1 2 Previous Next 15 Replies Latest reply on Aug 3, 2011 10:36 PM by parsec

    Intel DP67DE Throttling

    Ready_Eddie

      Hello All,

       

      I have DP67DE MB with a i5-2500K uP (dating myself) and am exploring overclocking via the turboboost settings.  Nothing extreme, say a ratio of 45.

       

      Running stressors such as Prime95 or Linpack result in the uP changing from 16 ratios to 45 as expected, but then after a minute or two the uP "throttles" back to a ratio of 33 and stays there.  There is not much in the forums about this for this board, but what there is confirms the result, and noone knows why.

       

      It does not appear to be thermal throttling.  I have a waterblock on the CPU and the 4 core temps never go above 52 degC.  All the other temps (from Intel Desktop Utilities and CPUID/HW Monitoring) are below 40 degC.

       

      I have been up and down the MSRs (using CrystalCPUID and BITS from biobits.org) and can't see what is triggering it.  I've peeked and even poked a few times.  The target is a ratio of 45.  On demand Clock modulation is not active.  For the cores, Thermal Stautus Log is 0,  PROCHOT# or FORCEPR# log is 0, TM1 and TM2 logs are 0, and Power Limitation log is 0.  Same for the Pkg Therm Status flags.  The C-state counters are holding steady.  The BIOs power settings (increased to move them out of the way) appear where they should in the RAPLs.

       

      The bios has a max current setting, but I can't find it in the MSRs.  The only potential location for it gives a GPF when read.

       

      So I'm stumped as to why to it is throttling.  The only thing I notice is the when you start the run, CPUID/HWMonitor show power ramping up to between 90 and 92 watts, as if  it is reading and averaging over several seconds.  When it reaches about 90 watts, the CPU throttles down to 33 ratios.

       

      My best guess is the MB Voltage Regulator is limiting power and/or current independently of the CPU and the BIOS settings.  If so, I don't see how it is telling the CPU to throttle.  Stealing clock cycles perhaps?  The VR temperatures are low, less than 30 degC.

       

      Is the VR current power/current likely the limiting factor?

       

      I see a design change notice for this board to upgrade it from a 3-phase to 4-phase VR.  I happen to have a B2 stepping board, can I swap for one of these down the road?

       

      Regards

        • 1. Re: Intel DP67DE Throttling
          parsec

          Given your description of the phenomenon and the environment in which it occurs, I'm wondering if you are experiencing the "Turbo Boost Technology Power Control " function of your CPU, and it is behaving as described in Section 6.4.1 of this document:

           

          http://download.intel.com/design/processor/designex/324644.pdf

           

          On the other hand, I am surprised I have not seen this occurring and discussed in testing of this series of CPUs by the various PC hardware review web sites.  I'll try to test this myself, although I have a somewhat different CPU.  If possible, you could check the values in the data locations described in that section.  I would be interested to know what the cause of this situation is, if you ever discover it.

          1 of 1 people found this helpful
          • 2. Re: Intel DP67DE Throttling
            parsec

            After doing some research, I believe I can answer some of your questions accurately.

             

            2nd Generation Intel Core CPUs set and adjust their Clock Speed with the CPU multiplier, and adjust CPU voltage using VID  (Voltage Identification).  This is done within the CPU itself, and the  CPU controls the voltage regulators, instructing them to raise or lower  the voltage they deliver to the CPU via the SVID bus.

             

            The  voltage regulators do not decide what voltage to provide to the CPU,  they respond to the CPUs requests.  While it is possible that the  voltage regulators can fail to provide the voltage requested by the CPU  due to some limitation or fault, normally they should have been designed  to provide the CPU at least it's specified voltages and current.  What  the CPUs reaction to not receiving the power it is requesting I don't  completely understand, but I imagine it will adjust itself accordingly  and possibly log an error regarding power delivery.  Regardless, the CPU adjusts the voltage regulators, which only obey the CPU, or fail to do so.  You may be able to increase the power available from the voltage regulators with BIOS settings.

             

            CPU clock speed, when  changed by Turbo Boost, is done by changing the CPU multiplier.  Don't  forget that Intel considers Turbo boost to be a temporary increase in  performance that is provided "opportunistically", basically meaning as  long as the CPU is operating within its thermal parameters.  That  determination is done by the CPU, and the method is generally described  in the section of the datasheet  I mentioned in my previous post.  Intel considers the maximum Turbo  Boost frequencies can or will be operating the CPU beyond it's maximum TDP rating, which is why Turbo Boost is temporary.

             

            Although  the description and functioning of Turbo Boost are provided by Intel,  it seems they are not written in stone.  Given an appropriate mother  board and BIOS settings capability, changes to the CPU multiplier via  Turbo Boost can be made permanent, meaning the multiplier won't  "throttle" down to a lower value, as long as the CPU does not overheat.   On my ASUS P67 mother board  with an i7-2600K, I set the Turbo multiplier to 42 on each of the four  cores (one must be set above the others to 43) and ran several  consecutive runs of Intel's Linpack stress testing program.   I did not adjust any power related settings.  I monitored the CPUs  clock frequency with several different programs during several test  instances to verify the accuracy of the clock speed.  Using a high  performance air cooled CPU cooler, I would reach a maximum CPU temp of  60C, and the CPU clock frequency stayed at 4200MHz for the duration of  the test, up to 15 minutes, as all my temps leveled off long before that  and IMO any change in the CPU multiplier/clock frequency should have  occurred in that interval.

             

            In your case what is likely happening is your Intel  mother board and BIOS settings are playing by the rules of Turbo Boost  2.0 as described in the section of the document I referenced.  If your  VR were the problem, I think they would fail to provide the necessary  voltage very quickly, and your multiplier would drop immediately.  You  might be able to change that behavior depending on the settings  available in your BIOS.  The BIOS Glossary document for all Intel mother  boards shows there are settings in the Processor Overrides, Turbo Boost  Tech. options that may allow you do accomplish this, but I don't know  if those are available in your boards BIOS.

            • 3. Re: Intel DP67DE Throttling
              Ready_Eddie

              parsec, sorry for the delay, I was out of town.

               

              parsec wrote:

               

              Given your description of the phenomenon and the environment in which it occurs, I'm wondering if you are experiencing the "Turbo Boost Technology Power Control " function of your CPU, and it is behaving as described in Section 6.4.1 of this document:

               

              http://download.intel.com/design/processor/designex/324644.pdf

               

              On the other hand, I am surprised I have not seen this occurring and discussed in testing of this series of CPUs by the various PC hardware review web sites.  I'll try to test this myself, although I have a somewhat different CPU.  If possible, you could check the values in the data locations described in that section.  I would be interested to know what the cause of this situation is, if you ever discover it.

               

              POWER_LIMIT_1 is 340h, 104 watts, my BIOS setting at 10% higher than the 95 watt default.  I have tried it much higher.

               

              POWER_LIMIT_1_TIME is 0Ah, .999424 sec, set by BIOS to 1 sec the default.  I have tried it at the 32 sec max setting.

               

              POWER_LIMIT_2 is 410h, 130 watts, my BIOS setting at 10% higher than the 118 watt default. I have tried it much higher.

               

              Note that IA32_THERM_STATUS at register address 19Ch has bits 0,1,2,3,4,5,6,7,8,9,10 and 11 all clear, all 4 cores, indicating no thermal or power threshold exceeded from the viewpoint of the CPU.  But it is running at a ratio of 33 as if Turbo Boost has been disabled.

               

              However, if I stress it a little bit more (LINPACK insteady of Prime95), then the CPU will after about a minute throttle all the way back to a ratio of 16, with bits 2 (PROCHOT# or FORCEPR#) and 3 (PROCHOT# or FORCEPR# log) being set.  I believe this is the MB/VR signaling the CPU via the bi-directional PROCHOT#.

              • 4. Re: Intel DP67DE Throttling
                parsec

                Hmm, interesting, and many things to check that are related to each other.  What is mysterious and contradictory is what occurs, throttling via CPU multiplier when you apparently see CPU and VR temperatures below the PROCHOT value, well below in the VR's case.  Why would the VR be sending a PROCHOT to the CPU when the VR is below 30C?

                 

                 

                The following statement from the 2nd Gen. Core Processor TSMDG Document  bothers me:

                 

                When operating in the turbo mode, the processor will monitor its own power and adjust
                the turbo frequency to maintain the average power within limits over a thermally
                significant time period. The package, processor core, and graphic core powers are
                estimated using architectural counters and do not rely on any input from the platform.

                 

                What I mean is that it sounds like Turbo will be throttled down simply because the algorithm is programmed to do so even when the thermal parameters are within acceptable values.  Does that statement mean that the Turbo boost is designed to be safe on any platform, meaning the worst case, lowest common denominator situation?  Thus high performance cooling is ignored and we're stuck with a dumbed-down throttling algorithm?

                 

                In my experience, that does not necessarily seem to be the case, as I described in my earlier post.  Exactly what is activated, disabled, etc, I do not know, but I don't have the issue that you do on my i7-2600K PC.  Is there an "ignore" value or "infinite" value for the time parameters you checked, for example, that my mother boards UEFI sets to allow endless Turbo boost frequencies?  Given that all the parameters you checked are as they should be, and your temperatures are correct, I am at a loss for an explanation, beyond my conjecture above.  I am still researching this situation, but all I can see at this time is your mother board and BIOS/UEFI may be the limiting factor.  Have you experimented with UEFI settings to obtain a different result?

                 

                Message was edited by: parsec

                • 5. Re: Intel DP67DE Throttling
                  Ready_Eddie

                  parsec wrote:

                   

                  The following statement from the 2nd Gen. Core Processor TSMDG Document  bothers me:

                   

                  When operating in the turbo mode, the processor will monitor its own power and adjust
                  the turbo frequency to maintain the average power within limits over a thermally
                  significant time period. The package, processor core, and graphic core powers are
                  estimated using architectural counters and do not rely on any input from the platform.

                   

                   

                  Yes, I saw that too. There is a MSR register that is accumulating CPU energy (don’t know why they start with energy or how they measure it).  If you take the derivative of the energy (ergo the BIOS power time interval settings) you get power.  If you divide the power by voltage, you get the CPU current.  I’ve tried this calculation with a stop watch and get close to what my tools say for power.

                   

                  The power/current settings are loading from my BIOS settings into the RAPL MSRs, as described in section 14.7 of Volume 3A System Programming Guide.  If the CPU calculated limits set here were being exceeded, then I would expect bits 10 and 11 (Power Limitation Status and its sticky bit log) of the IA32_THERM_STATUS  at MSE 19Ch to be asserted. Nope.  Bits 0-4 are also clear, so PROCHOT# has not been asserted by either the CPU or MB.

                   

                  The MSR that sets the current limit was recently added to the Sandy Bridge table of MSRs at 1ACh.  But when I try to read it using the Intel biosbits.org tool, I get a GPF.  I suspect it is not really there.  But then where is my board’s BIOS putting the current limit setting?

                   

                  Ultimately, all I can say it this behavior, dropping from a ratio of 45 to 33 after 65 seconds, is a mystery.  I can’t find anything in the MSRs that indicate the processor is in a throttled state, other than IA32_PERF_CTL at 199h requesting 45 for all 4 cores, and the IA32_PERF_STATUS at 198h saying the current state is 33.  It is as if the CPU doesn’t know it is in a throttled mode, and some external agent is doing it.

                  • 6. Re: Intel DP67DE Throttling
                    parsec

                    Eddie,  I certainly agree with your statement, it is a mystery.  I must say that I have no experience extracting the MSR register data, as you have been doing, although I am capable of reading and interpreting it.  I'm sure you have checked obvious things like new or updated codes in the registers, or that the programs reading the register are up to date, but you know more about that than I do.

                     

                    Focusing closer on the statement I posted from that document, I wonder about:

                     

                    The package, processor core, and graphic core powers are estimated using architectural counters and do not rely on any input from the platform.

                     

                    What does that mean and what are it's implications?  I interpret that as the Turbo throttling algorithm has hard-coded parameters (does not rely on any input from the platform) but is still triggered by something, and reduces the CPU multiplier arbitrarily.  That seems to simplistic, and could be implemented simply as a timer that is activated when the CPU is at the maximum Turbo mulitplier, and reduces it when it times out, assuming the CPU will be overheating, whether or not it actually is.  I am at the limits of my knowledge about this, although I find it interesting, and will try to find the time to see what is in my MSRs.

                     

                    There are many obvious, simple things such as your temperature readings are incorrect, or your cooling is not functioning correctly, which you likely have checked.  I am intrigued by my experience with my i7-2600K, and what I have read about over-clocking it and the i5-2500K, which is contrary to your experience, that being there is no throttling seen under the conditions you describe.  That is still a mystery to me but I am looking into it.

                    • 7. Re: Intel DP67DE Throttling
                      Ready_Eddie

                      parsec, thank you for your time and interest.

                      parsec wrote:

                      I must say that I have no experience extracting the MSR register data, as you have been doing, although I am capable of reading and interpreting it.

                      CrystalCPUID has a simple MSR walker as a sub-tool.  Although the main tool hasn't been revised for Sandy Bridge last time I looked, the walker is general purpose.  I have found that if an address is missing from what the walker displays under windows, biosbits under GRUB gets a GPF (General Processor Fault?) trying to read it, i.e. not used, doesn't exist.

                      parsec wrote:

                       

                      The package, processor core, and graphic core powers are estimated using architectural counters and do not rely on any input from the platform.

                       

                       

                       

                      The term "architectual" is Intel-speak for common across the processor family.  I suspect the "architectural counters" are the MSR_PKG_ENERGY_STATUS at 611h, which contains "Total Energy Consumed" combined with one of the time counters, say IA32_TIME_STAMP_COUNTER (TSC) at 10h:

                       

                      MSR_PKG_ENERGY_STATUS is a read-only MSR. It reports the actual energy use for

                      the package domain. This MSR is updated every ~1msec. It has a wraparound time

                      of around 60 secs when power consumption is high, and may be longer otherwise.

                       

                      Total Energy Consumed (bits 31:0): The unsigned integer value represents

                      the total amount of energy consumed since that last time this register is cleared.

                       

                      You wrote:

                      That seems to simplistic, and could be implemented simply as a timer that is activated when the CPU is at the maximum Turbo mulitplier, and reduces it when it times out, assuming the CPU will be overheating, whether or not it actually is.

                      I agree, too simplistic given all the information the CPU has.  This only seems to happen with Intel DP67DE and DP67BA boards, if it were something simple running open loop implemented by the CPU, I would expect more MBs to show it in the forums.

                       

                      The best user thread I've found is: http://hardforum.com/showthread.php?t=1574834, starting with post #52.  My searches seem to indicate other MBs work more as expected.

                      Regards

                      • 8. Re: Intel DP67DE Throttling
                        parsec

                        Eddie,  Thanks for the explanations and suggestions, I'll see if I can use the tools you suggested, as I'd like to check into the contents of these internal CPU registers, for this topic and others.  If I (ever) come up with something interesting, I'll post it in this thread, if you have the email notification turned on, you'll know.

                         

                        I've nothing else to offer you in your search for an answer.  The two Intel boards you mentioned belong to their Media Series, and I wondered if they were from the Executive Series, which are of the "corporate stable" type, and would likely be limited in their potential to operate outside of the rules, but that is not the case.

                         

                        Any chance that Windows is affecting what you experience, such as power management settings?  I'v seen them do that in the past, but you've likely already taken that into account.

                         

                        You seem to realize what the limiting factor is in general.  If you ever discover what exactly is the culprit, I'd be interested in hearing about it.  Good Luck!

                        • 9. Re: Intel DP67DE Throttling
                          downset

                          hello,

                           

                          i have the exact same problem

                           

                          i tried every possible bios setting, offset mode and static mode, increased every possible related value well above what i am comfortable with but the scenario is always the same

                          i tried with x40 and x38 multipliers, something that should be very easy with a dedicated air cooler like the mux-120

                           

                          if i run a load test, the overclock works perfect and cpu-z shows it runs at x40 (or x38)

                          however after 3 min it throttles down to x33, cpu temp is never more than 50 C

                           

                          the asus p8p67-M motherboard in my wife her pc overclocks to x44 without any trouble only difference in our setup is the motherboard.

                           

                           

                          if you guys find a reason or a setting i would be very interested to know

                           

                          would it help to contact intel about this? i feel if i spend the money on a p67 board and a K-series i5 i should be able to overclock

                          • 10. Re: Intel DP67DE Throttling
                            Ready_Eddie

                            No, I don't think anyone ever did.

                             

                            I was hoping Inel was monitoring this forum.  Please feel free to contact them and report what they say.

                             

                            The last thing I saw posted in May was that a fellow received one of the 4 phase versions of this board, and that it still throttled.

                             

                            My solution was to buy an ASRock board and put the Intel board (it's B3 replacement actually) up for sale on eBay.

                             

                            Regards

                            • 11. Re: Intel DP67DE Throttling
                              downset

                              i tried live chat on the intel website.

                              the support representative told me he could not help me as intel advices not to overclock even if your product alows it.

                               

                              when i asked why they sell a P67 MB and a K-series i5 if overclocking is not an option he send me the link to the following pdf:

                              http://downloadmirror.intel.com/19482/eng/DP67BG_PerfTuningGuide01.pdf

                               

                              that was all he could do.

                               

                              if you think about it a P67 limits fuctionality (can't use the build in GPU) and is more expensive

                              the K-series have no extra value on a P67 board and are more expensive

                              so you pay extra twice and have gained nothing, the prommised option to overclock (the sole reason for P67) is not supported and doesn't work

                              i can't understand why they even put this MB in the market

                               

                              probably will have to sell my MB too and get an other brand

                              • 12. Re: Intel DP67DE Throttling
                                Ready_Eddie

                                Intel advised not to overclock?  Therefore they can't talk to you?  That is really, really lame.

                                 

                                Here is the real story is from their site:

                                "Built-in processor frequency and memory frequency tuning
                                The Intel Desktop Board DP67DE is equipped with built-in capability to adjust processor frequency and memory frequency for Intel® Core™ processors with tuning capability to maximize computing performance as determined by the user (see note 3)."

                                Note 3: "Intel does not warrant against use of its processors and desktop boards outside of specification. Intel recommends users to ensure sufficient airflow and cooling of the system when enabling the processor and memory frequency tuning feature."

                                 

                                Lame.  You are a user yes?  You have insured sufficient airflow and cooling, yes?  If you dig deep enough, you can argue you are within specification by virtue of having done a proper thermal design (see their docs on case thermal requirements).  If you make a mistake or your risky cooling fails, I can understand that they don't want to replace fried CPUs, but to not talk to you about it not working as advertised???

                                 

                                Lame.

                                 

                                BTW, the DP67BG board has an entirely different voltage regulator design, and the forums say it OCs.  The ATX board corresponding to the DP67DE is the DP67BA, which has the same OC issues as the DP67DE.

                                 

                                Although I wonder what the "Intel Extreme Tuning Utility" for the DP67BG would do to the DP67DE.  Have you tried it?  (I don't have my DP67DE installed anymore, want to buy it? ;-)

                                 

                                Regards

                                • 13. Re: Intel DP67DE Throttling
                                  parsec

                                  "... the prommised option to overclock (the sole reason for P67) is not supported and doesn't work..."

                                   

                                  You should say that unlimited time over clocking via Turbo 2.0 multipliers does not work on Intel P67mother boards.  On my ASUS P67 mother board it does.  Of course, having a time limit is not expected when over clocking, but that seems to be the rule for Turbo 2.0.  The rules can be broken. but how do we know whether or not they will be?

                                   

                                  Intel does offer the Extreme Tuning Utility for the DP67BG, but only for that P67 board.  Perhaps unlimited-time over clocking is only available on that board.  The readme file for ETU still states that Intel guarantees their processors only up to their rated specs.  I'm not really surprised that Intel does not allow unlimited time Turbo multiplier over clocking on their non-extreme mother boards.  They have always been conservative regarding the usage and specs of their products.  For instance, the maximum rated memory speed for an i7-2600k is still only 1333, while memory and other mother board manufactures sell products that work well beyond that speed.

                                   

                                  It's pretty obvious that Intel has locked down the unlimited time Turbo over clocking on your boards, and it would have been nice to know that before purchasing them.  The info page for my ASUS P67 board does not state anywhere that over clocking is unlimited or even that it is available as an option.

                                  • 14. Re: Intel DP67DE Throttling
                                    downset

                                    the "Intel Extreme Tuning Utility" does not work on the pg67de MB

                                     

                                    very dissapointed, my board is for sale, i doubt i will even be able to sell it it is not very popular

                                     

                                    my girlfriend her ASUS P67-M is running 4.4ghz with the auto overclock function, without any tweaking with all power save fuctions enabled and it was just as expensive as the intel board...

                                     

                                     

                                     

                                    thanks for the replies and info eddy and parsec

                                    1 2 Previous Next