Real Temp - New temp program for Intel Core processors

[darkzone]

In this case, is how I thought, there is a problem with current OS installation. Well, I have work to do
Thanks for your time and patience, I'll keep you in touch about this

[unclewebb]

Don't do nothing too drastic yet darkzone. Give me a day or two to get the RealTempBeta version back on track.
With all of the information you've provided me, I might be able to come up with a simple fix. If everything else is working fine on your computer then there's no use doing an OS re-install just yet.

Are you using the Administrator account as your main account or are you using a separate user account?

[darkzone]

Yes, I use built-in Administrator account (yes, I know, that's stupid ) but I'm the only user on this machine. And no, I won't do anything drastic about this issue, just want to verify services and functions that I've stopped.

[Bludd]

Thank you. It has the same behaviour as 1.49 and 1.50

Just an update. CPU-Z 1.50 now shows 12x multiplier at idle, just as Real Temp. I think the change in power settings (to laptop) made all the difference.

[unclewebb]

Bludd: I think there are a lot of people running XP that don't realize that you might have to use the laptop power setting to get Speedstep to drop your multiplier down to its lowest value at idle. In Vista you have to go into the Power Options -> Processor power management and reduce the Minimum processor state to about 50%. I think CPU-Z 1.50 will still show you the full multiplier if you don't have laptop enabled. As far as I know, that's what Intel recommends should be displayed but I think showing the average multiplier gives a user a better idea of what's really going on.

On the Core 2 Dual and Quad Core chips that I've checked, dropping your multi down to 6.0 doesn't save you any power or create any less heat. At the wall, the difference is tough to measure. Usually less than 1 watt which is a normal fluctuation at idle anyhow.

The only way any of these power saving features will do you any good is if the core voltage drops down as well at idle. If you like to overclock then most boards won't drop your core voltage at idle so dropping the multi is kind of pointless. It doesn't really cost you anything but you also don't gain anything either.

darkzone: I know it's a bad idea and all that but I think a lot of users just use the Administrator account. Less hassle. I thought it was possible that maybe you had set up an account that had limited privileges or something like that. Hopefully that testing program I sent you will tell us something new.

[darkzone]

Well, the situation is blue, as we have a word here în my country
With new version of Set Thread which you sent to me, I've had these results:

- at first run program is working on first two cores only


- at second run I started RT 3.00 to see what happens


- at third run I set the program to work on last two cores only


- at fourth run I set the program to work on all four cores


Now, I hope this screens will be helpful for you to find out what is wrong in here.

[unclewebb]

Your screen shots are very helpful. This bug finally makes sense. For some reason, when an application starts on your computer, by default, it is only being allowed to run on core 0 and core 1. I don't know why because the Server 2008 HPC SP1 that I tried does not work like that. Normally when an application starts, it automatically is allowed to run on all available cores. That's what RealTemp 3.00 assumes and in your case, that assumption is wrong. It doesn't try to read the temperatures from core 2 and core 3 because it thinks the operating system has blocked access to them.

I should have a fix ready for you to test by tomorrow morning your time.



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

Edit: If you have mfc71u.dll on your system then you can try the Beta2 version:
burebista likes this one better because it's smaller and initially starts up quicker.
http://www.fileden.com/files/2008/3/...lTempBeta2.zip

I'm hoping this beta is much better than the last one and that the missing cores have finally been located.
I also threw in an extra button for Nvidia owners to make it easier to see the Minimum and Maximum GPU temperatures.
Core i5 support has been added as well as support for some of the early Core i7 based Xeon (ES) processors.

[darkzone]

OK, I'll check this tonight, and the new version of RT 3.00 also. I hope this time will work

But, a question still remains: why previous version, RT 2.82, runs just fine onto the same version of OS?

[nigelke]

I just wanted to say, thanks for your hard work on this program

[unclewebb]

nigelke: You're welcome.

But, a question still remains: why previous version, RT 2.82, runs just fine onto the same version of OS?

Good question. RT 2.82 did not ask the operating system what cores it's allowed to run on. RT 3.00 was trying to be more operating system friendly and decided to ask. Bad idea.

On your operating system, for some unknown reason, when RealTemp starts up, it's only being given access to the first two cores. RT 3.06 tries to force its way onto the other two cores.

It only took me 3 weeks, 1 new utility program and the downloading of 5GB of operating systems from MS to figure this out.

It will be interesting to see if this fix finally works.

Edit: You should also see if you have any other apps that are being limited to only run on your first two cores when they start up.

[darkzone]

Finally ... runs. IT'S ALIVE! Maybe because this is the 13th screenshot?


Program started onto all four cores at first run.

OK, thanks for your support, help and patience, you're THE ONE

[unclewebb]

It took a while but that dumb bug has finally been taken care of.



Now you can say good-bye to version 2.82.
Version 3.06 is much better.

When you start any program and look at SetAffinity... in the TaskManager it should automatically be running on all 4 cores. Have a look to see if you can find any other programs that are being limited to only your first two cores.

There might be something in your operating system / registry that needs a tweak so all of your cores, always get used.

[darkzone]

Yes, as I suspect, there is something wrong in my OS configuration. I started WinRAR (which is known as multithreaded optimized) and SetAffinity shows that the program runs onto first two cores only:


Now, I have to find out why.

Besides, I manually set the affinity for each running program when the system was ready to "up an run". Since then, I didn't verified anymore, being sure that all is OK. So, it was a mistake

[randomizer]

msconfig -> boot tab -> advanced settings...



Check out what it says for number of processors. Try unchecking it if it's checked, or vice-versa.

[darkzone]

That box is unchecked, Kevin told me to leave it so. I'll try tonight to check it and see what happens.

[Demo]

checking the box and selecting a number tells the OS how many CPU's to enable.

[unclewebb]

I just tried setting this to 2 and my Q6600 Quad instantly became a Dual Core according to Vista.
The TaskManager only shows two cores as well.



Usually when Number of processors is unchecked, the operating system defaults to using all available processors.
Maybe if you set your OS to 4 darkzone, all programs might be able to see all 4 cores.

Edit: I noticed that when Core Temp only sees two cores it thinks I must have swapped my Q6600 for an E6600.

[darkzone]

Well, the problem is deeper than I thought ... Checked or unchecked, 2 cores, 4 cores, it seems doesn't have any effect. So, must dig more deeper

[unclewebb]

burebista noticed some instances when the load reported by the RivaTuner / RealTemp plug-in was significantly different than the built in RivaTuner CPU.dll plug-in and asked me to come up with an explanation. The RealTemp method of calculating load is different than the traditional method so I was curious too of which method is the most accurate. Here's what I found during testing.

I decided to run 4 threads of Prime95 Small FFTs so the RivaTuner CPU.dll plug-in was showing 100% for all 4 cores and the RealTemp RTCore.dll plug-in was showing an average of about 99%.

I stopped Prime95 so the load instantly returned very close to 0%. I went back and did some screen captures which shows the results during this rapid load transition.

Time = 13:46:00



CPU.dll is reporting 100% for all 4 cores while the RealTemp plug-in is already down to 57.1%

Time = 13:46:01



The Riva CPU.dll plug-in is showing an average of 29.89% while the RealTemp plug-in is already correctly reporting 0.1%

Time = 13:46:02



Exact same as last second. The Riva CPU.dll plug-in is still reporting an average of 29.89% while the RealTemp plug-in is at 0.1%

Time = 13:46:03



Finally the RivaTuner CPU.dll plug-in has caught up to the RealTemp plug-in.

My conclusion based on the above is that the RealTemp RTCore.dll plug-in is much faster and more accurate at correctly reporting the load during rapid load transitions.

I don't think the RealTemp RivaTuner plug-in is always 100% accurate but in some situations I think it is more accurate than the TaskManager Load meter as well as other load meters based on similar code.

Edit: During the early part of this test I was using the Clock Modulation feature to adjust the Load. Traditional Load meters will continue to report this as 100%.

[randomizer]

Well, the problem is deeper than I thought ... Checked or unchecked, 2 cores, 4 cores, it seems doesn't have any effect. So, must dig more deeper

See what happens when you pick 1 core. Remember, you're dealing with Windows. Not everything works like it should I had to get that previous screenshot from Google because I can't even click on anything in my boot tab.

[Falkentyne]

Hi Unclewebb,

last week, I made a post asking about the 'load' issue in Realtemp.
The issue was that as the cpu speed gets higher, the load % seems to slowly decrease in realtemp.

I'm using a QX9650 and P5WDH board. either 2x2GB Ram or 1x2GB.
At 3 ghz (333x9)=default speed, Realtemp shows load of 99% with Prime 25.5 running.
At 3.7 ghz (370x10), realtemp shows load of 89%.
At 4 ghz (333x12), load is 74%. (same thing happens under Linpack when it gets to maximum stress, btw).
I forgot what it was at 4.1 ghz but it was in the high 60's.

Task Manager shows all 4 cores at 100% usage and process prime95 at 99%.

I actually noticed this problem many months ago but never thought about reporting it.

These tests were all after a fresh boot, with nothing running except necessary windows services, Prime95, Realtemp and CPUZ.

I tried the 3.06 beta and same problem.
You mentioned something in the other post about it but I didnt really understand it.

[darkzone]

See what happens when you pick 1 core. Remember, you're dealing with Windows. Not everything works like it should

I tried with one core, two cores etc., nothing changed. I'm waiting for i7 940 to be delivered (I already have mb and RAM), to setup new rig and then I'll reinstall the OS.

[unclewebb]

Falkentyne: I don't yet know if your problem is a RealTemp problem or maybe a problem with your CPU when overclocking. The number of people that have reported problems with the Load meter is very limited and during my testing when running 4 threads of Prime95 Small FFTs, it consistently shows in the high 99% range. When I recently did some testing for darkzone using Server 2008 HPC, on the same CPU it got as high as 100.0%. When I run 1, 2 or 3 threads of Prime, I get 25%, 50% or 75% more or less and it stays very consistently at those numbers unless something in the background kicks in and it briefly goes higher.

Use Prime95 Small FFTs when testing because I know the load it produces is very consistent. Are your percentage numbers consistent when you are running Prime95 Small FFTs? When you are overclocking to a higher MHz, does the Load meter initially start out at ~99% and then drop to 89% or 74% and then remain consistent at this amount or does it immediately only go to 74%?

There is a feature within Intel CPUs called Clock Modulation that can become automatically activated either by software or it can also become activated internally by the CPU itself. The percentage numbers you've mentioned have me wondering if maybe something like that is going on. Run Prime95 Small FFTs and go into the Settings window and you can play around with software activated Clock Modulation. The percentage numbers listed (87.5%, 75.0%, 62.5%, etc.) are Intel's approximations from their documentation. Depending on what multiplier your processor has, what you actually end up with will be different and won't equal those numbers.

My Q6600 has a 9.0 multi by default. My Load meter when playing with Clock Modulation will go from 99% to 88%, 77%, 65%, 55%, 45%,... Those first 6 steps seem to correspond pretty closely to 9/9, 8/9, 7/9, 6/9, 5/9 and 4/9. The last two steps I get are more like 3.5/9 and 2.5/9.

When Clock Modulation is activated on my Q6600, for every 9 pulses of the clock generator, 1 or more pulses gets ignored depending on how much modulation is being used. Programs like CPU-Z and RealTemp will still show your CPU running at full MHz but internally the CPU is acting sort of like a heart that is skipping a beat. This causes the CPU to run cooler since internally it's really not doing the same amount of work as a similar processor running at that same MHz which doesn't have any Clock Modulation going on.

Here's an example of my CPU at full load with no Clock Modulation:



When I set Clock Modulation to 50%, that seems to be equivalent to a load of about 5/9 or 55.5%.



The Task Manager and other traditional load meters continue to report this CPU working at 100% when internally speaking, it's only working half that hard.

It's obvious that this CPU is not working as hard because the average core temperature has dropped by about 10C compared to the first picture. The Task Manager continuing to report 100% Load does not accurately reflect what this CPU is doing while in this case, RealTemp does.

Do some testing and see if your RealTemp reported Load percentages correspond closely to what you'd get if you manually set Clock Modulation.

It's possible that this is nothing more than a big RealTemp bug but it's also possible that Clock Modulation is going on internally within your processor when you are overclocking. It's just a theory but it looks like the harder you push it, the more modulation kicks in.

I'll read the documentation and see if I can find a way to read the internal Clock Modulation status of a CPU. I think the user Clock Modulation and the hardware Clock Modulation are reported in separate registers. This bug is even more interesting than the last one.

Edit:

Model Specific Register (MSR) 0x19A is where software Clock Modulation data is stored within your CPU. User clock modulation is stored in the lower 5 bits. The upper bits are listed as Reserved by Intel but when your processor is reporting some lower than expected Load numbers in RealTemp, try doing a Read MSR on 0x19A. I don't know if it will show us anything but it's worth a look.

You can use my MSR tool for that:
http://www.fileden.com/files/2008/3/3/1794507/MSR.zip



You can click on the Read MSR button as much as you like and you won't hurt anything. If you enter random data into those two white boxes and go to different MSRs and click on Write MSR then you might get lucky and crash your computer. If that did happen, after you re-booted, everything would be back to normal so no worries.

Also try turning off TM1/TM2 in the bios when testing to see if that makes any difference when you're testing.

[burebista]

I'll read the documentation and see if I can find a way to read the internal Clock Modulation status of a CPU. I think the user Clock Modulation and the hardware Clock Modulation are reported in separate registers. This bug is even more interesting than the last one.

Kevin don't forget our good ol' friend: RMClock.

[unclewebb]

I've always liked RMClock 2.30
I tried a later version once but went back to 2.30.

I'm not sure how RMClock calculates its load percentage or if it will be able to report any internal throttling but it's worth a look.