of 57C...clearly an error

pic2
E6850 4.3ghz 1.6 vcore idling at -7C in 26C ambients.

[rge]

Hello there,

It looks like they didn't disclose anything. Just a point of reference for further research. A target Tj (desired?) which is individually calibrated for each unit. You can stick about that point, but merely can set it as a fixed value in RealTemp or any other monitoring software...

And... I don't see any information about PDC 2000 series, so I'm still stuck with my L2 stepping. Seems I'll have definitely downclock and undervolt, put the HS off and measure the Tcase with a multimeter...

And as for multimeters, guys... I don't have an IR gun and I will have to use my multimeter for temperature measurements. But I am afraid of doing that. AFAIK the temperature probes have something in common with conductivity. Do you think it's safe to place such a probe on an IHS?

It is fine to touch a multimeter probe on IHS...but you will not get accurate temps that way. I bought one, paid extra for calibration, only to find that out. There are surface temp probes for few thousand...little expensive for the task...but thermocouples are going to read way low as most of it is measuring temp of air when touching to a surface. Which is why intel embeds them in the IHS, and even then have to calibrate them.

[Mumak]

I'm really confused now what values to use for the DTS reading offset.
Tj Target probably not so why do they publish these values? Is Tj,max different from that Tj Target (never seen such a naming in documents)?

Moreover:
- the inaccuracy of DTS readings raises at lower temps
- certain CPUs report invalid DTS temperature below 50 C
- DTS was never meant to be used for temp reporting - just to catch the hot point (value -> 0)

Is it really usable, or better forget all DTS on CPUs prior to Nehalem ??

[unclewebb]

The first bunch of Core i7 is just more of the same.

"The PROCHOT# activation temperature is calibrated on
a part-by-part basis and normal factory variation may result in
the actual activation temperature being higher than the value
listed in the register."

Software will be able to read some sort of Target TJ value from a register but that's not a TjMax value so there's no way to convert that to an accurate core temperature. These new sensors will also have slope error, less than the current 45nm processors I assume, but it's still there.

"Future processors may report temperatures in °C"

Looks like the enthusiast community will have to wait for the next generation after Core i7 before they'll ever know something simple like what temperature their CPU is running at.

[randomizer]

So then, if the Target Tj is not Tj Max, then what the heck is it? Damn you Intel!

EDIT: Remember this? http://www.overclockerspulse.com/new...tjmax-for-all/
Looks like Mr. Inkley was being a bit misleading.

[unclewebb]

randomizer: Why not get your friends over at Tom's to download this guide:

Thermal and Mechanical Design Guidelines
http://download.intel.com/design/pro...nex/317804.pdf

Tell them to head to Appendix D so they can learn how to carve up the IHS of an old CPU that they must have in a drawer somewhere. Then they can run a thermal couple to the geometric center and at least get a Tcase reading with the heatsink attached. It would be a very interesting article to see how that temperature compares to and tracks core temperatures from idle to full load while using a variety of different programs and different loads.

In the good old days Tom's Hardware was always doing cutting edge stuff like that to get to the bottom of issues. With over 100,000 downloads of the last version of RealTemp, it's obvious that users are interested in this subject. This new Target TJ spec needs to be investigated further and the Core 2 Duo B2 Stepping seems like the best choice for that.

[randomizer]

Yea they must have a ton of ancient hardware lying around. Perhaps I should donate my E6600 and they can buy me a replacement It would be interesting to see if they can get it to run cooler than I can too.

EDIT: If you look at the presentation from August, it says:

"Approximate temperature in °C can be calculated by:
Tj = Tj_target - DTS value"

In the October presentation, this no longer appears on the slides. Also, the slide that contains the 45nm desktop Tj Targets used to just say Tj; although the new slide still just has Tj in the slide heading.

August


October


Intel seems to interchange their terms way too much. I really don't understand what this presentation was trying to prove. First Intel say Tj Target is Tj Max, then they are saying it isn't a specification at all and the values they gave are enough evidence to say that it can't be Tj Max. I wonder if Intel really know what any of these terms are. They obviously forgot that there is no (publicly released) B2 Q6600.