Real Temp - New temp program for Intel Core processors

[unclewebb]

rge came up with a great post a while ago about how to calibrate RealTemp.
http://www.xtremesystems.org/forums/...postcount=2429

The numbers he came up with were using his E8400 and right away the question was, "What about a Quad?"

I did some testing today at the recommended calibration point of ~1600 MHz and 1.10 volts. My E8400 was consuming 133 watts as measured with a Kill A Watt meter.

I swapped in my Q6600 - G0, and at the same settings and temperature, the meter showed 137 watts. During a CPU Cool Down Test with a Tuniq Tower, I've been seeing a 3 watt change of power consumption at the wall equaling a 1C change in core temperature. For a 45nm Dual Core that would be 1.5 watts per core per degree C.

If a Q6600 is only consuming an extra 4 watts at this setting (137w vs 133w) that works out to only 1 watt per core extra.

What this means is that a Q6600 at ultra idle should have core temperatures no more than 1C higher than the numbers that rge came up with.

I also put in my 65nm E2160 which only has 1MB of cache and it was using 129 watts in this test. In theory it should idle approximately 1C cooler than an E8400.

That's the thing about the ultra idle test. It doesn't matter too much what CPU you have, your results are going to be very similar to what rge found during his test plus or minus a degree.

[randomizer]

If the Intel stock cooler is supposed to keep it around 10-11C above ambient, that would mean 31-34C (ambient is about low 20s but I'm not certain as my only thermometer was shattered). Perhaps this combined with the wide margin of variation at Tj Max could be the reason why my DTS reports high temps. It might be at the extreme end of the calibration range.

[unclewebb]

Kyle who runs [H]ard OCP had this to say today in a forum:

"Hopefully will have an i7 milled to Intel spec so I can put thermistor in between it and the HSF in the next week or two."

I've wanted to see this sort of testing for a long time now.

I think Kyle can't believe how hot Core i7 is reporting its temps so he wants to see some numbers.

I was hoping for a test like this on a Core 2 Duo but the Core i7 sensors are far from perfect so this should give us some new info about how these sensors perform.

[randomizer]

Unless he's buying it, he'll probably get a cherry-picked chip with near-perfect sensors

[KURTZ]

hi there like i promise here are my sensors ... what you think Uncle?

however it was weird cause @ default i had the core1 died ... it's normal?

[unclewebb]

If you're running 4 GHz with 1.072 volts then there's no need to be concerned about the sensors. Your CPU looks like a good one.

At your normal overclocked MHz and core voltage, these sensors will be fine for reporting reasonably accurate core temperatures. All of the 45nm sensors stick at some point. When you get lucky, they only stick outside of your normal temperature range. Not many sensors can read reliably when you are more than 69 away from TJMax. That might be the sticking point for Core 0.

Most of the tweaks to RealTemp lately have been for better Core i7 support but it never hurts to have the latest version:
http://www.fileden.com/files/2008/3/...alTempBeta.zip

[KURTZ]

yes Uncle, but the core1's sensor was really died @ default frequency (333mhz) with 0 movements ... BUT not when i launch PRIME cause @ FULL (100%) to 62,5% it lives, but not under this percentage ...

anyway, another question: in the previous post you can see that the second core has 7 movements (@4000 obviously) but when i tried to reboot the sensor has only 5 movements ... why this happens?

thx for your support

edit: this is my latest shot

[unclewebb]

In your previous screen shot, both sensors move from top to bottom. Usually the sticking point is consistent from day to day but it's possible that it can change.

The Cool Down Test should be consistent after a reboot if you are using the same settings and haven't changed your cooling any between runs. Post two screen shots so I can have a look. For maximum consistency you also need to make sure you're not using your computer while the test is running and that there is nothing significant running in the background changing your results. Run at default MHz with C1E / SpeedStep enabled and it might be easier to see your official sticking points.

[sakis_the_fraud]

@unclewebb

i changed my 8400 C0 to an 8400 E0 and after a fast run of the sensor movement test i noticed that the movement was stuck at 0!

C0

http://www.xtremesystems.org/forums/...95#post3429195

E0



bug of realtemp, or something is wrong with my proccessor?

[Warboy]

unclewebb, The X33xx series CPU is not auto detected correctly, They should have a TJmax of 95c.

[astaris]

Should you be seeing a 6C to 10C difference? Probably not. Is it normal? Unfortunately it does seem to be the norm.

Intel has talked about slope error but my Cool Down Test is leading me to believe that TJMax isn't nearly as fixed as we all once assumed. There could be as much as 10C of error at TJMax so two similar CPUs or similar cores could report vastly different temperatures. I'm still learning about this stuff so when you get a chance tomorrow post a screen shot of that test. It helps me see what the biggest contributor to the differences in temperatures is.

Thanks T_Flight. There are some nice charts about VID in the documentation but they don't include what register in the CPU you're supposed to read that value from or how you're supposed to interpret it. Anyone with a Core i7 should post a screen shot so I can see if I did good or bad.

Section 2.5 of Volume 1 of the Core i7 Datasheet tells me almost everything I need to know, almost.
http://download.intel.com/design/pro...hts/320834.pdf

About VID reading, i think that the VID_max should be at IA32_PERF_STATUS MSR (address 198H), bits 32-39. VID_max shoud be the VID of the processor at P0 (maximum performance level).
The actual voltage in mv should be:
V_mv=V_min+VID_max*6.25
V_min should be 0.5 V from the core i7 VID table, but in this case i think that the V_min is the minimum voltage obtained with VID7=0 (the same logic in fact applies to core duo mobile processors), so V_min shoud be 0.81875.
So in the end the correct formula shoud be:
V_mv=818.75+VID_max*6.25

[WaterFlex]

I can`t understand. What is TJmax for Xeon E3110?

[unclewebb]

sakis_the_fraud: Readings of 0 0 in the Sensor Movement Test is the first sign that your sensors may get stuck at lower temperatures and the CPU Cool Down Test seems to confirm that. From 62.5% down to Idle, both of your sensors are stuck at 56 away from TJMax. Here is what sensors that don't get stuck will look like during this test:

http://www.xtremesystems.org/forums/...postcount=2489

You can buy a Kill A Watt meter on EBay for about $15. Plug your computer into that and monitor its power consumption. During each step of the CPU Cool Down Test, you will see power consumption decrease. If the power your CPU is consuming goes down then its heat output and core temperature will also go down as long as your CPU fan speed, cooling capacity and room temperature are all kept constant.

Intel has said that when these sensors get 50 or more away from TJMax, they have a tendency to become saturated which is another way to say that they get stuck. E0 does not always equal better sensors.

Warboy: The X33xx series CPU is not auto detected correctly...

I went to Tom's and had a look at the TJ Target tables which is based on information Intel released at the last IDF. The first table is for 65nm Desktop CPUs and as far as I can tell, every value in that first table is wrong. By wrong I mean that using Intel's TJ Target values as TJMax will, for most users, give you less accurate core temperatures, not more accurate.

At IDF, Intel also showed that TJMax is not a fixed value. TJMax for any individual processor within a series can fall into a wide range of different temperatures. For 45nm CPUs the range might be plus or minus 5C. The range for 65nm sensors might be the same or it might be different. TJ Target for 45nm CPUs might be closer to the average TJMax for 45nm. TJ Target might represent the low end of TJMax for 65nm. Are you confused yet? Me too.

I've spent a lot of time during the last few months going over Intel's IDF news releases, reading between the lines, asking questions, doing some more testing and trying to understand what is really meant by TJ Target. My conclusion is that if you use the Intel released TJ Target values as TJMax, your reported core temperatures, in most cases, will be wrong.

That's just the first table I'm looking at. If using any value in that first table as TJMax results in less accurate core temperatures, then it's impossible for me to pick through the rest of this information trying to find out which values are correct and which ones aren't.

RealTemp, Core Temp and Everest let you use whatever TJMax you want so if you believe that the information released by Intel accurately represents TJMax then use it. I disagree with that and won't be changing TJMax that RealTemp uses based on information from Intel's IDF news releases.

astaris: IA32_PERF_STATUS MSR (address 198H) used to contain information about VID for Core 2 Duo processors but it does not seem to contain VID information for Core i7 CPUs. The one Core i7 that I saw didn't seem to have anything stored in the upper 32 bits of that register. The lower 32 bits have also changed compared to Core 2 Duo.

RT MSR Tool



I think the lowest 5 bits contain the CPU multiplier for Core i7.
0x17 = 23 decimal

That's from a Core i7 940 in Turbo mode which has a multi of 22 + 1 for the Turbo.

This value changes by 1 depending on load and if Turbo mode is being used but it doesn't seem to have any relationship to VID anymore. Core i7 owners can download my tool and test their own CPU. Post a screen shot if you see something other than your multi hiding in there.

Bits [63..16] of this register are listed as Reserved and are not publicly documented by Intel and the bottom 16 bits are not clearly documented by Intel either. The docs only say "Current performance State Value" but it doesn't explain how to interpret those bits. I have a plan B to try and find where VID info has moved to but I don't like my chances at the moment.

WaterFlex: I'd like to know what TJMax is for your E3110 too.
If I could find an E0 stepping at a good price I'd probably buy one and do some testing with it.

I think Intel's IDF presentation says TJ Target = 95C and RealTemp probably uses TJMax = 100C. Try running your CPU at 1600 to 2000 MHz with the core voltage set to approximately 1.10 volts. If your sensors aren't stuck then compare your reported temperatures to your room or water temperature with your case open and it might become obvious what your TJMax is or isn't. Post some screen shots and I'll help you out.

[msgclb]

astaris: IA32_PERF_STATUS MSR (address 198H) used to contain information about VID for Core 2 Duo processors but it does not seem to contain VID information for Core i7 CPUs. The one Core i7 that I saw didn't seem to have anything stored in the upper 32 bits of that register. The lower 32 bits have also changed compared to Core 2 Duo.

RT MSR Tool



I think the lowest 5 bits contain the CPU multiplier for Core i7.
0x17 = 23 decimal

That's from a Core i7 940 in Turbo mode which has a multi of 22 + 1 for the Turbo.

This value changes by 1 depending on load and if Turbo mode is being used but it doesn't seem to have any relationship to VID anymore. Core i7 owners can download my tool and test their own CPU. Post a screen shot if you see something other than your multi hiding in there.

Bits [63..16] of this register are listed as Reserved and are not publicly documented by Intel and the bottom 16 bits are not clearly documented by Intel either. The docs only say "Current performance State Value" but it doesn't explain how to interpret those bits. I have a plan B to try and find where VID info has moved to but I don't like my chances at the moment.

I just happen to be doing some memory tests when your MSR Tool popped up. Here's an 920 using x12 multi. I believe C is 12.

[unclewebb]

Thanks msgclb. 0x0C is equal to 12 decimal so MSR 0x198 does seem to contain the multi for Core i7 but not much of anything else that I can see.

With Core 2 Duo/Quad when you used CPU-Z and went into the cpuz.ini file and set

Sensor=0

it would report VID instead of your core voltage. That's the number zero and not the letter O. If it is set to 1 then it reads and displays your real time Core Voltage in that box.

Can you give that a try to see if CPU-Z can be used to report VID for Core i7 users? So far, Core Temp and RealTemp are in the dark about VID for Core i7.



With Core Duo based processors, VID and multis could be found in MSR 0x198



but Intel has obviously moved where the multi is located and eliminated the VID data in here. They are allowed to do stuff like this without issuing any documentation of the changes. By using the word Reserved throughout their documentation, that allows Intel to change the meaning of MSRs however they like, whenever they like. We call this, "a monopoly"

[elmysterio]

Intel says 90C, RealTemp uses 100C.

No, intel says 80C.

[msgclb]

I changed my multiplier to 14. As you see the MSR changed from C to E. Also I change CPU-Z option to Sensor=0. From the image it's clear that this doesn't work with a Core i7.



Here's my CPU-Z v1.48.5 register dump.
cpuz.txt

[astaris]

astaris: IA32_PERF_STATUS MSR (address 198H) used to contain information about VID for Core 2 Duo processors but it does not seem to contain VID information for Core i7 CPUs. The one Core i7 that I saw didn't seem to have anything stored in the upper 32 bits of that register. The lower 32 bits have also changed compared to Core 2 Duo.

RT MSR Tool



I think the lowest 5 bits contain the CPU multiplier for Core i7.
0x17 = 23 decimal

That's from a Core i7 940 in Turbo mode which has a multi of 22 + 1 for the Turbo.

This value changes by 1 depending on load and if Turbo mode is being used but it doesn't seem to have any relationship to VID anymore. Core i7 owners can download my tool and test their own CPU. Post a screen shot if you see something other than your multi hiding in there.

Bits [63..16] of this register are listed as Reserved and are not publicly documented by Intel and the bottom 16 bits are not clearly documented by Intel either. The docs only say "Current performance State Value" but it doesn't explain how to interpret those bits. I have a plan B to try and find where VID info has moved to but I don't like my chances at the moment.

Ok, i'm wondering if register IA32_PERF_Control (0X199) changed too, because in the old processors it stored the target vid+fid, so in performance 0 level it should store the default vid of the processor.

[astaris]

Ok, i'm wondering if register IA32_PERF_Control (0X199) changed too, because in the old processors it stored the target vid+fid, so in performance 0 level it should store the default vid of the processor.

nope, i asked a friend with 920 to read the MSR: it contains only the target multi

[unclewebb]

That's what I'm finding. There used to be several MSR locations where VID information used to hide but with Core i7 they're all gone. It's almost as if Intel doesn't want us knowing what VID our new chips are so users don't RMA them for some lame reason. I plan to come up with a small program in the near future that will try to automatically sniff out where VID info might be hiding now in Core i7. If not even CPU-Z knows where VID is hiding then we might have trouble figuring this out. Since TJMax can now be read from each Core i7 CPU, maybe Intel thought they needed to hide a new specification that enthusiasts like to know.

elmysterio: When word got out that Intel's original choices for TJMax for 65nm were a little out to lunch, Intel decided to bump a lot of them up by 10C. As far as I know, they only released their new and improved TJMax values to the Coolest (aka. Core Temp developer).

It goes beyond believable when they prepare a presentation with slides and PDF documents to finally release TJMax to the public after 2 years of hiding this information and no one in charge notices that most of the important TJMax values for 65nm are wrong. Even after the presentation, they don't correct anything until the complaints start to come in. For some reason, in my eyes, their credibility is about zilch.

Users will have to flip a coin and decide who they're going to believe. If you think Intel is giving you the straight goods then set TJMax in RealTemp or the program of your choice to that value. Be prepared for some real low ball temperatures.

When I got to line 1 in the first table and saw that TJMax is 70C for my E6400, I laughed a little, , and then decided to take the rest of this news release with a grain of salt. Users have been reporting sub ambient idle temperatures for the B2 series since they were first released and that's when most software was using TJMax = 85C. TJMax = 90C for my E6400 B2 is a lot closer to the truth than 70C, 80C or 85C.

[randomizer]

No, intel says 80C.

That's a screenshot of the tables on Tom's, which haven't been updated with the "correct" values. Since nobody believes the new ones either, there doesn't really seem to be a need to "fix" those tables up.

[i43]

Hi uncle and Thx for updates

i was just checking latest public? beta RT v2.88 - is it correct now it auto-detect TjMax 100C for 45nm. QX9650 CPU's?
is this 100C official now? .. just to be sure ?

[unclewebb]

For the QX9650, Intel says TJ Target = 95C. RealTemp will be using TJMax = 100C. I don't own a QX9650 and I've never tested a QX9650. My decision to set TJMax the same as the rest of the 45nm Dual and Quad Core processors is based on user testing and my general belief that the disclosure of TJ Target by Intel does not accurately represent TJMax for a lot of processors. Intel did not release any information about the range of error in TJMax so TJ Target numbers that TJMax is based on are meaningless. They say that TJ Target and TJMax should be about the same but for many 65nm processors, these two values are different by 10C.

I wish things were clearer. I wish Intel had released some actual engineering type documentation to back up their news release but they chose not to. Overall, there are too many inconsistencies in the documentation that's been released that I've decided not to trust any of it.

If you let your QX9650 idle at 2000 MHz and ~1.10 volts, I think your reported core temperatures are more accurate when TJMax is set to 100C.

[Demo]

Is the TjMax information some kind of a trade secret or something ?

Following that TjMax of 90 for the Q6000 series results me in getting lower than ambient temps in 1 of my cores so either 95~100c is the real Tj for the Q6600

edit:

heres my sensor test at 3.2Ghz(400*8) 1.335v at bios, 1.304 windows idle and 1.248~1.256v on 100% load.



sensor test at 1.2Ghz(200*6) 1.275v at bios, 1.240 windows idle and 1.224v on 100% load.

[burebista]

Guys, I really don't understand why you want to know/rely on that pesky TJMax? Any software reads DTS which shows distance to TJMax (whatever TJMax it is). Keep it >20-30 and forget TJMax value.