RealTemp General Discussion

[Krout]

I am lost here. I thought that the real value (temp of a core) shouldn't depend on what you set TJMax to. The temperature is read from the CPU internal sensor, right? So, how can I manipulate it?
I guess I don't understand what the TJMax is. What is it? More important: what is TJ Max value for Q6600?
Thank you

 

[burebista]

I thought that the real value (temp of a core) shouldn't depend on what you set TJMax to. The temperature is read from the CPU internal sensor, right?

Of course it depends of TJMax value. What is read from CPU internal sensor is distance to TJMax and that's the only one which matters for you CPU thermal behavior (throttle/thermal shutdown). Look in your screenshot, despite your TJMax settings distance to TJMax remains ~same across all your cores. Here I made an example myself.
Did you read uncle's documentation?

 

[unclewebb]

Intel did not include a thermometer within their CPUs so software can not read the core temperature directly.

TJMax is the maximum temperature that your CPU can get up to before it reaches the thermal throttling point. This will cause it to slow down if it gets that hot. For a Q6600, the thermal throttling point is typically 90C for the Q6600 - B2 stepping CPUs and 100C for the Q6600 - G0 processors. In the RealTemp settings window just click on the Defaults button and it will read your CPU and set this correctly. The Set TJ Max button is for advanced calibration purposes and most users don't need to adjust this.

The temperature sensors in these CPUs work backwards compared to a normal thermometer. As your CPU gets hotter this sensor counts down towards zero. When it reaches zero, your CPU has reached its maximum safe temperature and it is designed to automatically slow down to protect itself from damage. The direct reading from this sensor is displayed by RealTemp in the Distance to TJMax box. RealTemp then uses a simple formula to convert this number into a temperature reading.

CPU Temperature = TJMax - Distance to TJMax

If you have a Q6600 - G0 (check CPU-Z to find out) then the formula would be:

CPU Temperature = 100 - Distance to TJMax

The first core, Core 0, tends to be the most accurate. In your screen shot it shows that this number is 58. If you put that number in the above formula then your CPU Temperature should be reported as 100 - 58 which equals 42C. That sounds normal at idle.

Set TJMax back to Defaults and this program will report your temperatures just fine.

You can download the latest version here:

http://www.fileden.com/files/2008/3/3/1794507/RealTempBeta.zip

 

[Krout]

Thanks everybody.
It is getting clear now. So far I understand that the most important value is TJMax. I believe Intel refers to it as TCC activation point. Unfortunately they (Intel) don't disclose this value for any specific processor. So, this value is a guess...maybe educated, but still a guess.
I also understand that the core temp (in RealTemp or any other software of this kind) is as accurate as the TJMax value for the processor, and the most important value is the distance to TJMax. As long as I don't get too close to 0 for the distance to TJMax - I am fine, right?
I am doing this for my Q6600 but ultimately I am getting ready to OC i7, which I am getting very soon.

Another questions: does RealTemp interface with PECI?
What is Prime95?
Thanks.

 

[Mussels]

As long as I don't get too close to 0 for the distance to TJMax - I am fine, right?

Yes.

Although i prefer to leave ~20C til TJmax at load, to cover for hot days/extra heat from somewhere unanticipated.

Another questions: does RealTemp interface with PECI?
What is Prime95?
Thanks.

Dont know about PECI.

Prime95 is one of many CPU stress test programs - current favourite is OCCT on the linpack test

 

[Krout]

Thank you.
I guess the question about PECI is to the developer of the RealTemp..
I just downloded a Beta RealTemp and ran Load Tester. It shows 100% load but my Windows built-in CPU load meter displays only 26% load even after 10 minutes (sreenshot shows 3 minutes, but I am looking at it as I type this). How should I read this? Is it because it's Beta?
Thank you

 

[Mussels]

Thank you.
I guess the question about PECI is to the developer of the RealTemp..
I just downloded a Beta RealTemp and ran Load Tester. It shows 100% load but my Windows built-in CPU load meter displays only 26% load even after 10 minutes (sreenshot shows 3 minutes, but I am looking at it as I type this). How should I read this? Is it because it's Beta?
Thank you

well, it IS beta try OCCT as i mentioned above

 

[Krout]

I am running OCCT right now. Looks like a good program. However too much red (former Soviet - Perestroika) and fine print. Thanks for recommending it.

 

[unclewebb]

The Load Tester provides 100% load to 1 core. You have 4 cores so fully loading one core is equivalent to 25% load on your CPU. Add in some background tasks and an overall load reading of 27% sounds right to me. RealTemp agrees with that gadget on your Desktop so it must be OK. You can run multiple copies of Load Tester if you want to create more load.

RealTemp reads temperature data directly from the CPU registers. No PECI.

 

[Krout]

Thank you unclewebb, it does make sense now - one core.

 

[unclewebb]

Prime95
http://www.mersenne.org/freesoft/

This program can load up all 4 cores as it searches for Prime numbers. It's a very popular stress testing program.

Load Tester is the direct opposite of Prime95. LoadTester puts a very gentle load on your CPU and is good for testing turbo boost on the new Core i5 / i7 processors as well as EIST testing.

 

[Mussels]

Prime95
http://www.mersenne.org/freesoft/

This program can load up all 4 cores as it searches for Prime numbers. It's a very popular stress testing program.

starting to lose popularity now, as other programs (OCCT linpack, intel burn in test) get the same results, but faster.

your system may error 10 hours into prime95, but achieve the same result in OCCT linpack in 10 minutes.

another plus is that OCCT linpack is x64 - it tests all of the CPU, and all of the ram without leaving parts untested like the x86 ones do

 

[unclewebb]

Funny thing is I have the opposite problem at the moment. I'm testing some memory and on my board it can run hours of LinX x64 when testing 3GB of memory but Prime 95 x64 Blend fails in 30 seconds or less. It's nice when you are stable enough to pass both tests but my motherboard with the old 965 chipset doesn't seem to like these 2x2GB modules. They fail Prime at any MHz, any memory voltage, any timings, etc., etc.

I quit. I think I'll go get a P55-UD4P and start playing with something new. My P5B Deluxe has had a long life.

 

[Mussels]

Funny thing is I have the opposite problem at the moment. I'm testing some memory and on my board it can run hours of LinX x64 when testing 3GB of memory but Prime 95 x64 Blend fails in 30 seconds or less. It's nice when you are stable enough to pass both tests but my motherboard with the old 965 chipset doesn't seem to like these 2x2GB modules. They fail Prime at any MHz, any memory voltage, any timings, etc., etc.

I quit. I think I'll go get a P55-UD4P and start playing with something new. My P5B Deluxe has had a long life.

well there you go, if after a test run you still have issues, try another program!

try higher NB volts, since those boards were designed for 1GB modules at lower clocks, it may be overly stressing it.

 

[Krout]

The Load Tester provides 100% load to 1 core. You have 4 cores so fully loading one core is equivalent to 25% load on your CPU. Add in some background tasks and an overall load reading of 27% sounds right to me. RealTemp agrees with that gadget on your Desktop so it must be OK. You can run multiple copies of Load Tester if you want to create more load.
.

Here is another shot. It looks like Load TEster does not load a single core, but distribute the load between all four cores

 

[unclewebb]

If you want LoadTester to be locked to a specific core then you can use the Task Manager Set Affinity... option to do that.
LoadTester can provide 100% load to a single core in nice equal steps of 10%. How you distribute that load is up to you.

 

[leeakred]

Real Temp Calibration??

Hi everyone. I have always used real temp in the past, but i have recently installed a q9450 in one of my rigs, and not sure how to accuratley calibrate real temp to show the nearest correct temps.

I have read alot of the posts in these forums on the subject, but wanted to see if i could get some advice from the man him self

Ive had trouble re attaching my waterblock to the new CPU, as its a tight fit on my mobo ( P5Q Deluxe, using Swiftech H20-220 compact).
So worried about the fact i hope its seated correctly (re seatted 3 times) i want to try get the most accurate temps, as they seem alittle higher than id like at the moment.

I have overclocked the q9450 to 3.0ghz, with out any voltage increase. and attached the CPU cooldown test etc below;

How should i calibrate real temp? And do you think my temps are to high?

Sorry if this question has been asked many times before, but by reading other posts, every CPU is slightly differant? Thought i would try get the best answer from the experts.

Thanks for any help / advice!

 

[leeakred]

sorry, had trouble with image

 

[leeakred]

So what does anyone think. How much do i need to calibrate the TJmax by?
And do you think the temps are to high?

Thanks again for any advice.

 

[pothead0666]

can real temp be compatible with amd 64 athlon

 

[unclewebb]

leeakred: Your sensors look OK for slope error but my opinion is that Intel does not accurately set TJMax. My best guess would be to use TJMax = 100, 105, 100, 108. I find that Core 0 tends to be the most accurate which is why I leave that one at 100 and adjust from there. Perfect temperatures from idle to TJMax don't exist for 45nm Core 2 Quad processors but I think that calibration will be an improvement.

pothead0666: Unfortunately I haven't become rich writing free software and dealing with the infinite number of CPUs that Intel cranks out is enough work to keep me busy. I wish I had more time and some AMD systems to test on but I don't have either.

 

[leeakred]

Thank you! I will calibrate and see how it looks then.

I ran folding@home on all 4 cores (using vm ware) and GPU, to try stress whole system and see what the temps were like. Core 0 was hitting 60-61 max. Do you think that is fine? Its just with the water cooling (all be it lower end water cooling), i didnt know whether to expect it to be a bit lower temps. So still not sure wether to re-seat the block and re-applie TIM?

Thanks again!

 

[unclewebb]

leeakred: With Intel Core CPUs, people tend to be way too concerned about temperatures when they really don't need to be. What I learned from project RealTemp is that as long as your computer is stable and it's not thermal throttling, you can ignore your core temperature. It's just a number and there are so many places for possible error that it might not even be that accurate of a number, especially on 45nm Core 2 Quads. These CPUs are designed to run reliably with core temperatures over 90C so running them at 60C is fine and well within spec.

The inconsistency of these sensors from one to the next on the same CPU has caused a lot of remounting of heatsinks and water blocks but the vast majority of the problem is sensor error. If your computer is running fine then I wouldn't bother remounting anything.

 

[Mussels]

leeakred: With Intel Core CPUs, people tend to be way too concerned about temperatures when they really don't need to be. What I learned from project RealTemp is that as long as your computer is stable and it's not thermal throttling, you can ignore your core temperature. It's just a number and there are so many places for possible error that it might not even be that accurate of a number, especially on 45nm Core 2 Quads. These CPUs are designed to run reliably with core temperatures over 90C so running them at 60C is fine and well within spec.

The inconsistency of these sensors from one to the next on the same CPU has caused a lot of remounting of heatsinks and water blocks but the vast majority of the problem is sensor error. If your computer is running fine then I wouldn't bother remounting anything.

i can back this up too.

hot intel CPU's throttle before they take any damage - the only reason for concern is that the high temps tend to do bad things to nearby components like northbridges, VRM's and capacitors, causing THEM to overheat.

 

[leeakred]

Thanks for the advice!

Think i will see how it goes then, becasue at the moment everything is stable (touch wood), and il try not to put to much worry in the current temps.

Thanks again!