New Intel Core 2 Multiplier / Core i7 Turbo Testing Tool

U

unclewebb

[H]ard|Gawd
Joined
Jun 21, 2006
Messages
1,807
http://www.fileden.com/files/2008/3/3/1794507/Turbo.zip

This new tool follows the method outlined in the Intel Turbo white paper for Core i7 and at full load is the most accurate method to calculate the CPU multiplier.

The reason this is important is that some motherboards at full load running high stress programs like LinX will start to cycle the Turbo (+1 multiplier) feature on and off rapidly. This new i7 Turbo tool should catch that situation sooner than any other software presently available. At full load, as soon as the calculated multiplier starts to drop, Turbo mode is starting to cycle on and off.

Most software only reads a single multiplier value from a model specific register (MSR) within the CPU. This is not the preferred method.

If you notice a lot of drift in the calculated multiplier at idle, that's a sign that your bios settings for EIST and C1E do not agree with your Windows Power Options settings. In Vista you need to go into the Control Panel -> Power Options and set your Minimum processor state to the appropriate percentage to either fully enable or fully disable your energy saving features. Try using 50% or 100% for this setting. You may find that your bios version does not properly handle turning off some things like EIST if that is your goal.

poweroptions.png


Windows XP is a little different. To fully enable EIST and to get your computer to use a reduced multiplier at idle, (6.0 Core 2 / 12.0 Core i7), you might need to go into the Power Options and set the Power Scheme to Portable/Laptop computer.

If you have a Core i7 and enable Turbo mode in the bios, you can also use this tool to toggle Turbo mode on and off from within Windows.

This tool was designed for the Core i7 but I decided to support Core 2 processors as well. The method outlined in the Intel Turbo white paper also applies to the Core 2 CPUs that I've tested.

Here's a test I came up with to check how software deals with odd ball multipliers and situations that you don't read about in any book.

A Quad CPU has two Dual Core CPUs inside of it. Using MSR 0x199 you can trick it so that one of the Dual Cores uses one multiplier while the other Dual Core uses a completely different multiplier, even at full load running Prime95.

For my example I set core0 and core 1 to use a 9.0X multi and core2 and core3 to use a 6.0X multi. Right click on CPU-Z and you can get it to read any individual core.

CPU-Z correctly shows one side of the CPU running at 3000 MHz and the other side at 2000 MHz. RealTemp averages those two values and reports 2500 MHz. RealTemp also averages the 9.0 and 6.0 multipliers and displays 7.5.

The average multi read from the MSR is shown as 7.500 and the calculated multi also shows the correct 7.5000. Not a lot of tools can handle this situation correctly. :D

trickt.png


At idle, calculating the multiplier with Intel's method can result in the occasional random value but if your system is set up right, the amount of multiplier drift should be minimal.

At full load running Prime Small FFTs, the calculated multiplier should be steady as a rock. Any drift at full load is the first sign of a problem.

If you learn something new about your CPU using this tool then post your results. This is the first general release of i7 Turbo so if there are any problems or anything you don't understand, let me know. Thanks.
 
Last edited:
Here's a good example of a problem that this new tool was able to solve.

A user contacted me yesterday to tell me that RealTemp was reporting his multiplier for his Q9450 at only 7.0 at full load when CPU-Z and other tools were reporting it at 8.0.

Upon further investigation he discovered that there is a problem with his old bios not correctly supporting his new 45nm Quad. At full load, core 0 and core 1 were using the correct 8.0X multiplier while core 2 and core 3 were only using a 6.0X multiplier.
The overall average reported by RealTemp as well as the Turbo tool is the correct 7.0.

mystats3.jpg
 
turbo5b.jpg


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

I just wanted to let users know that this handy tool for testing Turbo throttling has been updated.

The Asus P6T Deluxe is one of the boards that gives users no way to turn off TDP based Turbo throttling in the bios.

On a Core i7-920 with Turbo enabled, the Calculated Multiplier and the MSR Multiplier should be locked on 21.000 when the CPU is fully loaded. Anything less at full load is a warning sign of the +1.0 Turbo cycling on and off.
 
Interesting, didn't know that something like the example in the 2nd post could happen with C2Qs... Not terribly useful for my desktop (rather just have it going 100% all the time and simply use S3/suspend if I wanna save power) but this'll be nice to know/have once Westmere is out. :D
 
The i7 Turbo tool is mostly used to test for Turbo throttling. This usually happens somewhere around 4GHz and 1.40 volts. High core temperatures around 85C can also trigger Turbo throttling on many Asus boards that give you no way to turn TDP based Turbo throttling off.

That might not be such a bad thing but it would still be nice for Asus to give users the option to disable this feature.
 
Last edited:
great tool, any way to check which boards have the option "TDP based Turbo throttling off". I just ordered gigabyte ex58 ud4p and it's not opened yet..
 
I'm keeping a list and so far only the Asus P6T line is guilty of not having this option.

My friend rge had this problem on his Gigabyte Extreme when he first got it with an early F3 bios but they fixed this problem a while ago so your UD4P should be fine.

With water or a top notch air cooler and a D0 chip, 4 GHz is not that hard to reach so the Asus TDP limit is becoming an issue for some users. I'm not sure what this option is labeled in the Gigabyte bios but you'll figure it out. At least you've got a nice tool now to check for this issue. :D

Edit: From what I've heard, Gigabyte doesn't have a bios option to turn Turbo throttling off because it is always turned off like it should be on any enthusiast board.
 
Last edited:
unclewebb said:
turbo5b.jpg


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

I just wanted to let users know that this handy tool for testing Turbo throttling has been updated.

The Asus P6T Deluxe is one of the boards that gives users no way to turn off TDP based Turbo throttling in the bios.

On a Core i7-920 with Turbo enabled, the Calculated Multiplier and the MSR Multiplier should be locked on 21.000 when the CPU is fully loaded. Anything less at full load is a warning sign of the +1.0 Turbo cycling on and off.
Click to expand...


I haven't seen my cpu vary from it's x21 @ 200 under 100% prime, intel burn test, memtest, etc. And this is on a Asus P6T deluxe v2 latest bios. I've watched for it too but now you got me worried. I mean I know it's only 200mhz but damnit, I want my 200mhz on at all times! :p But like Impulse, I turn off all power managment cpu features though so maybe that's it.
 
KVW: It's all about core voltage. If your sig is accurate, you're able to run 4.2 GHz with out needing excessive core voltage. That's a good thing.

The problem typically occurs at 4 GHz or above when the core voltage is up near 1.40 volts. You still have some head room so keep overclocking. Let me know when that Calculated Multiplier starts to fluctuate when running LinX. :)
 
http://www.fileden.com/files/2008/3/3/1794507/Turbo.zip

-Fixed incorrect multiplier result when only 1 core was active. Was showing +1 when it should have been showing a +2 Turbo boost. Thanks rge for finding and testing that out.

-added an INI file to remember the previous Load Filter used.

-added the CPU name string.

-added the ability to toggle C1E but only for Core 2 users so far.

-added an INI option called Delay=0 which controls how long of a delay in milliseconds there is before the multiplier is read from the MSR. The range is 0 to 10 ms.

When reading the multiplier from a CPU, the act of reading this information momentarily loads the CPU which can increase the value of the multiplier returned. Most software that reports the multiplier will use a tiny delay just before reading the multiplier to give the CPU a chance to settle down.

Delay=0 will result in a more responsive and accurate MSR Multiplier while Delay=1 will give the CPU a millisecond to settle down and at idle will result in an MSR Multiplier that doesn't float around as much. Default is Delay=1.
 
Last edited:
unclewebb said:
http://www.fileden.com/files/2008/3/3/1794507/Turbo.zip

I've updated this tool once or twice and have included a separate program called MultiBoost to help with the DFI 12X multiplier Windows 7 bug.

Give this thread a bump if either tool is useful to you.
Click to expand...

Bump! Just bought a T3eH8 and am wondering if DFI is going to release a BIOS update to fix this Win7 bug. At least I'm not upgrading to Win7 until a couple months after it's finalized.
 
I'm not sure who is more to blame but I'm pretty sure there will be bios updates and there will be Windows 7 updates before it is released. Until then, the MultiBoost tool is proving to be a simple and effective fix for this problem. It's working so well that I plan to build its functionality into RealTemp in the near future. All you will need to do is add MultiBoost=1 to your RealTemp.ini file and you probably won't even ever notice the stuck at 12 multi problem.

Windows 7 has gone from build 7100 to beyond build 7200 so this problem might be fixed already.
 
unclewebb said:
KVW: It's all about core voltage. If your sig is accurate, you're able to run 4.2 GHz with out needing excessive core voltage. That's a good thing.

The problem typically occurs at 4 GHz or above when the core voltage is up near 1.40 volts. You still have some head room so keep overclocking. Let me know when that Calculated Multiplier starts to fluctuate when running LinX. :)
Click to expand...

I pushed it to 4.4ghz the other day (21x210), i was actually able to boot up windows at the same voltage, 1.25v but wasn't prime95 stable. To get prime95 stable, I had to push the voltage up to 1.325 but was hitting some seriously scary temps (90c) so I couldn't run prime for too long... However my turbo still didn't fluctuate, at least not according to cpu-z. Does LinX really push the CPU that much harder? Also does cpu-z not catch the fluctuations? The reason why I'm curious now is maybe it's the cause of stuttering/hitching I see in my oblivion that I don't remember seeing with my 4.4ghz e8400 wolfdale. :confused:
 
A $15 Kill-a-Watt meter was a good investment. It clearly shows that LinX takes significantly more watts to run than Prime Small FFTs. More watts coming out of your wall equals more watts going directly to your CPU and more heat that needs to be dissipated.

Also does cpu-z not catch the fluctuations?
Click to expand...

CPU-Z will catch Turbo throttling when it is severe enough. When it first starts to happen, it might initially miss it.

Here's a quote from 003 on the XS forums:

Notice how i7 Turbo is reporting a multiplier of 20.875 (1/8 threads running at 20x), but all other programs still report 21x.
Click to expand...

When he pushed his CPU a little harder, all software showed a 20X multi but i7 Turbo was the first program to show him that there was a problem.
 
Here's a good example of what Turbo throttling looks like when running Prime95 on an Asus P6T.
Not all software detects this problem when it first starts to happen. Thanks glide1.

throttle.png
 
I transferred the i7 Turbo code back into RealTemp 3.30 to help catch Turbo throttling as soon as it begins.

Here is some testing done by rge at XtremeSystems
------------------------------------------------------------------------------

Early throttling caught by RT, not yet by cpuz down to 23.8 from 24. My i950 starts to throttle almost right at 100W tdp.

rtthrottles238pst.jpg


Again throttling to 23.6, not yet caught by cpuz.

rtthrottles236pst.jpg


Complete throttling/turbo disengaged by 120W, now shown by both cpuz and RT

rtthrottles23cpuz23pst.jpg
 
Last edited:
There is now a limited edition bios out for the Asus P6T Deluxe v1 that automatically disables the TDP/TDC based Turbo throttling. It has been tested and cross flashed onto a v2 board so far with excellent results. No more Turbo throttling.

Depending on testing, Asus might be releasing this feature for other versions of the P6T as well but all will be limited edition bios versions. They won't add this feature to the generally available versions that you can download from their website.

If you have a P6T and would like to bypass the TDP/TDC safety check then send me a PM and I will help you get this new bios.
 
flirting with disaster? I also noticed my temps got very close to 90*C :S

tjmax.png
 
I got those temps at full load @3.9GHz with 30C ambient. Not really a problem but I definitely didn't wanna go any higher until the ambients dropped, like now. :)
 
DragonQ said:
I got those temps at full load @3.9GHz with 30C ambient. Not really a problem but I definitely didn't wanna go any higher until the ambients dropped, like now. :)
Click to expand...

hmmm i have the same but at stock speeds, maybe its the crappy intel cooler/paste + my enclosed case
 
86 5.0L said:
hmmm i have the same but at stock speeds, maybe its the crappy intel cooler/paste + my enclosed case
Click to expand...

+ dust in the hs/f? always something to be careful of.

yeah thats it, Bloomfield is a huge die that consumes lots of power. That cooler is not a huge unit that takes a large change in surface temperature to equalize with the thermal output of your processor.

NCIX regularly puts the Arctic cooling Freezer 7 pro/64 on sale for $20. At that price its a great unit, able to quietly keep a 130W TDP below a delta of 40C.
 
86 5.0L: It looks like your board supports a +2 turbo boost when lightly loaded. Many boards only have a maximum of +1 turbo boost as soon as you start overclocking the FSB. The Intel DX58SO board can access a +2 when overclocking the FSB but it's the only board that I know of that can do this.

Try running the other program in the download called i7 Turbo V8. It monitors the multiplier on each thread. When running a single threaded app like Super PI, you'll probably see a multiplier on one thread at or near 22.0. You usually need to have C3/C6 enabled in the bios.

Intel says the +2 turbo is only available when 1 of the 4 cores is active. It was thought that you would need to disable 3 cores in the bios which kind of defeats the purpose but it's also possible if 3 cores go inactive because they've entered the C3/C6 sleep state.

If any of the 3 cores that are inactive wake up to process background tasks and become active, the maximum turbo boost will drop down to +1. The turbo can constantly cycle between +1 and +2 hundreds of times a second depending on load so you need a program like i7 Turbo V8 to give you an accurate look at the average Calculated Multiplier.

If you can, post a screen shot of i7 Turbo V8 while running a SuperPI benchmark. It should be interesting on your board. The one thread running SuperPI should stand out with the highest average multiplier.
 
Last edited:
MrWizard6600 said:
+ dust in the hs/f? always something to be careful of.

yeah thats it, Bloomfield is a huge die that consumes lots of power. That cooler is not a huge unit that takes a large change in surface temperature to equalize with the thermal output of your processor.

NCIX regularly puts the Arctic cooling Freezer 7 pro/64 on sale for $20. At that price its a great unit, able to quietly keep a 130W TDP below a delta of 40C.
Click to expand...

Its a new build, so no dust, I was thinking of going with a CoolerMaster V8

unclewebb said:
86 5.0L: It looks like your board supports a +2 turbo boost when lightly loaded. Many boards only have a maximum of +1 turbo boost as soon as you start overclocking the FSB. The Intel DX58SO board can access a +2 when overclocking the FSB but it's the only board that I know of that can do this.

Try running the other program in the download called i7 Turbo V8. It monitors the multiplier on each thread. When running a single threaded app like Super PI, you'll probably see a multiplier on one thread at or near 22.0. You usually need to have C3/C6 enabled in the bios.

Intel says the +2 turbo is only available when 1 of the 4 cores is active. It was thought that you would need to disable 3 cores in the bios which kind of defeats the purpose but it's also possible if 3 cores go inactive because they've entered the C3/C6 sleep state.

If any of the 3 cores that are inactive wake up to process background tasks and become active, the maximum turbo boost will drop down to +1. The turbo can constantly cycle between +1 and +2 hundreds of times a second depending on load so you need a program like i7 Turbo V8 to give you an accurate look at the average Calculated Multiplier.

If you can, post a screen shot of i7 Turbo V8 while running a SuperPI benchmark. It should be interesting on your board. The one thread running SuperPI should stand out with the highest average multiplier.
Click to expand...

Heres my screenshots,

disable 3 cores, no HT


4 cores, no HT,


4 cores, + HT, notice the 3rd cores multi?
 
unclewebb said:
Many boards only have a maximum of +1 turbo boost as soon as you start overclocking the FSB. The Intel DX58SO board can access a +2 when overclocking the FSB but it's the only board that I know of that can do this.
Click to expand...

My Asus P6T Deluxe V2 goes up to 22x on one or two cores when only one or two cores are loaded, e.g. when playing a game that only uses one core.
 
86 5.0L: Interesting numbers. In your first pic you are definitely getting the full +2 turbo. In the second picture you are only getting a +1 turbo. In your third picture you are getting a bit of both. The turbo is continuously cycling from +1 to +2. The thread doing most of the work with the highest load (62.47%), is also the same thread that is reporting the highest average multiplier (21.485). The average multiplier for that core means the +2 Turbo is being maintained for 48.5% of the time. The other 51.5% of the time it is being limited to +1.

The more you are able to minimize the background activity on your computer, the more time it will be able to spend at +2.

You might also want to try a program like HyperPI.

http://www.virgilioborges.com.br/virgilioborges/hyperpi/

It's a front end for Super PI that gives you the option to run multiple threads of Super PI. The biggest difference between it and Super PI is that Super PI automatically shifts the load around to multiple threads. Hyper PI when running a single Super PI calculation will lock it to only one thread. This might allow your CPU to maintain a +2 turbo for a bigger percentage of time.

Do you have C3/C6 enabled in the bios? Motherboards are interpreting the Intel specs a little differently from one board to the next and even one bios version to the next. When C3/C6 is enabled and you force the load to a single thread, you might see the individual multipliers drop down between 12 and 15 on the inactive cores when running a single threaded app. You can also go into your Control Panel -> Power Options and play around with the Minimum processor state when testing.

My Asus P6T Deluxe V2 goes up to 22x on one or two cores when only one or two cores are loaded
Click to expand...

What software are you using to monitor your multiplier? With a Core i7-920, the only way to get a 22X multiplier is to either disable 3 cores in the bios or you can get a 22X multiplier on one core if the other 3 cores are inactive because they are in either the C3 or C6 sleep state. Anything beyond that is typically software reporting error. As soon as two cores are active, the multiplier instantly drops down to +1. This happens so quickly that most software misreports it. The only accurate way is to use the Intel recommended high performance timers and calculate an average over a 1 second interval.

Here's a quote from the Intel Turbo White Paper:

"The number of active cores at any given instant dictates the upper limit of Intel® Turbo Boost technology. For this discussion, a core is considered ‘active’ if it is in the “C0” or “C1” state; cores in the “C3” or “C6” state are considered ‘inactive’."

Here are the specs for the 920:

"1/1/1/2 are the available frequency bins (+133 MHz)of Intel® Turbo Boost Technology that are available for 4, 3, 2, 1 active cores respectively."

The Xeon X5570 is more generous with available bins of Turbo Boost:

"2/2/3/3 are the available frequency bins (+133 MHz)of Intel® Turbo Boost Technology that are available for 4, 3, 2, 1 active cores respectively."

With Core i5, Intel is going to be giving users more bins of turbo boost as long as motherboard manufacturers implement their design properly.
 
Last edited:
unclewebb said:
86 5.0L: Interesting numbers. In your first pic you are definitely getting the full +2 turbo. In the second picture you are only getting a +1 turbo. In your third picture you are getting a bit of both. The turbo is continuously cycling from +1 to +2. The thread doing most of the work with the highest load (62.47%), is also the same thread that is reporting the highest average multiplier (21.485). The average multiplier for that core means the +2 Turbo is being maintained for 48.5% of the time. The other 51.5% of the time it is being limited to +1.

The more you are able to minimize the background activity on your computer, the more time it will be able to spend at +2.

Do you have C3/C6 enabled in the bios? Motherboards are interpreting the Intel specs a little differently from one board to the next and even one bios version to the next. When C3/C6 is enabled and you force the load to a single thread, you might see the individual multipliers drop down between 12 and 15 on the inactive cores when running a single threaded app. You can also go into your Control Panel -> Power Options and play around with the Minimum processor state when testing.
Click to expand...

my bad I forgot that info, C3 is enabled in the 1st and 2nd pic but not the third
 
unclewebb said:
With a Core i7-920, the only way to get a 22X multiplier is to either disable 3 cores in the bios or you can get a 22X multiplier on one core if the other 3 cores are inactive because they are in either the C3 or C6 sleep state. Anything beyond that is typically software reporting error. As soon as two cores are active, the multiplier instantly drops down to +1. This happens so quickly that most software misreports it. The only accurate way is to use the Intel recommended high performance timers and calculate an average over a 1 second interval.
Click to expand...

Well I have C-states enabled so that's fine.
 
I saw over at Xtreme that someone else had a dual gainestown setup as well..

I have 2 x X5550

I have noticed that if i load 16x threads, only one cpu with kick on the turbo mode.

The only way it seems to turbo on both is two load each cpu 50% ...

Temps are fine ... 55C @ full load

It is really weird ... Any Ideas?
 
Last edited:
I was testing out the new Asus P6T7 Supercomputer board with a Xeon W3520 and noticed with this tool that the multiplier was dropping down from 21x when my temps were reaching around 80C under load. The system was overclocked but voltages were not very high.
 
burningrave101: This is called turbo throttling and is a well known issue with the Asus P6T motherboards. Here is what Intel has to say in their November 2008 Turbo White Paper.

1.2 Dependencies / Algorithm

Intel® Turbo Boost technology core frequency upside availability is ultimately constrained by power delivery limits, but within those constraints, it is limited by the following factors:
• The estimated current consumption of the processor
• The estimated power consumption of the processor
• The temperature of the processor
----------------------------------------------------------------------------------------------

Some of the Gigabyte boards simply have this "feature" disabled. Other boards give you an option to enable or disable this in the bios. The Asus P6T series doesn't give you any option to control this. The P6T7 Supercomputer isn't so super when pushed hard.

Asus has finally started to listen to users who have been complaining about this for a while now. They have released two bios versions so far for the P6T Deluxe v1 which finally have turbo throttling disabled. You won't find these special bios versions on their website so if anyone needs it, let me know.

I've been told that they are also working on releasing some similar bios versions for 3 other P6T motherboards but I don't have any idea which ones yet.

Do me a favor and post a screen shot of CPU-Z and i7 Turbo when turbo throttling first starts to happen. CPU-Z ignores this problem when the multiplier first starts to cycle between 21 and 20 but by calculating the average multiplier the way Intel recommends, you can catch this problem as soon as it starts to happen.

EvilAlchemist: i7 Turbo does not yet support Gainestown. I might try to create a special version for you to help you out. I've been thinking about adding Gainestown support to RealTemp as well but my software testers haven't been dedicated enough. It's hard to develop for a platform that I don't own.
If your testing is accurate, it might be a bios issue.
 
Last edited:
http://www.fileden.com/files/2008/3/3/1794507/Turbo.zip

Just a minor update. I added initial support for Core i3 / i5 / new i7 and even i9 as long as hyper threading isn't enabled. :)

Intel documentation is doesn't exist at the moment so no guarantees that it will correctly identify the name of one of these new CPUs but the program does read the multiplier correctly. It will be fun to see +3 or +4 bins of Turbo boost some day soon. :D

i7 Turbo also works correctly on Core 2 Duo or Quads and will show your average multiplier more accurately than any other program.

The V8 has been dropped from the name and the original i7 Turbo has been retired so now there is only one version to maintain.

If you are still using RealTemp 3.00, update to RealTemp 3.30. This will improve the consistency of your i7 Turbo results since they both use the same internal timers. The RealTemp download contains both programs.

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

Here's a screen shot of rge running 8 hours of Prime without the calculated multiplier budging by 0.001 the whole time. Great overclock and temps.

rgeturbotest.jpg
 
Last edited:
i'm running an Asus P6T so i probably can't disable Turbo throttling.

which do you guys think would be better if I wanted to achieve a stable 3.8ghz?

1. Turbo off. BCLK=190, Mult=20x (190 * 20 = 3.8ghz)
2. Turbo on. BCLK=180, Mult=Auto (180 * 21 = 3.78ghz)

i'm trying to weigh the advantages/disadvantages of the Turbo feature
 
Well the advantage is not having to raise your BLCK as high, which will help with stability if your motherboard is not among the best and also should reduce your case temperatures slightly.
 
You must log in or register to reply here.
Back
Top