Why is my 2500k throttling?

defaultluser

[H]F Junkie
Joined
Jan 14, 2006
Messages
14,398
My system:

Intel DP67BA with latest BIOS
2500k, base clock at 33, but turbo set to 40-43, auto voltage, low power state activated.
Turbo boost short power max = 200w
Turbo boost sustained power max = 200w
Windows 7 64 with power plan set to "balanced"

So I've recently been playing around with Turbo Boost overclocking (the only option supported by my cheap motherboard), and checking out how my system operates using the Intel Extreme Tuning Utility.

I was a bit disappointed to discover that Turbo Boost stops after a very short time even though I keep my CPU very cool. But I was even more *pissed* to find that when running Prime 95 torture test with 4 threads, my CPU at stock 3.3 GHz was regularly throttled to 1.6 GHz!

Everything I've read says that if you don't exceed TDP, and you keep the processor significantly below TJMax, it should never throttle. But I'm not seeing that here.

It cools down a bit (from low 60s down to 50s) and after a minute or so it resumes full speed (3.3 GHz). The temperature stabilizes again at 60, after which it runs for a couple of minutes, and then it repeats.

Included is a screenshot of Real Temp, to show I'm nowhere near TJMax, so I have no damn clue why my system is throttling.

LisOZ1K.jpg


I knew when I bought this Intel board that it would be no overclocker's dream, but I seriously expected it to be able to supply rated TDP. This processor is not even drawing 70w at 3.3 GHz according to Real Temp.

Is there some Windows configuration option I'm missing? Is there a temperature I could be missing? Any help here would be appreciated.

I'll go take a closer look at my BIOS options and see if there's anything obvious relating to current/power/temperatures that I missed.

EDIT: tried High Performance power option in Windows. All that got me was a higher idle clock - still throttles to 1.6 GHz after I run Prime 95!
 
Last edited:
Because Intel designed the DP67BA that way. You can't fix it. Intel reserves permanent overclocking for its Extreme motherboards; the only Extreme motherboard in the P67 line is the DP67BG.

I fought with my Intel DP67DE (so similar it uses the same BIOS as the DP67BA) for months, and never could make it keep an overclock, so I passed it on to my father-in-law and got an ASUS P8P67 Pro 3.1 that OCs with a single click and keeps it. I don't know if a P67 board would be the right choice anymore for a 2500K.

For more info, see the thread I made about Intel P67 motherboards, particularly the final post:

http://hardforum.com/showthread.php?t=1574834
 
Because Intel designed the DP67BA that way. You can't fix it. Intel reserves permanent overclocking for its Extreme motherboards; the only Extreme motherboard in the P67 line is the DP67BG.

I fought with my Intel DP67DE (so similar it uses the same BIOS as the DP67BA) for months, and never could make it keep an overclock, so I passed it on to my father-in-law and got an ASUS P8P67 Pro 3.1 that OCs with a single click and keeps it. I don't know if a P67 board would be the right choice anymore for a 2500K.

For more info, see the thread I made about Intel P67 motherboards, particularly the final post:

http://hardforum.com/showthread.php?t=1574834

Thank you so very much! I didn't realize we had a thread on these craptastic boards. Here is the best post:

If you continue to stress the CPU while in the 33 ratio state, you will sometimes see it drop to a ratio of 16, especially if stressing with OCCT/Linpack instead of Prime95. When this occurs, bits 2 and 3 of the IA32_THERM_STATUS MSR show as set, indicating "PROCHOT# or FORCEPR# has been asserted by another agent."

I don’t think the CPU is doing it, rather it is the (under designed) motherboard VR 3 phase circuitry. I don't know how it implements the 33 ratio, but the 16 is when it cries for help from the CPU via this signal.

Using HW monitor, I note that the drop to 33 occurs whenever the CPU power goes a little above 90 watts, even when the bios power settings are changed (they appear in the RAPL MSRs, but these move the CPU logic limits, not the motherboard VR). This should not be happening this low, as this CPU has a TDP of 95 watts. If you have a minimal stock cooler, the temp and/or current should be approaching their limits at the TDP wattage by design, not 5% under it.

And I see the exact same behavior, despite my excellent cooling. Even at stock BIOS settings (NO OVERCLOCK) I occasionally see it throttle to 1.6! Well, that's the last time I bother purchasing something Intel-branded that is not a CPU.

Guess I'll look around for a z77 motherboard then! At least stock is discounted to move right now :D
 
When fully loaded, if your CPU initially runs at a multiplier higher than 33 but quickly drops down to 33, that usually means you are reaching the Turbo TDP limit for your CPU. Some manufacturers set the Turbo TDP value extremely low and on your board it might be set to 60 watts. If you don't have an option in the bios to adjust this then try running ThrottleStop and click on the TPL button. If the manufacturer leaves this register unlocked then you will be able to adjust this value much higher with ThrottleStop. Sadly, this feature is locked by the bios on many motherboards. If ThrottleStop is grayed out and does not let you adjust this then that is what is going on and the only way to unlock this feature is with a modified bios.

The next problem is when it drops down to 1600 MHz. Intel CPUs have a feature called BD PROCHOT or bi-directional processor hot. This allows other items on a motherboard to send a signal to the CPU which tricks the CPU into thinking it is too hot. The result is that it will immediately start thermal throttling and drop down to 1600 MHz because it thinks that it is over heating. It is usually the voltage regulator that sends this signal but it can be anything on your motherboard. ThrottleStop also has another feature that lets you turn off the BD PROCHOT signal path so the CPU will ignore this signal and should continue to run at its default speed of 3300 MHz.

Whether BD PROCHOT is checked or not in ThrottleStop, your CPU will still throttle if it ever reaches the thermal throttling temperature of 98C. There is no way to turn that off so your CPU will still be safe from over heating.

ThrottleStop 5.00
http://www.techinferno.com/downloads/

Intel might be trying to protect some under designed component or two on your motherboard. Using ThrottleStop to bypass these safety features is at your own risk.

Here is an updated version of RealTemp with a few more features for the Sandy and Ivy Bridge CPUs.

RealTemp T|I Edition
http://www.techinferno.com/2012/11/23/realtemp-techinferno-edition-now-available/
 
Thanks so much unclewebb!

ThrottleStop was able to disable the external BD PROCHOT, and my CPU has stopped throttling. This had no effect on my turbo dropoff, but that's okay, it was what I expected.

Although my Turbo TDP can be adjusted in the BIOS, it has no effect. I cannot increase the package current limit from 97A, and this is likely the culprit. So perhaps I could get around this with some undervolting? Might be worth a tweak session :D

Anyway, since I now have a solution to my largest complaint, I think I will pass on the replacement motherboard. But thanks for the information both of you!
 
Are the power limits in the TPL window locked in ThrottleStop? Can you post a screen shot of that window? Is it set to 60 watts like it seems to be?

The 97 Amp limit is usually adequate and doesn't need to be increased. I am pretty sure it's the TDP Power limit holding you back now.
 
Are the power limits in the TPL window locked in ThrottleStop? Can you post a screen shot of that window? Is it set to 60 watts like it seems to be?

The 97 Amp limit is usually adequate and doesn't need to be increased. I am pretty sure it's the TDP Power limit holding you back now.

The TDP is set to 200w. It was originally set to 95 (sustained) and 117 (peak), and I was able to alter these values in the BIOS. I figure 200 is plenty.

N2N2GIX.jpg


I cannot alter the package current limit, so are you certain that's not the culprit?
 
My P8Z77 board sets the power limits to 255 255 to make sure they don't get in the way of full turbo boost. Your limits of 200 200 and 97 amps should not be killing all turbo boost. Hopefully Disable Turbo is NOT checked in RealTemp. Can you post a screen shot of ThrottleStop when doing some full load testing?

When you change the Package Current Limit in ThrottleStop and then click on OK, this should change. When you re-open the Power Limits window does it show the new value? The Current Limit Register is not locked so in theory you should be able to change this.

What does the ThrottleStop TRL window show? Does it show the same turbo multipliers as your bios is showing?

If everything looks OK, can you try running a ThrottleStop log file when testing. You can copy and paste the data to www.pastebin.com
 
Last edited:
My P8Z77 board sets the power limits to 255 255 to make sure they don't get in the way of full turbo boost. Your limits of 200 200 and 97 amps should not be killing all turbo boost. Hopefully Disable Turbo is NOT checked in RealTemp. Can you post a screen shot of ThrottleStop when doing some full load testing?

Disable Turbo is not checked in either RealTemp or ThrotttleStop.

VN1QfZc.jpg


When you change the Package Current Limit in ThrottleStop and then click on OK, this should change. When you re-open the Power Limits window does it show the new value? The Current Limit Register is not locked so in theory you should be able to change this.

I can set the field, but the apply button is not available (grayed-out), so I don't think it's set at any point. The setting is also not saved (resets to 97 every time I reboot), which makes me suspect that it's not alterable. I set it to 200 just now, clicked ok, and started another heavy processing test. Turbo dropped after a few minutes like it has before.

What does the ThrottleStop TRL window show? Does it show the same turbo multipliers as your bios is showing?

It matches my BIOS exactly. The turbo is definitely happening, just not sustained.

BFirX3H.jpg


If everything looks OK, can you try running a ThrottleStop log file when testing. You can copy and paste the data to www.pastebin.com

As you can see I have enabled this option in the screenshot above, but I don't know where it is saved. The is no obvious file in the folder containing the executable. The only file modified today by ThrottleStop is the ini file (I did a windows search).

I see no options to set the log location. Do I have to activate the "saving" of the file by checking and unchecking the box?
 
Did you unzip ThrottleStop into its own folder somewhere convenient like your desktop? When I click on the Log File option, it will create a folder in the ThrottleStop directory called Logs and inside there will be a log file with today's date on it. I think this log file is updated once per minute and if you exit ThrottleStop, this directory and log file should be created immediately so all data will be saved. If you run ThrottleStop from within a zipped file or if the ThrottleStop folder is located on your hard drive where you don't have full read / write privileges then that could block ThrottleStop from creating the log file or saving settings to the INI file.

The current limit should get reset to 97 every time you reboot. ThrottleStop only makes temporary changes and does not change anything in your bios. After you change this value in ThrottleStop and click on OK and then immediately re-open the Power Limits window, does ThrottleStop show the new value? If it does then it is changing this value within the CPU. Increasing this value is only going to increase your full load multiplier if this is the thing that is holding you back. If your CPU is running under the 97 amp limit when fully loaded then raising this limit even higher is not going to change anything. Your screen shot shows 72.1 W and 1.2860 V so that implies only 56 Amps ( 72.1 / 1.2860)

Does CPU-Z show your core voltage or does it only show VID voltage? If you can reduce your core voltage by setting it manually in the bios then this can prolong the amount of time Turbo Boost will last for. The AUTO voltage bios setting when overclocking often times results in far more voltage than you need which increases power consumption and creates extra heat.

Have you tried clicking on the Set Multiplier box and adjusting that lower and then higher until it says Turbo while fully loaded?

There are a variety of ways for Intel to throttle or limit Turbo Boost on a CPU so I am just trying to understand what method Intel is using on this board. When starting a full load test do you initially get the full 40 multiplier and then does it slowly step down to 35 or 33? A log file with the More Data option selected would show that and will show how long Turbo Boost is working for. Your screen shot above shows you are getting some Turbo Boost but just not as much as you should be getting. Instead of full load, if you run 2 threads of the TS Bench, does your multiplier stay over 40?

When testing with ThrottleStop, don't run XTU at the same time. Both programs are writing data to the same CPU registers so it is possible for these 2 programs to interfere with each other.

If you don't make any progress with the Set Multiplier adjustment, can you switch to the Windows High Performance Profile and then open up a command window with Admin privileges. In Windows 7 I think you can type in cmd in the Start menu - Search box to find this and then right click on thecmd program and choose the Run As Administrator option. In this command window type in

powercfg -qh >C:\power.txt

This will query your power profile including the hidden values within this profile and it will output the data to a file called power.txt in your main C: directory.

You can copy and paste the info in the power.txt file to www.pastebin.com

The mobile UM CPUs can have Turbo Boost limited by a secret Windows power setting but I don't think the desktop processors are being limited using this method. Always good to check just to make sure.
 
Last edited:
EDIT: I stopped running XTU for the following

Okay, here are some of my observations:

1. The turbo boost amount varies based upon the difficulty of the load (in terms of processor resource utilization). Prime 95 torture test (benchmark, strains AVX units) only goes up to 35x. Normal encoding of a video with Handbrake (4 threads, 90-ish utilization) gets 37x. Dual-core loads are not as power-limited and can do 40x. Single core can do 42x. But again these all end after a period of time, and won't restart the turbo unless the system sits idle for specified amount of time (seems to be targeting a certain power duty cycle).

2. I was able to get ThrottleStop logging to work by running it as Admin (forgot that apps can ALTER existing files, but require ADMIN to create). Here is a log of two runs of Prime 95 4 thread:

http://pastebin.com/embed_iframe.php?i=bVfPATQk

first run = balanced performance Windows profile
first run = high performance Windows profile

In both cases I twiddled with the set multiplier option as you suggested, and it made no improvement.

Based upon that detailed log I have to agree that it's not the platform current that's responsible. That's way less than the peak allowed current.

3. Here are my Windows power settings:

http://pastebin.com/embed_iframe.php?i=mMdK5Kmc
 
Last edited:
Code:
   DATE       TIME    MULTI   C0%   CKMOD  CHIPM   BAT_mW  DTS   GPU     VID   POWER
2013-07-05  17:53:46  16.00    2.9  100.0  100.0        0   70    26   0.9757    6.6
2013-07-05  17:53:46  16.00    4.6  100.0  100.0        0   69    26   1.0007    5.8
2013-07-05  17:53:46  16.00    9.5  100.0  100.0        0   69    26   0.9757    7.4
2013-07-05  17:53:46  16.00    4.9  100.0  100.0        0   69    26   0.9757    7.0
2013-07-05  17:53:46  16.00    3.4  100.0  100.0        0   70    26   0.9757    6.1
2013-07-05  17:53:47  31.00   60.6  100.0  100.0        0   58    26   1.3711   34.3  <------VID Peak
2013-07-05  17:53:47  38.93   99.5  100.0  100.0        0   56    26   1.3711   62.6
2013-07-05  17:53:47  38.60   99.4  100.0  100.0        0   55    26   1.2810   66.6
2013-07-05  17:53:47  35.33   99.8  100.0  100.0        0   54    26   1.2860   73.0
2013-07-05  17:53:47  35.02  100.0  100.0  100.0        0   54    26   1.2910   71.9
2013-07-05  17:53:47  35.00  100.0  100.0  100.0        0   53    26   1.2860   72.2
2013-07-05  17:53:47  35.00  100.0  100.0  100.0        0   53    26   1.2910   68.1
2013-07-05  17:53:47  35.00  100.0  100.0  100.0        0   53    26   1.2860   75.9

Can you go into the bios and set the CPU core voltage manually? If you look at the log file you can see that as soon as the CPU starts to throttle up to full speed, it requests 1.3711 volts. This might be triggering something in the voltage regulator which results in immediate throttling and prevents the CPU from getting up to and maintaining its full overclocked speed. I believe that this throttling is controlled by the chipset or by the CPU.

If you were able to manually set the core voltage to about 1.20 to 1.25 volts, that might free up some head room and allow the full load multiplier to go a little higher.

I don't know if there is a simple solution for this problem but could you run the MSR Finder tool? Fully load your CPU like above and when it throttles back to the 35 or 33 multiplier, can you click on the Snapshot button and get a snapshot of your CPU registers. The log file will be hiding in a similar location so run this as as Admin too.

MSR Finder
http://www.mediafire.com/?b2tg2yfjt27csy3

There might be something in there that gives me a clue to the throttling method that Intel is using but it is likely something in the chipset and won't be easy to solve because Intel does a good job of hiding the full documentation for their chipsets. When doing this test, don't have XTU or ThrottleStop running. On an Asus or Gigabyte board with a $30 heatsink, you could probably run fully loaded with a 45 multiplier instead of the 35 or 33 that this Intel board is limiting you too.

Edit: I haven't found anything obvious in your Windows power profile that is causing this.
 
Last edited:
Back
Top