Real Temp - New temp program for Intel Core processors

[iddqd]

is there a version that works with Phenom IIs?

[unclewebb]

is there a version that works with Phenom IIs?

Unfortunately, no. Core Temp supports AMD.

It's been full time work just trying to keep up with Intel, especially when they start doing asinine things like recycle old model numbers for their new processors, not once but twice.

There was enough confusion with two steppings of the E6300 which were built using two totally different cores. Now they've decided to re-use that part number one more time for a CPU built with 45nm technology that runs at 2.80 GHz compared to the original 65nm E6300 CPUs that ran at 1.86 GHz.

http://processorfinder.intel.com/Lis...earchKey=e6300

[unclewebb]

http://www.fileden.com/files/2008/3/...mpBeta_330.zip

-incorrect multiplier being reported when running CPUID HW Monitor was fixed.

-Disable Turbo Mode option added for Core i7

-added the ability to toggle C1E and added EIST reporting.

-updated the RivaTuner plugin. The individual Load Meters require the main load meter to be active as well.

-added MultiBoost=1 INI option for any Core i7 board that is stuck at a 12X multiplier when using Windows 7. (DFI / MSI)

-added AverageMulti=1 INI option for Core 2 based CPUs which uses the same timers as Core i7 uses to accurately calculate the average multiplier over 1 second intervals.

-got rid of the "Not Responding" message during the XS benchmark.

-initial Core i5 support.

[stasio]

Good job Uncle

[drnip]

You ever thought about CPU voltage readout on Real Temp? Would be nice.

[stasio]

You ever thought about CPU voltage readout on Real Temp? Would be nice.

Then,will be no more "RealTemp"

[-X-hellfire]

I´m rebuilding my computer, see sig, had my motherboard in for RMA but they
didn´t find anything but my SATA RAID 5 is still gone, for now I got a
SATA 1TB harddrive in IDE-mode and testing everything, don't even got any
nvidia forceware drivers installed running on an old 14" monitor @ 800x600
resolution (colors look ok) but seriously thinking of digging up the
old 21" Sony (colors is very grayish)



Did this test before I underclocked it as unclewebb described in post #3509
and Speedstep C1E idle @ 1.600 GHz and load @ 2.400Ghz,
test calibrated 0.0 0.0 4.5 2.5 (just some numbers dialed in)

idle @ 1.600 GHz and load @ 2.400Ghz, test calibrated 0.0 0.0 4.5 2.5
1b
1b-00-idle-C1E-2400-9x266-1V27-1x(000)-P95-MEM800-tcal





I underclocked it as unclewebb described in post #3509 and Speedstep C1E
Prime95 with 25% load on 1 core @ 1.600Ghz, test calibrated 0.0 0.0 4.5 2.5
2a
2a-25-load-C1E-1600-6x266-1V10-1x(025)-P95-MEM800-tcal


2b
2b-00-idle-C1E-1600-6x266-1V10-1x(000)-P95-MEM800-tcal





test calibrated 0.0 0.0 4.5 2.5 (just some numbers dialed in)
2c
2c-00-idle-C1E-1600-6x266-1V10-1x(000)-P95-MEM800-tcal-sett







1 x Prime95 with 25% load on core 0 @ 1.600Ghz
3 x Prime 2004 75% load on core 1, 2, 3 @ 1.600Ghz
PrintScreen at 99% "heatup load" as close as possible to 100%
Maximum cooling 5 fans at 100% and, topfan 50%
in: 2x120mm, door:80mm on gpu
cpu:120mm on TRUE, out:120mm, twin 2x80mm)
calibrated 0.0 0.0 0.0 0.0 (see 2c)

3a
3a-99-load-C1E-1600-6x266-1V10-4x(099)-P95-MEM800-0cal






0 x Prime95 with idle on core 0 @ 1.600Ghz
3 x Prime 2004 75% load on core 1, 2, 3 @ 1.600Ghz
PrintScreen at 31.4% "CPU load" and "test #9"
calibrated 0.0 0.0 0.0 0.0 (see 2c)
3b
3b-75-load-C1E-1600-6x266-1V10-3x(075)-P95-MEM800-0cal


3c
3c-00-idle-C1E-1600-6x266-1V10-0x(000)-P95-MEM800-0cal


(btw 1V10 is 1.10V, 3x(075) is 75% x P95 is Prime95, MEM800 is memory @ 800 Mhz, 0cal is not calibrated at 0.0 0.0 0.0 0.0 etc )


What are good values for the calibration (see 2c) ?

[unclewebb]

Hello -X-hellfire. How the hell have you been? I haven't seen you around here for a while.

Since writing RealTemp, I have come to the conclusion that TJMax is not consistent on all 4 cores of a Quad core processor. Intel quietly agreed with this at their IDF news conference last year when they showed a graph that showed there is some error in how they set TJMax at the factory. They were too embarrassed to get into specific details though because with 45nm Quads, the TJMax difference core to core can be huge.

Before adjusting your temps with calibration factors, it's a good idea to try to estimate the difference in TJMax. On my Q6600 G0, my best guess is TJMax = 100C, 100C, 105C, 105C for the four cores.

The best test for this is to run Prime95 Small FFTs and heat your CPU up as much as possible. If you can get the core temperature up over 65C that would be great. By this point, the 4 sensors have very little if any slope error and most of the difference observed is differences in TJMax. For your CPU I would set core 0 and core 1 to TJMax = 100C and then you will need to adjust core 2 and core 3 to a higher TJMax so all 4 report pretty much the same thing during this Small FFT high temperature stress test.

After that is done then you can go back to idle and adjust core 2 and core 3 with calibration factors to equal out the idle temps. Your core 3 might be OK after TJMax is adjusted. Core 2 tends to sag the most and it will likely need a calibration boost even after TJMax is adjusted.

A lot of users don't like getting their core temperature up high enough to be able to find out this kind of information so I haven't talked too much about it. Lots of users probably don't agree with my different TJMax values from one core to the next but it's the only reasonable explanation I could come up with. The final result for my Q6600 is 4 cores that track each other very closely from idle to TJMax.

I took my CPU up to its max while running P95 Small FFTs to keep the load as equal as possible on all 4 cores.



Intel's IDF version of the truth where they blamed most of the issues on the slope error of these sensors only tells part of the story. To me, this screen shot is a clear sign that TJMax is simply not consistent for all 4 cores. I've seen so many screen shots that I'd also be willing to say that it is likely deliberately set higher on many Quad CPUs for core 2 and usually core 3 as well.

The Core i7 series is also a good example of this. When running P95 Small FFTs there is almost always an exact 5C difference in temperature from core 0 to core 3 after things have stabilized when the core temperature is above 70C. I think the end core is deliberately set to TJMax=105C but I have no way to prove this.

If you're curious, do some overclocking, add in some core voltage and get your core temps higher into the range where slope error is not a factor. Don't use any calibration factors and use the default TJMax and post a screen shot so I can have a look.

[randomizer]

I wonder why they would set 2 cores higher, assuming they did. Perhaps it is to maintain decent performance when two cores begin to throttle. When two cores are throttling it will lower the heat output of the whole package so I guess it would help keep the remaining two cores in thermal "check" without lowering their processing power.

EDIT: It could be the layout of the die as well.

[unclewebb]

Hi randomizer. Nice to see another old timer drop by.

One test I did a long time ago was to run Prime 95 Small FFTs and then lock it by using Set Affinity... in the Task Manager so it would only run on core 0 and core 1 and then reverse it so the same load is only running on core 2 and core 3 with next to nothing on core 0 and core 1. With the shrink to 45nm, the difference between the two sets of cores during this kind of test is minimal. It shows that when you have 4 cores within a cm of each other, there isn't a hell of a lot of difference in the core temperature between any of them even when two cores are doing all the work and the other two are idle. When all 4 cores are working equally running Prime Small FFTs, the quick heat transfer between cores should have them all at pretty much the same temperature after Prime has been running for a while. There's just no reasonable explanation why they would be 10C different.

In testing, the reported temperatures are never equal which is one of the reasons I think TJMax is not consistent across the cores. If TJMax is not consistent then the "why" is an interesting thing to consider.

On the Q6600 it is possible to have the two Dual Cores running with two different multipliers. Maybe they offset the TJMax so if CPU 0 started to throttle, it would start to cycle down to a 6.0X multiplier, letting it cool off while CPU 1 would still be able to run at full speed using its 9.0X multiplier. This would cause a smaller drop in performance compared to all 4 cores dropping down to 6.0X at the exact same time. Your explanation / theory makes a lot of sense randomizer.

We've seen that these CPUs don't go ka-boom at 90C or 100C or 110C or even higher so setting TJMax slightly higher on CPU 1 wouldn't cause any issues and should smooth the transition if there ever is any thermal throttling.

It would have been nice if last year's IDF presentations included some more meaty engineering documents instead of Core i7 is wonderful documents prepared by the PR and fluff department.

We will never know the whole truth about these sensors so I'm open to any reasonable explanation that can cover all the data that's been collected.

Edit: Kind of funny. I was just helping another user on another forum understand his CPU today with this ~10C difference between his two Dual Cores in his Q9650.
http://forums.guru3d.com/showpost.ph...9&postcount=18

[-X-hellfire]

Hi unclewebb

yeah it's bee a while and I get just been using my Q6600 @ 3.2Hz a moderate overclock, worked great until distaster struck and my RAID 5 died, not going into more of that (link), anyway I still got my T-Balancer setup as you can see but since the RMA I bought new fans, which are much better.

I going to test the Q6600 with your suggestions of a 60-65C overclock but found out after yesterdays underclock marathon with all fans on full throttle that one of the fans is noisy, the new 120mm case out fan, is very annoying even though I switched Attenuator (a PWM filter) on the T-Balancer, still the same "whistle" sound. Have bee trying to fix it but it wont go away, had the same problem with the old fan, but did some customized curves for the fan which worked pretty good but takes forever to get very good.

Thought the new fan would fix it but its the same, this way "off-topic", but my serious RealTemp calibration of the Q6600 depends on having a working cooling system, that includes keeping the noise pollution under control. I used my NoiseGard Sennheizer headphones yesterday which reduces whistles and fan noise very good, however I right now want to eliminate the culprit to the problem. The Q6600 have a TRUE and the same fanbrand and it has no whistle, an external fan (same brand) for the firewall is also silent.

What I wanted to do was to save a ghost with all T-Balancer profiles before I started the serious overclocking, a good plan and backups are no1 before hell breaks out.


Will start the backups soon as I'm as curios as you are of what the result of the RealTemp calibration will be.

(sorry for the "Off-topic" post but I´m usually end up there what ever I do :innocent: )

[unclewebb]

Your Q6600 seems to be following a very familiar pattern. Core 0 and core 1 are very similar while core 3 sags a little and core 2 sags a lot. When you start doing some testing at higher temperatures I'm sure this is going to make a lot more sense. The testing with acebmxer on the Guru 3D forum went very well today. No two CPUs are identical but it's surprising how many of them follow a very familiar pattern. It goes way beyond this just being random chance. The 65nm Quads don't have the sensor sticking issues at normal temperatures the way the 45nm Quads do so your Q6600 should be easy to figure out.

[randomizer]

The only exception to the norm is, of course, my E6600 It seems to have a Tj Max closer to 72C than 80C (Tj Target) or 90C (what Real Temp uses I think) from memory. It actually made Intel's first IDF presentation look ok!

EDIT: Forgot to say nice to see you drop by Comp

Hi randomizer. Nice to see another old timer drop by.

Hey. I got your PM about the other tool you've been working on, but I don't have i7 so I can't say much about it

In testing, the reported temperatures are never equal which is one of the reasons I think TJMax is not consistent across the cores. If TJMax is not consistent then the "why" is an interesting thing to consider.

Perhaps Benson will enlighten us as to whether or not Intel deliberately set Tj Max higher on 2 cores, and if yes, then why. I'll have to see if I still have his email address. It might be another "trade secret" but I can't see what kind of problems would be caused by telling us. Then again, look how long it took them to tell us fake Tj Max figures.

[KURTZ]

just a Q guys, is there a program like RT for the AMD's processors?

[Russ_64]

It is called CoreTemp - look in AMD thread .......

[unclewebb]

http://www.fileden.com/files/2008/3/...507/RTCore.zip

I did a few updates to the RealTemp / RivaTuner plugin today for any RivaTuner fans. By default, it uses the Calculated Multiplier for Core 2 and Core i7 now.

I also added Battery Level monitoring support for mobile users to the plugin.

If you have a Core i7, enable hyper threading and draw a screen full of graphs so I can see that it works.

randomizer: The i7 Turbo program works on Core 2 as well. You can use it to learn about your multiplier. I was hoping you were ready to retire your E6600 by now so I could get my hands on it for some testing. Send it my way and I'll send you back an upgraded chip.

[randomizer]

randomizer: The i7 Turbo program works on Core 2 as well. You can use it to learn about your multiplier. I was hoping you were ready to retire your E6600 by now so I could get my hands on it for some testing. Send it my way and I'll send you back an upgraded chip.

I thought you were kind of joking about that earlier (that plus I forgot you ever said it after all this time ). What will you send in return? I might need a new board and I don't want to be out of service for too long so I'd rather have the new board already if needed Let's hope the chip doesn't start running normally for you just so that it doesn't get itself drilled.

[unclewebb]

I have a Q6600 G0 and an E8400 C0 in stock. Take your pick. Neither one is a golden chip. The E8400 has run a 10.3 second Super PI time when it was on the edge of self destruction at 4.5 GHz. Definitely not Prime stable but lots of fun all the same. It's max Prime stable is down around 4000 MHz to 4050 MHz. I run my Q6600 more conservatively at 3 GHz and I have never pushed it to extreme levels. My old motherboard with the 965 chipset isn't great with Quads so I didn't think there was much point. Either chip is good enough for me so take your pick. Your E6600 is kind of like acquiring a rare fossil. It might provide us with some new information and be the missing link in the evolution of TJMax.

Send me a PM and we can work something out.

Edit: I lied. It actually ran 10.250 in a 1M of SuperPI.

[CompuTronix]

Reposted from the bottom of the previous page.

Hi unclewebb and rge,

Sorry I haven't posted or PM'd for quite some time, however, I've continued to follow your excellent work, as well as every post on every page, every day.

Your points are well taken concerning Intel's deviations in Tjunction Max values, however, from a different perspective, I'd like to share with you my research and empirical data, which I've continued to acquire for my Core i7 and Core 2 Temperature Guide over at Tom's.

During the past 2+ years, I have extensively tested, and recently re-tested and analyzed processor temperatures using three data points which consist of idle, 50% load and 100% load at stock settings with Prime95 Small FFT's. I then repeated each test at overclocked settings with higher Vcore and temperatures. The objective was to observe and carefully document the thermal relationships between CPU temperature and Core temperature, sensor linearity characteristics, slpoe error behaviors, and power consumption.

Calibrations were based upon the 5c thermal Gradient between the Analog Thermal Diode and the DTS value, of which we're both familiar from certain Intel engineering documents. All testing was conducted using a standardized test setup under controlled conditions at 22c ambient on a variety of motherboards, chipsets and CPU cooler combinations. Real Temp was used to cross-reference Core temperatures in SpeedFan.

Although the CPU temperature offset calibration value may shift on certain motherboard when BIOS test settings are changed, CPU temperature was verified or re-calibrated in SpeedFan prior to acquiring each measurement, so that information from the Analog Thermal Diode would provide a valid and accurate point of reference. This assured that the results were both consistent and repeatable.

The following assortment of 65nm an 45nm processor variants were tested:

(1) E2160 L2
(1) E4500 M0
(1) E5200 R0
(1) E6400 B2
(4) E6600 B2
(1) E6850 G0
(6) Q6600 G0
(1) Q6700 G0
(1) E7200 M0
(1) E8500 E0
(3) Q9650 E0
(6) i7 920 C0
(1) i7 940 C0

Of these 28 individual processors, 17 of which are quads, my findings show that the Analog Thermal Diode is typically linear from low idle temperatures thru very high load temperatures. Testing also revealed that the 5c Gradient between CPU temperature and "Mean" Core temperature is relatively constant from 4c to 6c, which conforms with the detailed thermocouple testing performed by rge.

The most interesting point is that when you originally conducted your IR testing, except for a few low order variables concealed from your Fluke IR gun on the IHS, apparently you were closer than you may have since thought. For example, the Core 2 processors which I've tested that are suposedly Tjunction Max 100c have shown values that average from 97c to 98c, with the exception of the i7's, which are nearly spot-on between 99c and 100c.

As such, it's likely that the differences often seen in the Core 2 quads between each pair of dual cores is 100c and 95c, which yields a Mean Tjunction Max value of 97.5c. This obviously coincides with my findings, and offers further insights toward Intel's new term "Tj Target".

Comp

[unclewebb]

Hi Comp. It's been a while. I thought you didn't love us anymore.

For the Core 2 CPUs I've tested, throttling seems to start when the digital thermal sensor is reporting 3 or sometimes 2 so if TJMax is 100 then it does seem to happen at 97C or 98C. I thought this was something Intel did deliberately so the CPU would slow down and keep the core temperature from ever reaching 100C. During my 3 hour Prime95 Small FFT run with the CPU fan turned off, only once did one core touch 100C so I thought Intel's design was working pretty damn good.

I saw a screen shot at X-Bit labs which showed a Core i7 at 99C and it hadn't tripped the thermal throttle bit yet so RealTemp was still showing OK in the Thermal Status area. As soon as that bit gets tripped, even once for a microsecond, RealTemp will display LOG and if thermal throttling is in progress then that changes to HOT.

Based on user feedback, I think the official difference in TJMax between the two Dual Cores within a Quad was increased when the 45nm Quads came out. My best guess is about 5C for 65nm and maybe as high as 10C for the 45nm Quads.

I gave up trying to buy one of every CPU that Intel makes. There's no way I can come up with a large enough sample or even consistent and accurate enough testing procedures to be 100% certain about any of this. Intel knows the truth about their calibration procedures but they weren't willing to tell the whole story at the IDF conferences last year and most users have lost interest in this topic. At least they were willing to admit that TJMax is not a fixed, written in stone value and that some variation does happen and they also admitted to the slope error of these sensors. That was less info than I hoped to get from IDF but realistically, more than I expected to hear Intel admit to.

They were right about one thing. Their Core i7 temperature sensors are much better.

[KURTZ]

guys, but RT recognize the i7 processors' VID?

[KURTZ]

It is called CoreTemp - look in AMD thread .......

yes, but this can't do a sensor test for the cpus ...

[unclewebb]

KURTZ: There is no documented way to read VID from a Core i7 CPU. Before RealTemp was updated to support Core i7, it used to show some random numbers for VID but it shouldn't show anything now. If it is still possible to read VID, Intel isn't saying anything about it.

If the programmer of Core Temp wants to do a multi level temperature step down test for AMD CPUs then I would be happy to show him what I do to create lower temperature steps while Prime 95 is running. It just uses the Intel On Demand Clock Modulation feature while the test is running. This reduces the heat output of any CPU in gradual steps and at the end of the test it kills the Prime95 test and lets it sit there at idle for the last step. That would be a useful feature for the AMD guys so I hope someone does this. I don't have any AMD hardware to develop on so it won't be me.

Edit: I did some more testing tonight and discovered why RealTemp and the plug-in can display very high load numbers at idle on mobile CPUs. The Load % being displayed is based on how hard the one active core is working. With 45nm mobile chips, one core can get turned off at idle while the other core will start running internally at half the normal FSB speed due to an Intel feature called, "Dynamic FSB Frequency Switching." The result at idle is a single core running at 800 MHz so the true load on that core is actually quite high. It's working like a dog to get all the work done that needs to be done while running at a fraction of its normal speed while its partner is having a sleep. The reported Load % looks bizarre and has absolutely nothing to do with what Task Manager reports for CPU Usage. There is some definite meaning behind a big idle load number when you are monitoring a mobile CPU.

Here's the latest RealTemp / RivaTuner plug-in:
http://www.fileden.com/files/2008/3/...507/RTCore.zip

[JohnZS]

Once again thank you for your continued efforts at RealTemp Unclewebb
I have a quick question, the % Load monitor is not working correctly on my QX9650, I am using version 3.30Beta. During Intel Burn Test and Prime 95 Large or Small FFT's CPU is reporting @ 89% load, yet under Windows Task Manager it is 100% Load.
Any ideas?
When my CPU is @ Stock the % Load Monitor in Real Temp works.
Also have you altered the TjMax rom 95 to 100? for the C0 stepping QX9650? (Quad Core Extreme)?
Thanks
John

[unclewebb]

Hi JohnZS:

There are two possible issues with the Load meter. If you are using the Clock Modulation feature of an Intel CPU then the RealTemp load meter will report less at full load than the Task Manager does. If that is happening, that would be a good thing.

If the RealTemp Load meter works fine when you are using the default multiplier for your CPU but screws up when you are using a different multiplier then that would be a bad thing. A user with an early QX processor was helping me with this issue a long, long time ago. If you want to be the new helper then I can send you some stuff to test for me. Volunteer testers with unique CPUs are always needed.

As for the endless question, "What is TJMax?", I can honestly say, "I don't know." Intel doesn't seem to know and neither do I. If you still have that wonky Quad that you used to have then I think TJMax varies by as much as 10C from core to core. I don't know if your cores are TJMax 95 to 105 or maybe TJMax 100 to 110 or some combination in between. Intel never released enough information to clarify this. With sticking sensor issues and sensor slope error, it can be very difficult to accurately prove this. A sad situation but I'm only the messenger.

I just had a look at my code and here's a note I wrote a long time ago:

"Intel says QX9650 TJ Target = 95C but it seems more like TJMax = 100C during testing so leave it as is"

The truth is that there probably doesn't exist a QX9650 that has all 4 cores with the exact same TJMax. Once I get the RivaTuner plugin updated, maybe we can look at your numbers one more time to try to come up with a good guess at TJMax.