1. Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Need Help with RealTemp 3.42

Discussion in 'RealTemp' started by bushisland, Nov 28, 2009.

  1. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    Hi,
    I am having a problem with the calibration. Attached is some screen captures.( Hope they are OK) I did a before sensor test at my Regular settings, then I did the low idle calibration . I then did another sensor test at my regular settings. Does not look right to me. Temps are not close and Distance to TJMax does not add up to 100. I'm confused. Please help. Thank You. CPU is Q9450 and OS is win7 64 bit. http://forums.techpowerup.com/attac...ent.php?attachmentid=30932&stc=1&d=1259375217
     

    Attached Files:

  2. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    I'd like to be able to help you but I can't easily see those first two screen shots. If you can't be bothered to crop it down to what's important then at least upload it somewhere like

    www.imageshack.us

    so it doesn't get reduced in size to the point where I have to strain to see it.

    It looks like you've got 4 sensors that likely have a unique TJMax. Run Prime95 Small FFTs at your normal overclock for about 5 or 10 minutes and post your results. That will let me see how much error is error in TJMax and how much is slope error. I originally thought that TJMax was consistent across all 4 cores and the whole problem was slope error but after seeing enough of these calibrations, I've learned that the opposite is closer to the truth.

    Post some info that I can actually see and I'll try to help you out.

    Don't worry about the old Distance to TJMax plus Temperature must add up to 100. That would be true if TJMax was exactly equal to 100 but Intel admits that 100 is kind of a ball park figure with no stated error.

    RealTemp 3.49
    http://www.fileden.com/files/2008/3/3/1794507/RealTempBeta.zip

    Most of the recent improvements have been for better ATI support which probably doesn't effect you.
     
  3. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    Sorry, My mistake.
     
  4. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    No problem. The easier it is for me to see, the easier it is for me to help you. At least none of your sensors are getting stuck in their normal temperature range. I should be able to come up with an easy calibration to get your temps to line up a little better. These sensors aren't that accurate and my calibration suggestion is based partly on experience and partly on guessing and some assumptions. Knowing that these sensors are not that great, it's not worth getting too technical trying to get perfect numbers out of them. That's simply impossible with the quality of sensors and the data about them that Intel has provided.
     
  5. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada

    Attached Files:

  6. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    That's better.

    I find that somewhere around 65C or 70C these sensors show very little slope error so by then it becomes clear that the difference in temperature is mostly a difference in TJMax. You're right on the lower edge of that temperature range.

    If you can, just run Prime95 Small FFTs for about 5 or 10 minutes until the temperatures stabilize. It might get another degree or two hotter than the above picture.

    Edit: I've started comparing numbers but I have to go. I'll stop by tomorrow morning and tell you what I think.
     
    Last edited: Nov 28, 2009
  7. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    OK here is is .I took a snap shot at 100% load after 10 minutes and another after stopping Prime95 at idle. Room temperature was 23C and case temp was 23c. I used a good temp probe on a Kaise AC/DC Digital Clamp Meter (Model SK-7701) to obtain Temps.
    .http://forums.techpowerup.com/attachment.php?attachmentid=30945&stc=1&d=1259410501
    htthttp://forums.techpowerup.com/attachment.php?attachmentid=30946&stc=1&d=1259410501p://forums.techpowerup.com/attachment.php?attachmentid=30944&stc=1&d=1259http://forums.techpowerup.com/attachment.php?attachmentid=30947&stc=1&d=1259410501410501
    Thanks
     

    Attached Files:

  8. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada

    Attached Files:

    Last edited: Nov 28, 2009
  9. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    The sensors for your first 3 cores have similar slopes while core 3 is significantly different.

    For the first 3 cores I would adjust TJMax to 99, 103 and 105.

    For the 4th core, if you want accurate load temperatures, I'd use TJMax = 102. Because of the slope error, this will have it about 3C different at idle compared to the other ones. If you want it to be the same at idle then you could set it to TJMax = 99C and keep in mind that it is going to be reading about 3C too low at full load. There's no happy medium.

    Using calibration factors to correct the slope would require using a very high TJMax value with a large negative calibration factor. This would give you temperatures very similar to the other cores from idle to your normal full load but once you went beyond that range it would start reporting significantly too high. If you want to see how this looks try setting Core 3 to TJMax = 112 with a Calibration Factor of -9.6

    I tend not to bother with fancy calibrations that look nice over a limited range. Easier to just set TJMax to about 102 and keep in mind that it's not perfect. I wish these sensors didn't have slope error and I wish Intel did a better job of calibrating and documenting them but they don't.
     
  10. pantherx12

    pantherx12 New Member

    Joined:
    Jan 2, 2009
    Messages:
    9,714 (4.56/day)
    Thanks Received:
    1,699
    Location:
    ENGLAND-LAND-LAND
    Hi what would you reccomend I set my TJmax's to?

    Thank you : ]
     

    Attached Files:

  11. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    For the early Quads, I've seen quite a few that core 0 and core 1 were set to one number and core 2 and core 3 were set 5C higher. In your example, try using TJMax = 90, 90, 95, 95. Slope error is not an issue for these early sensors. I used to think it was until I discovered that most of the issue is that TJMax is not consistent across all 4 cores. My theory is that this is done deliberately by Intel to keep all 4 cores from reaching the thermal throttling point at the exact same time which would result in a noticeable drop in performance. Unfortunately, there is no documentation to back up this theory. It's the only reasonable explanation I've been able to come up with to cover the data I've seen and collected.
     
  12. pantherx12

    pantherx12 New Member

    Joined:
    Jan 2, 2009
    Messages:
    9,714 (4.56/day)
    Thanks Received:
    1,699
    Location:
    ENGLAND-LAND-LAND
    Thanks for the quick reply.

    I had them set to 85/85/90/90 to level the temperatures myself.

    Sounds like a logical explanation to me.
     
  13. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    The best way to check which way is more accurate is to reduce your core voltage as close to 1.10 volts as possible and set your bios to 266 x 6.0 ~ 1600 MHz and then boot up and compare your idle temperatures to your room temperature near your open computer.

    I've never specifically tested that Xeon model. With the above test, it's usually clear whether the first two cores are 85C or 90C.

    TJMax = 85C became a legendary number from some Pentium 4 era manual. It was used by most early temperature monitoring programs and continues to be used by most programs but I haven't seen any relevance for Core 2 Duo or Quad processors. Only Intel knows the real truth and they're not saying.
     
    pantherx12 says thanks.
  14. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    Thank You Sir.(I was away for a day) I will set these new values, retest and send info back to you.
     
  15. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    The original calibration formula I developed for RealTemp was able to handle CPUs like yours much better. Unfortunately, based on generic information released at an IDF conference by Intel, I changed my formula and not for the better. I was depending on Intel to finally release some accurate engineering information about the sensors they use but instead most of the information released was more of the PR variety.

    I think I'm going to add an extra variable back into my formula so users like yourself that are interested in calibrating will be able to come up with a much better approximation for situations like core 3. I'll try to post something here in a day or two. Send me a PM if I forget.
     
  16. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    I have the settings set as advised. You are dead on. I'm happy:) See attachments.
    htthttp://forums.techpowerup.com/attachment.php?attachmentid=30982&stc=1&d=1259511946p://forums.techpowerup.com/attachment.php?attachmentid=30983&stc=1&d=1http://forums.techpowerup.com/attachment.php?attachmentid=30984&stc=1&d=1259511946259511946

    In your opinion, what is the highest temperature that you would run this CPU at 24/7?
     

    Attached Files:

  17. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    As I said, I think I can tweak my formula so all 4 sensors will track very close together from idle to TJMax but what you have so far looks like an improvement.

    I generally don't make specific temperature recommendations because there isn't much need to. Most users do some testing with Prime95 or similar and see the highest temperatures that they'll ever see. During normal use, they never get up to a similar temperature.

    I concentrate on stability. If your computer is 100% stable and it is not thermal throttling then I don't see any need to be concerned about the core temperature. Your cooling solution is more than adequate to be running at 3.4 GHz. Intel puts a 3 year warranty on their CPUs for a reason. They're very reliable even when users overclock them.
     
  18. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    OK, if you do not post here in a few days I will PM you, as requested. Yes it is better to have a stable system. My screen freezes up after about a hour into Prime95. I know the red line for this CPU is 1.36V. I'm at 1.28125V now, so I have some head room left. Also, I want to say thank you for all the help you are giving.
     
  19. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    Unclewebb,
    Based on the new TJ Max settings you have given me , is it OK to do a low idle calibration now?
     
  20. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    You can give it a try. A lot of the 45nm sensors get stuck at lower temperatures so it might not tell you anything new. Take it down to 2000 MHz and 1.10 volts and see how your idle temps compare to your room temperature near your open case.
     
  21. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    I gave you the wrong info, will edit in a min.
    Room temperature near case is 25C
     

    Attached Files:

    • temp.jpg
      temp.jpg
      File size:
      48.9 KB
      Views:
      141
    Last edited: Dec 1, 2009
  22. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    That looks correct to me. 8C over room temperature for a Quad at those settings is what you should see so 33C in a 25C room is perfect. Hopefully I have some time tomorrow to tweak my calibration formula.
     
  23. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada

    Attached Files:

  24. bushisland

    bushisland

    Joined:
    Nov 28, 2009
    Messages:
    42 (0.02/day)
    Thanks Received:
    0
    Location:
    Ontario, Canada
    I goofed. Pass on this
     
  25. unclewebb

    unclewebb RealTemp Author

    Joined:
    Jun 1, 2008
    Messages:
    949 (0.40/day)
    Thanks Received:
    422
    You may have sensors that are getting stuck. Is your case computer case open?

    I test with an open case and usually on its side to reduce the chance of heat transfer from the graphics card.r

    You can try another Cool Down Test at 2000 MHz and 1.1 volts and maybe you will be able to see if any sensors stop moving.
     

Currently Active Users Viewing This Thread: 1 (0 members and 1 guest)

Share This Page