C2D temps revisited....

Discussion in 'Overclocking & Cooling' started by TheRapture, Mar 31, 2007.

  1. SuperKeijo

    SuperKeijo Gawd

    Messages:
    705
    Joined:
    Jan 18, 2007
    Ok, I ran the sucker without a heatsink.

    [​IMG]

    Fans are spinning! it's running!

    Idling in windows:

    [​IMG]

    This was the hottest spot I found on the IHS.

    Windows reports:

    [​IMG]

    This is confusing. The thermometer reports 1-4c higher than core temps in 0.94 which shouldn't be possible. But then again over 10c more to the IR reading sounds a bit too much also. So, maybe..

    -The TJunction is actually 90-95c? (This would actually make the most sense to me after all this)
    -IR reading is a few degrees off?

    The c2d ran absolutely fine at 600MHz and 0.7v (in bios) without a heatsink! I orthosed 10 minutes and the temp (acording to ct 0.94) never rose above 55c, pretty nice! I'm sure it would be fine up to atleast 1,0-1,2GHz at under 1v. That would be pretty sweat, running a cpu at over 1GHz without a heatsink!
     
  2. unclewebb

    unclewebb [H]ard|Gawd

    Messages:
    1,799
    Joined:
    Jun 21, 2006
    Thanks SuperKeijo for joining me on the crazy side of the fence! :D

    Yesterday when I pushed the power button I wasn't sure what to expect and was a little nervous for some reason but I think I'll head back to the "lab" today for some more testing.

    Now that I know that TjMax=85 is correct for my processor at high temps, that means that the on chip digital thermal sensors (DTS) are not very accurate when a Core 2 Duo is reading low temps. In previous testing at a 27C idle when my heatsink was still attached, the DTS based core temp numbers were at a minimum 6C too low.

    Users got suspicious when CoreTemp started reporting numbers below ambient but that doesn't necessarily mean that the assumed TjMax is wrong. It's more of a sign that the DTS isn't accurate at reporting low core temperatures and shouldn't be used for that.

    Yesterday at 60C and above the DTS seemed well calibrated to the real core temperature but at 40C where you were testing I think it is still off by a couple of degrees which is exactly what you were finding.

    You might want to try a small boost in your voltage and try running at 50C or 60C to check for accuracy.
     
  3. BillParrish

    BillParrish [H]ardness Supreme

    Messages:
    7,520
    Joined:
    Aug 25, 2006
    I dont understand all this at all, its not like Intel does not make the datasheets with revisions readily available and the thermal trip temps published. But it looks like Big Fun :D

    In searching for information on the "range" of DTS values that would be vaild, I tripped over this mess"

    http://www.intel.com/design/core2duo/documentation.htm

    The documents found here are not marketing bullshit, but (imo) akin to a certified letter from God. Perhaps I am too trusting but I think not.

    For instance in the latest specification update on C2D desktop processors, this caught my attention (bold is mine):

    AI65. A Thermal Interrupt is Not Generated when the Current Temperature is Invalid
    Problem: When the DTS (Digital Thermal Sensor) crosses one of its programmed thresholds it generates an interrupt and logs the event (IA32_THERM_STATUS MSR (019Ch) bits [9,7]). Due to this erratum, if the DTS reaches an invalid temperature (as indicated IA32_THERM_STATUS MSR bit[31]) it does not generate an interrupt even if one of the programmed thresholds is crossed and the corresponding log bits become set.
    Implication: When the temperature reaches an invalid temperature the CPU does not generate a Thermal interrupt even if a programmed threshold is crossed.
    Workaround: None identified.
    Status: For the steppings affected, see the Summary Tables of Changes.

    http://download.intel.com/design/processor/specupdt/31327914.pdf page 42.


    Affected processors:
    AI = Intel® Core™2 Extreme processor X6800Δ and Intel® Core™2 Duo desktop
    processor E6000Δ and E4000Δ sequence

    Applies to these steppings:
    B1 B2 L2

    Hmm the hits just keep on coming, not as worrisome but still not good.

    AI12 X- X- X- No Fix -Programming the Digital Thermal Sensor (DTS) Threshold May Cause Unexpected Thermal Interrupts


    So... the question is, will the cpu still thermal trip ? What would the interupt (that is not being sent) do ? Shutdown the machine ? Alarm ? Fan speed ? Order pizza ?


    Chapter 13 section 5 of this document for more info:
    http://www.intel.com/design/processor/manuals/253668.pdf

    It appears catastrophic trip is not affected I am still chewing on it all, however it appears somehow that despite sounding very bad, it may not be such a big deal after all. It may however explain some of the "odd" results the testing is seeing.

    Following the documentaion:
    http://www.microsoft.com/whdc/system/bus/PCI/MSI.mspx
    Message-Signaled Interrupts: The Basics
    MSIs provide a device and its driver an alternative to traditional line-based interrupts. To generate an MSI, a device delivers a “message” to a specific memory address. The Windows operating system provides to the device and the device’s driver both the message content and the address to which the message is delivered. The hardware mechanisms by which devices support MSIs are defined in the PCI 2.2 and PCI 3.0 specifications.

    It appears this interupt (that apparently is/may not be generated is serviced by the APCI (advanced programming and control interface)
    • LVT Thermal Monitor Register (FEE0 0330H) — Specifies interrupt delivery
    when the thermal sensor generates an interrupt (see Section 13.5.2, “Thermal
    Monitor”). This LVT entry is implementation specific, not architectural. If implemented,
    it will always be at base address FEE0 0330H.

    Bah, you have to be a member to get info on the APCI/PCIE spec.
     
  4. CoW]8(0)

    CoW]8(0) [H]ard|Gawd

    Messages:
    1,427
    Joined:
    Jul 25, 2005
    What exactly is considered an 'invalid' temperature?

    And to explain, an interrupt is basically an event that triggers the CPU to stop whatever its currently doing and run code to handle the interrupt (ie an interruption, and hence the term).

    The seriousness of this issue hangs upon what they mean by an 'invalid' temp? Is it a negative temperature? A temperature value that overflows?

    I believe the MSI specification you referring describes how devices on the PCI bus generate interrupts. Basically, the operating system tells the PCI device the specific register (and the content format) of where to trigger an interrupt. I don't think its directly related to the thermal interrupt issue.

    Well I haven't gone through this entire thread but I think I see the problem and the explanation. I believe the problem is within the Core Temp program, nothing to do with Intel (except maybe some bad documentation). Core Temp must be using a bad offset in version .94 Beta which was fixed in version .95. The reference was adjusted from 85C to 100C which in turn adjusted the Core 0 and Core 1 temperatures by +15C.

    Here's the changelog on the Core Temp site for version .95: http://www.thecoolest.zerobrains.com/CoreTemp/history.html
     
  5. unclewebb

    unclewebb [H]ard|Gawd

    Messages:
    1,799
    Joined:
    Jun 21, 2006
    An invalid temperature occurs when the processor continues to heat up and operate after DTS=0.

    On my processor that has a TjMax = 85C, when DTS = 0 the core temperature is 85C. When the core temperature reaches 86C, the DTS is undefined or invalid. The DTS is a 7 bit value so It actually wraps around to 127 and begins counting down again as the processor continues to heat up.

    CoreTemp 0.95 was adjusted to correct for very low temperatures on some processors by changing the TjMax from 85C to 100C. It wasn't adjusted for the early revision B2 processors like I have. CoreTemp 0.95 continues to use a TjMax = 85C for the B2 processors.

    I believe this adjustment or fix in CoreTemp 0.95 is likely wrong. It's possible that the DTS is simply not 100% accurate at very low idle temperatures. Adding 15C to the reported temperatures by changing the TjMax might make the reported low temperatures look more believable but now the higher temperatures could be being reported 15C higher than the actual temperature.

    All I know is that for my revision B2 E6400, when CoreTemp uses a TjMax=85C, the core temperatures reported are exactly equal to the readings from an IR thermometer pointed directly at the bare processor's IHS from about 48C to 85C. At lower temperatures like 20C, CoreTemp reports temperatures that are too low by at least 6C. It ends up reporting idle core temperatures that are below ambient which is impossible but TjMax=85C gives me excellent accuracy on the high end. The IR thermometer may not be reading the true core temperature but it does read the true ( IHS ) heat spreader temperature and I know when that hits 85C, DTS = 0 and my processor is throttling.

    If this is all way too confusing and you're not sure what to believe then stick to using the DTS. Don't let it get below 3 or thermal throttling will begin which slows your processor down to try and reduce its heat. If you are overclocking close to your processors maximum abilities then I would recommend not letting the DTS get below 15. You need extra headroom when you are pushing your processor to the max.
     
  6. CoW]8(0)

    CoW]8(0) [H]ard|Gawd

    Messages:
    1,427
    Joined:
    Jul 25, 2005
    The most important value shouldn't be the absolute core temperatures, it should be the difference between your core temperatures and Tjmax. And this always remains the same, regardless of whether the Tjmax is 85C or 100C.

    I believe the calibration for the sensor of your CPU may have been very inaccurate from the beginning. And coincidentally, it was off by about 15C which was the issue with Core Temp between versions .94 and .95.

    Intel had the intentions of having the Tjmax at 100C, but bad calibration meant that the CPU would always throttle at what it thought was 100C (or when the Tdelta-CoreTemp = 0). Based on this, there may be alot of Core 2 Duo's out there that throttle above or below 100C by a significant amount.
     
  7. speedlever

    speedlever Limp Gawd

    Messages:
    447
    Joined:
    Sep 29, 2006
    Very interesting (and lengthy!) reading of this thread.

    FWIW, here's my e4300 and stock HSF running on an ASRock 4coredual-VSTA mobo with 2x1 Gb Kingston value DDR2-667 ram, Fortron 450w PSU and eVGA 6600GT-AGP video card. I used AS Ceramique for the TIM (cleaned the factory TIM off with 91% alcohol and coffee filters).

    The e4300 is an L2 with batch number Q649.

    Minimum temp test (This board is sorta unique: OC mode can be untied, and here the CPU freq was dropped to 100 MHz). Ambient fairly warm: estimated at 27*C.
    [​IMG]

    Idling at 266 MHz (oc from 1.8 to 2.4)
    [​IMG]

    Under Orthos load at 2 hours:
    [​IMG]

    I also have an e6600 on an Asus P5BE with DDR2-800 ram running stock settings with a Noctua NH-U12F HSF. My SpeedFan temps are very close to the e4300 temps. Alas, I cannot run TAT or CoreTemp on the e6600 as that system is running Vista Ultimate 64 bit. The e4300 is running Vista Home Premium 32 bit.
     
  8. twogunpete

    twogunpete n00b

    Messages:
    54
    Joined:
    Mar 13, 2006
  9. unclewebb

    unclewebb [H]ard|Gawd

    Messages:
    1,799
    Joined:
    Jun 21, 2006
    twogunpete: Can you try posting those images again. I seem to have misplaced my magnifying glass! :D
     
  10. cryptonomicon

    cryptonomicon n00b

    Messages:
    17
    Joined:
    Jun 6, 2007
    What's the current consensus then that comes out of this discussion?

    If I currently want to use one app to monitor temps on a C2D, what should it be?

    Or is it two apps, one for the idle temps and one for the max temps perhaps?
     
  11. twogunpete

    twogunpete n00b

    Messages:
    54
    Joined:
    Mar 13, 2006
  12. cryptonomicon

    cryptonomicon n00b

    Messages:
    17
    Joined:
    Jun 6, 2007
    i'd honestly like to know what others here are going with in light of the doubts this thread throws on which software to go by in monitoring C2D overlock procs, so consider this a bump back to the front page for that purpose...
     
  13. SuperKeijo

    SuperKeijo Gawd

    Messages:
    705
    Joined:
    Jan 18, 2007
    Unclewebb and I came to the conclusion that Tjmax = 85 is actually the one to go with (for now). The idle temps do appear a little too low with the 85c readings but load temps seem to be more accurate. The IR tests without a heatsink are in favor of the 85c readings and 100c seems to be a little too high. Me, personally I'm going to go with the Tjmax = 85c for now. I'm not really that interested in idle temps anyway so it doesn't matter to me if they're a bit off..

    If someone has some more info/testings regarding the whole temps issue, please share!
     
  14. Evoic

    Evoic Gawd

    Messages:
    856
    Joined:
    Apr 30, 2007
    Right. So if I understand this thread correctly.......

    It's safe to assume idle + 8-10c and load would be closer to what's actually displayed.
    Correct?
     
  15. Warl0rd

    Warl0rd Limp Gawd

    Messages:
    170
    Joined:
    May 28, 2007
    Only if your reading the real Delta value (DTS) instead of the temp calculated using a tjunction .
     
  16. TheRapture

    TheRapture [H]ardness Supreme

    Messages:
    6,871
    Joined:
    May 31, 2000

    I am using CoreTemp .95, with the options set to show the DELTA between the current temp and TjMax.......typically under heavy load I see I still have 30c or so left before I even approach throttling....so I never see what the REAL temp is, just what the DTS is reporting before it hits the max suggested temp where throttling will occur.

    I am not too anal about temps, my stuff is set up correctly, contact is good, voltage is well under limits, and airflow is decent. Oh yeah and it is 100% stable. That's what matters.
     
  17. SuperKeijo

    SuperKeijo Gawd

    Messages:
    705
    Joined:
    Jan 18, 2007
    Certainly seems that way, yes.
     
  18. twogunpete

    twogunpete n00b

    Messages:
    54
    Joined:
    Mar 13, 2006
  19. SuperKeijo

    SuperKeijo Gawd

    Messages:
    705
    Joined:
    Jan 18, 2007
    twogunpete, read the thread. :rolleyes:
     
  20. twogunpete

    twogunpete n00b

    Messages:
    54
    Joined:
    Mar 13, 2006
  21. TheRapture

    TheRapture [H]ardness Supreme

    Messages:
    6,871
    Joined:
    May 31, 2000
    TAT is not showing any throttling going on....? Makes no sense....your cpu temp should not change according to which software you use, rather, the REPORTED temps will show differently...but changing software should not make the difference between overheating or not....

    What cooler are you using, you should be NOWHERE near 90c with any decent cooler unless you live in the sahara....:p
     
  22. twogunpete

    twogunpete n00b

    Messages:
    54
    Joined:
    Mar 13, 2006
  23. Warl0rd

    Warl0rd Limp Gawd

    Messages:
    170
    Joined:
    May 28, 2007
  24. jmackay

    jmackay 2[H]4U

    Messages:
    3,929
    Joined:
    Nov 23, 2006
    Now I'm way confused...60.1c is for e6600
     
  25. Evoic

    Evoic Gawd

    Messages:
    856
    Joined:
    Apr 30, 2007

    That's totally wrong......it has to be.
    I had an E4300 for testing last week and the chip was above 60c with no issues.
    I had 1.6v BIOS, 1.54v Actual going to the chip and under full load it was 68c in TAT, probably about 75c total on the CPU.

    No throttling, no locking, and ran SuperPi in under 16 seconds without so much as a hiccup.

    I think Intel is taking the "safe" route and covering their butt in the event that someone has a failure due to additional voltage increases.
     
  26. unclewebb

    unclewebb [H]ard|Gawd

    Messages:
    1,799
    Joined:
    Jun 21, 2006
    Those temperatures on the Intel website for desktop Core 2 Duo processors are NOT core temperatures. The Intel temperature spec of 60.1C or 61.4C is not a core temperature and it can not be compared to what a program like CoreTemp 0.95 or TAT reports.
     
  27. Evoic

    Evoic Gawd

    Messages:
    856
    Joined:
    Apr 30, 2007
    .....Well. If the temperature is measured at the center of the heatsink that sits flat on the CPU - what would you call that?
     
  28. unclewebb

    unclewebb [H]ard|Gawd

    Messages:
    1,799
    Joined:
    Jun 21, 2006
    Intel calls that 60C number the case temperature.

    The number that CoreTemp reports is based on the on chip digital thermal sensors ( DTS ) that have been strategically located by Intel at the hottest points on each core. There are multiple sensors ( I believe 8 ) per core. The sensors are compared and then the highest temperature ( lowest DTS reading ) is returned for each core.

    Depending on what type of software you are running, different areas of the core will run at different temperatures. Intel designed the DTS to get around the previous limitations of an average diode temperature.

    The only 100% Intel documented temperature for a Core 2 Duo desktop processor comes directly from the DTS and CoreTemp 0.95 can report that number in real time. Just keep in mind that it is not an absolute temperature but a relative temperature. It tells you exactly how many degrees you are away from the processor starting to thermal throttle itself.

    Here's a good picture of a Core 2 Duo. The center of the heat spreader is actually on the bottom edge of the 2 cores between core0 and core1. A temperature measured there is not always, if ever, going to represent the hottest part of the core.
     
  29. Evoic

    Evoic Gawd

    Messages:
    856
    Joined:
    Apr 30, 2007
    So.....you're telling me that there is a 10c, 15c, or even 20c difference in temperature in the few millimeters difference between the DTS reading and the "Case center".

    With all due respect - as I know you've spent quite a fair bit of time on this subject - I respectfully disagree.

    Intel is spec'ing the number too low.
    Plain and simple.

    How they get to that number, can be debated - depending on technique.
    WHY they chose that number is a better question for my interests.
     
  30. unclewebb

    unclewebb [H]ard|Gawd

    Messages:
    1,799
    Joined:
    Jun 21, 2006
    Evoic: One previous test I did was to run one core at full load and the second core at zero load. This was a steady load test where the temperature of both cores was allowed to fully stabilize and the difference between the hottest spot on both cores was an average of 6C. The difference between the hottest spot on one core compared to the coldest spot on the other core may have been slightly higher but there's no way to test for that.

    In the normal operation of a cpu you have large instantaneous temperature gradients. Start and stop Orthos a few times while monitoring with CoreTemp and you'll see how the temperature can change quite rapidly, both up and down.

    I don't think an instantaneous temperature difference of 20C or 25C in the millisecond or microsecond world of a cpu is totally out of the question. When a processor is mostly idle you won't see any large temperature gradients. When a cpu is working and is rapidly performing and switching between a variety of tasks, I think a 25C difference between an 85C core temperature reading measured at the hottest point on the chip and a 60C average temperature reading measured at a spot on the opposite side of the heat spreader at the edge of the cores is not unreasonable. Also keep in mind that Intel is assuming that most users will be using their OEM heatsink and fan which doesn't have the cooling performance of the after market coolers that most of us are running.

    An analog Tcase temperature can not be measured as accurately as reading a digitally calibrated on die sensor so it's possible that Intel left a little more head room in this 60C number to be on the safe side.

    The bottom line is that it doesn't really matter what Intel did or what that 60C number is all about. My cpu didn't begin to thermal throttle until the DTS got down to 2 which is equivalent to a CoreTemp absolute reading of 83C. The throttling is pretty mild here and most users wouldn't even notice a difference. At DTS=0 and beyond, which is reported as an absolute temperature of 85C for my E6400, the throttling starts to get more aggressive. Avoiding throttling and maintaining stability are the only important things to keep an eye on.

    At 3600 MHz and a load voltage of 1.48 volts my processor loses stability if the DTS drops below 25 but is 100% Orthos stable when the DTS is over 25. When fully overclocking your processor, the throttling point becomes a non-issue because you will lose long term stability and your computer will reboot long before you get near the throttling point or the maximum Intel rated temperature of your cpu.
     
  31. Cousin Patty

    Cousin Patty Gawd

    Messages:
    634
    Joined:
    Jul 21, 2005
    Ran my e6600 which is OC'ed to 3.2ghz last night with Orthos for over 7 hours and core temp said the cores didnt get above 57 degrees Celcius. I'm happy with that. My HSF I had before I got my Nautilus kept the processor at 58 degrees idle...so thats a big improvement. Even reseated the thing multiple times.

    I'm just gonna try lowering some voltages and then I think I'm done. System is OC'ed to a decent level and my system's stable. So I'm happy.
     
  32. unclewebb

    unclewebb [H]ard|Gawd

    Messages:
    1,799
    Joined:
    Jun 21, 2006
    Evoic: Here's some numbers to back up the size of the temperature gradients that I was talking about.

    I used TAT and cycled full load onto Core0 on and off a couple of times.

    Using CoreTemp with the logging set to a 1 second interval you can see that parts of the core can change temperatures up and down by up to 25C within one second and that's using a Tuniq Tower. An OEM cooler would likely provide even greater instantaneous changes. The maximum temperature gradient between the two cores was 13C.

    I don't believe that a sensor positioned at the center on top of the IHS as per the Intel Tcase spec, is going to change temperatures as quickly as parts of the core can.

    [​IMG]
     
  33. Evoic

    Evoic Gawd

    Messages:
    856
    Joined:
    Apr 30, 2007
    So.....in essence, you're saying that the cores can reach much higher temps for a moment (as much as a 25c difference for a split second), but that Intel's determination to place their peak number at 60c is correct ? ?

    If the cores are heated to their near throttle max - let's call it a steady 82c......you don't feel that the center of the "case" will easily read above 60c for extended periods of time? (above Intel's spec)

    I think we may be arguing 2 different points here.
    My estimation is that Intel's 60c is grossly under-rating what the CPU can absorb and still continue to function properly and efficiently without throttling.

    You're arguing how the measurement of said temperature can be interpreted.
    (Which, by the way, either way it is interpreted, they're still off by a longshot) :)
     
  34. unclewebb

    unclewebb [H]ard|Gawd

    Messages:
    1,799
    Joined:
    Jun 21, 2006
    You're absolutely right. Sorry for getting a little side tracked.

    When I was running my E6400 with the heatsink off and with the IR thermometer pointed directly at where Intel recommends you measure the Tcase temperature, my E6400 continued to run fine at 85C.

    I had a look through the Intel datasheet. The 60.1C number for Tcase is a recommended maximum for heatsink designers when you are operating this processor at its Thermal Design Power (TDP) of 65 watts. When we are overclocking and over volting our processors and running TAT or Orthos on both cores for extended periods of time, these processors are using far more than their Intel designed 65 watts.

    I think the 60C number is a good safe minimum and if your Tcase temp never goes over 60C then you are guaranteed that no part of your processor will be at risk of heat related damage.

    If you are using CoreTemp which monitors the on chip digital thermal sensors then the 60C number is not relevant. The digital sensors do a much better job of tracking all of the hot spots on the processor so Intel was able to design the thermal throttling point and thermal shut down point at a higher level based on the DTS data. Your processor is running within the Intel spec and will function properly as long as the DTS is reporting a positive number above zero.

    Chapter 5 Thermal Specifications and Design Considerations from the Intel data sheet has more detailed information about the Tcase 60C number.

    "Intel recommends that complete thermal solution designs target the Thermal Design Power (TDP) indicated in Table 27 instead of the maximum processor power consumption. The Thermal Monitor feature is designed to protect the processor in the unlikely event that an application exceeds the TDP recommendation for a sustained periods of time. For more details on the usage of this feature, refer to Section 5.2. In all cases the Thermal Monitor and Thermal Monitor 2 feature must be enabled for the processor to remain within specification."
     
  35. Evoic

    Evoic Gawd

    Messages:
    856
    Joined:
    Apr 30, 2007
  36. TheRapture

    TheRapture [H]ardness Supreme

    Messages:
    6,871
    Joined:
    May 31, 2000

    If you would READ the Intel spec, they say the temp of 60c or 61c is on the TOP of the INTEGRATED HEAT SPREADER, not the INTERNAL core temp.

    TAT and Speedfan as well as CoreTemp read the INTERNAL diode.

    I respectfully disagree, you are wrong, unclewebb is right. Intel says It is measured at the geometric center on the topside of the processor integrated heat spreader.
     
  37. Astral Abyss

    Astral Abyss 2[H]4U

    Messages:
    2,660
    Joined:
    Jun 15, 2004
    You guys are looking for something that isn't there. What Intel is saying is that 61.4C is the highest temperature you can achive and still maintain the 65W power rating of the processor. AMD does the same thing. It does not say anything about the absolute maxium the processor can handle, and obviously that is nowhere close to the maximum... Anything more and you will be out of Intel approved "design specs".

    Please read:

    Thermal Design Power: (Also referred to as Thermal Guideline) The maximum amount of heat which a thermal solution must be able to dissipate from the processor so that the processor will operate under normal operating conditions. (emphasis added)

    Normal operating conditions does not include overclocking or overvolting...
     
  38. Evoic

    Evoic Gawd

    Messages:
    856
    Joined:
    Apr 30, 2007
    Read the post directly above this one, and feel free to apologize for your tone.

    Apparantly, it's a little easier to misinterpret than you thought. :rolleyes:

    Thank you for pointing out what appears to be a more clearly worded description of what we're looking at, Astral.