i7 Turbo - Tool to test for Turbo throttling

[Brama]

rge has been doing some testing and it looks like the newest version of i7 Turbo is going to show when one core jumps up to a +2 Turbo ratio even with 4 cores enabled. Stay tuned for some interesting results.

I am tuning up and down your frequency to get this info as soon as possible.

Very, very interesting and it probably could explain why sometimes I have 2 or 3 GFlops of difference using LinX with the same exact setup.

[unclewebb]

rge only found a +2 when he was not overclocking. As soon as he changed the FSB by 1 MHz, the +2 was gone. He proved that it does exist if 3 of your 4 cores are idle. If any one of the three cores is doing anything at all then the Turbo drops back to the regular +1 multi.

+2 exists but was completely useless on his Gigabyte board because it is disabled the moment you start overclocking. Most XS users have a tendency to do some overclocking. The ones that don't, have been strung up out back!

In a handful of hours I will release the V8 version that rge was using to do his testing. It also shows how much the other cores are able to idle down. With fewer background tasks, the single core that is doing all the work can spend more time at the +2 Turbo setting. It was interesting to finally see some real evidence of what's actually going on in Core i7 CPUs.

[unclewebb]

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

I decided what the hell. Why screw around all day while people are lining up waiting to do some testing.
Post some screenies if you learn anything new.

I'm just starting to understand what this tool is telling me. I believe it is accurate and I will explain it further for anyone that doesn't get it or doesn't agree with what they are seeing.

The download includes both versions. i7 Turbo V8 should show the calculated multiplier for all 8 threads if you have hyper threading enabled. When it shows one thread with an average multiplier at 14 and the other at 24 on the same core, I think that is a sign that the hyper thread has been put to sleep and has entered a C3 or C6 sleep state. You will need to enable these I think and probably lock your MHz to default in the bios to see the mythical +2 Turbo. Check out rge's posts on the previous page.

[rge]

Has anyone any info about turbo throttling on D0s.
I mean, when does it start (cpu voltage, cpu temp)?

It is completely governed by your mobo manufacturer bios implementation, not by cpu, the cpu simply has that info available for use or non use.

Your bios manufacturer can choose to ignore turbo limits and throttling does not occur, so you just get a permanent +1 multi with turbo enabled. Most mobo manufactures have either disabled turbo throttling, or given the user the option in the bios to turn it on or off.

Asus is the exception, many of their mobos/bioses essentially screw you, and your cpu will throttle when you need the power the most, ie under stress. I believe some Asus mobo bioses implement throttling at temps >85C or total watts around 100 to 120. But how it is implemented current, watts, temps is completely up to the manufacturers whims via bios.

[enkie]

new bios 0007 (beta) for the p6t deluxe that fixes the problem but still nothing for the P6t6 ws revolution was in spite of a topic on the asus forum.

[crashmax]

ASUS P6T7 WS SuperComputer BIOS 0210 (first revision) and Turbo throttling...

[xpatar]

I just crossflashed successfully from 0504 V2 to 0007 V1.
Turbo throttling has been resolved and everything is working like a charm.

Here a few pictures:


Multis don't change at full cpu load, pay attention at low cpu power consumption (unreal).

While @ idle:


Thanks Kevin

[unclewebb]

I just crossflashed successfully from 0504 V2 to 0007 V1.
Turbo throttling has been resolved and everything is working like a charm.

Stuff like that brings tears of joy to my eyes.
Your overclock looks much nicer now. Glad I could help.

Can someone explain the cross flashing procedure for the v2 boards? I'm sure there are a few users that would like to try this but are a little scared about screwing up their board or not being able to go back to v2 if they decide they don't like bios 0007. Do you lose any functionality of a v1 board after a cross flash?

[xpatar]

So far I haven't notice any loss in functionality and according to ASUS the boards are almost the same. V2 doesn't have the sas controller.
As for the crossflashing procedure Amurtigress an XS member shows the way to crossflashing:
http://www.xtremesystems.org/forums/...&postcount=521

[xpatar]

After crossflashing from v0504 to v0007 I managed to raise my oc without turbo throttling,
bios version 0007 proves to be stable in the long run stress test:

Prime95 9 hours stable


I'm going to check also with the new bios version 0006

[unclewebb]

I was hoping that ben805 could check out the new 0006 bios version but unfortunately he was a bad boy and has been Banned.
Maybe it's just a timeout.

Anyone that can run LinX at 4.4 GHz should be exempt from getting banned. If you are out there ben and have any interesting results from the new bios then you can send them to the email address in the About... box of RealTemp and I will be sure to share them.

It will be interesting to see if the CPU Power Value changes at all with 0006. Probably not. I think that's how they are tricking the CPU into staying at full power, all the time. Good for them.

[unclewebb]

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

Just added a small feature so you can minimize i7 Turbo to the system tray.

[xpatar]

Updated to bios version 0006



Again the same results, no turbo throttling, plus low cpu power consumption (unreal).
I can raise my oc @ 4.4GHz but that takes more than 1.6v Vcpu to keep it stable which results in very high temps

[unclewebb]

There was no way before bios 0007 and bios 0006 that you could run 4322 MHz at 1.56 volts and be hitting 90C without immediate turbo throttling down to a 20X multiplier. Thanks for testing that xpatar.

Bios 0006 shows a date of July 7th, 2009 so it's definitely a newer version than 0007.

[xpatar]

Yes, turbo throttling was like a nightmare for the oc

[WeeMaan]

Is this a sign of turbo throttling?



Where can I get the 0006 bios to P6T deluxe v2 that everyone is talking about?

[unclewebb]

Keep an eye on the calculated multiplier while your CPU is at full load. It should be steady at 21.000 and the standard deviation below it should be close to 0.000 which shows that all threads are reporting the same thing.

For accurate results, I suggest turning off any software that monitors motherboard voltages. On some motherboards, this can cause inaccurate results.
I like OCCT but when testing, try running LinX or Prime95 Small FFTs to keep it simple. RealTemp, CPU-Z and Core Temp don't seem to cause any issues with i7 Turbo. OCCT might be OK too but I haven't seen any testing of it yet.

This post in the P6T thread has a link to the 0006 bios if you need it.
http://www.xtremesystems.org/forums/...&postcount=991

i7 Turbo 6.2
http://www.fileden.com/files/2008/3/3/1794507/Turbo.zip

Use i7 Turbo V8. It's similar to the original but shows the calculated multiplier for each thread. At full load these should all be consistent at 21.000.

[WeeMaan]

Keep an eye on the calculated multiplier while your CPU is at full load. It should be steady at 21.000 and the standard deviation below it should be close to 0.000 which shows that all threads are reporting the same thing.

For accurate results, I suggest turning off any software that monitors motherboard voltages. On some motherboards, this can cause inaccurate results.
I like OCCT but when testing, try running LinX or Prime95 Small FFTs to keep it simple. RealTemp, CPU-Z and Core Temp don't seem to cause any issues with i7 Turbo. OCCT might be OK too but I haven't seen any testing of it yet.

This post in the P6T thread has a link to the 0006 bios if you need it.
http://www.xtremesystems.org/forums/...&postcount=991

i7 Turbo 6.2
http://www.fileden.com/files/2008/3/3/1794507/Turbo.zip

Use i7 Turbo V8. It's similar to the original but shows the calculated multiplier for each thread. At full load these should all be consistent at 21.000.

Okay, I tested linX now at 21*200 and the calculated multi does drop from time to time, sometimes it goes up to 22x to.



I'm going to flash the bios to 0006 to try it out as soon as I can find a solution to a minor problem.
After I have done a bootable cd with Nero and put Afudos ES and 0006.rom and started it up then I cant use my usb keyboard. :S

And one other thing, if I turn on any powersaving feautures (C-state, speedstep, C1E) I get a bsod just after the windows 7 logo appears, every time.
Is this something the 0006 bios might correct?
Because the powersavings feautures is important to me.

[icecpu]

Does any one know why when I run LinX it throttling???
But it does not throttle when run prime95 or OCCT
I'm using windows 7 64bits

[unclewebb]

icecpu: That is common. LinX puts a bigger load on your CPU than Prime95 does so it's likely that LinX will show turbo throttling before other less stressful testing programs.

WeeMaan: Go into your Control Panel -> Power Options and set your Minimum processor state to 100% when testing. This will help keep your multiplier from dropping down at idle and will make your testing more consistent.

When testing with i7 Turbo, run a log file. Start with a mild overclock and make sure that you are getting consistent results from i7 Turbo. With turbo enabled, it should be locked on 21.000 and very steady at full load. A log file lets you see whether a low Calculated multiplier value is a one time random occurrence or as soon as the load and temperature goes up, the multiplier starts dropping down.

The only problem that I've seen with i7 Turbo is some programs on some motherboards that monitor motherboard voltages can cause the occasional dropout in the calculated multiplier. HW Monitor and Everest and perhaps OCCT might be a problem. That's why it's a good idea to test first that you are able to get consistent results from i7 Turbo.

[unclewebb]

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

This update tries to filter out a few more of the random values that sometimes show up for the Calculated Multiplier at full load.

If you are having trouble getting consistent results at full load then post or send me your i7 Turbo log file so I can have a look.

SendSpace is a good site to upload a log file too.

Also let me know what operating system you are using and what other monitoring software you are using when testing and any other details you can think of.
Thanks.

Edit: RealTemp 3.00 can cause issues with i7 Turbo results when the internal timers overflow and wrap around back to zero. When you are testing, I recommend that you update to RealTemp 3.30 which handles timer wrap around better.

http://www.fileden.com/files/2008/3/...alTempBeta.zip

[Stao]

What is the exactly meaning by turbo throttling?
The i7 Turbo v8 shows that the multiplier minimum changes from 21.000 t0 15.997, but that only happens 0.5 sec once per 15 mins when I use prime in-large to stress my CPU. Is that means turbo throttling?

[unclewebb]

At full load when running a stress program like Prime on a Core i7 920, your multiplier should be locked on its maximum of 21.000.

Turbo throttling is when the multiplier will start to cycle back and forth between 21 and 20 as the load and temperatures go up. When this first starts to happen, the CPU multiplier will be bouncing back and forth between 21 and 20 hundreds of times a second as the +1 Turbo feature is engaged and disengaged rapidly. When this issue gets more severe, the multiplier will be stuck at 20 and you won't be getting any turbo boost at full load.

i7 Turbo V8 uses the Intel method to calculate the average multiplier and should be able to report the first signs of any problems.

If you go into your control panel -> Power Options you should be able to set your minimum processor state to 100% when testing. One multiplier dropout every 15 minutes is nothing to worry about. Run an i7 Turbo V8 log file. It will clearly show a one time random drop out versus your multiplier constantly reporting a number less than 21.000. Post it here or send me a PM if you want me to have a look at it.

[Stao]

Thanks for the reply unclewebb, here is some screen shoot about it.




By the way, I don't see the power option that I can change my processor usage.





And here is the log file shows that I have turbo throttling once per 5 min.

[unclewebb]

Good news Stao. No signs of turbo throttling. The occasional glitch in i7 Turbo V8 is likely being caused by RealTemp 3.00 which doesn't do a very good job of dealing with the internal timers that have a problem with overflowing about every 5 minutes.

In the post just before you asked your original question it says this:

-----------------------------------------------------------------------------------------------------
Edit: RealTemp 3.00 can cause issues with i7 Turbo results when the internal timers overflow and wrap around back to zero. When you are testing, I recommend that you update to RealTemp 3.30 which handles timer wrap around better.

http://www.fileden.com/files/2008/3/...alTempBeta.zip

--------------------------------------------------------------------------------------------------------

RealTemp 3.30 is new and improved so it and i7 Turbo V8 are able to share the system timers in a more friendly manner and your multiplier should be steady at 21.000 now.

rge sent me a log file recently of i7 Turbo V8 data where the calculated multiplier didn't budge once during 4 hours of testing.

Usually if you have EIST enabled in the bios there is that Minimum/Maximum processor state option. I'll have to go play with Windows 7. What motherboard are you using and bios version? DFI had some recent bios updates to take care of some Win 7 compatibility issues. You'll probably have to find another user with a similar system to yours to compare with. How low does your Calculated multi drop at idle?