Real Temp - New temp program for Intel Core processors

[Monstru]

C1E was enabled. Power pforiles were set on balanced. So I guess the results are ok. What puzzles me are the temps WAY to low from the sensor test.

[donmarkoni]

Good to know unclewebb,

prolly my stupid fault anyway.

I don't think it's your fault. I have the same problem. Stumbled upon it some weeks ago, IIRC when 19x.xx drivers came out. Reinstalling didn't help. Uninstalling and then installing helped, IIRC.

BTW, too many "IIRC", so I'll give it a try, cause right now don't have GPU temps anywhere but Everest.

EDIT: Uninst/inst didn't help. Used DriverSweeper http://www.guru3d.com/category/driversweeper/ then installed 191.07 and it works great.

Sorry for all this relatively off topic chat, but this is proof that missing temps has nothing to do with RT. Maybe there is many driver leftovers on this 18x.xx to 19x.xx crossover and regarding it has some power management additions, it could be the reason for mess up.

[unclewebb]

Monstru: The version of RealTemp you are using is not correctly recognizing your processor so it is not reading the TJMax value from the CPU like it should be doing. Perhaps Intel has lowered this. When RealTemp finds an unknown CPU, it takes a wild guess and assumes 100C. If that is wrong then your temperatures might not be reported correctly. I made a lot of progress today so I should have something ready for testing by tomorrow. Core Temp was updated recently. Maybe see what that says for TJMax.

donmarkoni: Thanks for figuring that out. Hopefully your fix works for other users. This bug came about somewhere around the late 180 series Nvidia drivers.

[Monstru]

Ok Kevin, I will wait for the new version. I will also try the latest CoreTemp a little bit later.

[unclewebb]

Unwinder, the programmer of RivaTuner, has also had some issues lately with Nvidia's latest drivers and GPU temperatures. Here's his fix:

http://www.evga.com/forums/tm.asp?m=...key=%EF%BF%BD#

Thanks for that info burebista.

[randomizer]

I'm using 191.03 and my temps show up in Real Temp 3.30 RC 10.

EDIT: 150 page milestone has been reached

[Monstru]

Here you go Kevin, with the beta:

[Astratuner]

I don't think it's your fault. I have the same problem. Stumbled upon it some weeks ago, IIRC when 19x.xx drivers came out. Reinstalling didn't help. Uninstalling and then installing helped, IIRC.

BTW, too many "IIRC", so I'll give it a try, cause right now don't have GPU temps anywhere but Everest.

EDIT: Uninst/inst didn't help. Used DriverSweeper http://www.guru3d.com/category/driversweeper/ then installed 191.07 and it works great.

Sorry for all this relatively off topic chat, but this is proof that missing temps has nothing to do with RT. Maybe there is many driver leftovers on this 18x.xx to 19x.xx crossover and regarding it has some power management additions, it could be the reason for mess up.

Thanks donmarkoni,


the driversweeper re-install methode did it!!!!

[unclewebb]

Thanks Monstru, that looks better. Intel must be using TJMax = 105C on your CPU. Core 0 has always been the most accurate temperature wise. How does this look now compared to your room temperature? These new Dual Cores are going to run a little cooler than the Quads but it might still be a couple of degrees too low. I'll try to change RealTemp so it correctly calls your CPU a Core i5 instead of a Core i7. Intel has so many CPUs that are overlapping each other in specs these days that it's becoming more work for programmers to keep them all sorted out.

Can you try running this program and post your results?

http://www.fileden.com/files/2008/3/...507/MSR1AD.zip

It reads Model Specific Register (MSR) 0x1AD which should show me how many bins of turbo boost this CPU is capable of. I don't have any Intel documentation to go by so it's always best to ask the CPU to tell me what it can do.

[Monstru]

No problem Kevin, I will use the program and report back. My room temp is 25 oC.

[Monstru]

Hey Kevin, I ran MSR1Ad and it says

1818191A

[unclewebb]

With a room temperature of 25C, it looks like your sensors read a little on the low side. The original Core i7-9xx series had very consistent and accurate sensors at idle and at TJMax. I can't yet say the same thing based on some of the socket 1156 screen shots I've seen. I hope Intel didn't decide to cheap out again like they did with the 45nm Core 2 sensors.

I've also seen another issue recently where on a retail Core i7-920, the TJMax written into the processor was 93C instead of the usual 100C. This resulted in temperatures being reported that looked to be too low compared to every other Core i7-920. More core temperature games?

Edit: Here's how those numbers in MSR 0x1AD translate. There are 4 numbers in that MSR that correspond to the maximum multiplier available when 4, 3, 2 or 1 core is active. It might just be a coincidence that you have 4 numbers in there or another sign that your CPU might have started life as a Quad. Maybe it's like an AMD chip where the mobo guys found a trick to unlock some hidden cores.

Those numbers are in hexadecimal and so the numbers in decimal are 24, 24, 25, 26

For your CPU, when two cores are active, the maximum multiplier is 25 which you already know. When only one core is active, the maximum multiplier can go up as high as 26. If you can disable a core in the bios, you should be able to see the full 26 times multiplier all the time. The other way to get this is to enable C3/C6. This allows one core to go to sleep which Intel classifies as an inactive core. When running a single threaded activity, you should see the multiplier continuously cycle between 25 and 26. The multiplier on the hardest working core will approach 26 but any background processing will activate the second core and kick you back to a 25 times multiplier. This multiplier cycling is happening hundreds of times a second based on load when C3/C6 is enabled. Post a screen shot of RealTemp or i7 Turbo if you can see this happening.

[unclewebb]

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

Lots of minor changes behind the scenes including better support for Intel's new socket 1156 CPUs as well as their new Core i7 mobile CPUs.

Changes include:

-the use of the i7 Turbo method to calculate the multiplier for all CPUs. If this causes a problem for older Core 2 CPUs that may have the internal timers disabled then you will need to add this line to the RealTemp.INI file to use the previous method:

MSRMulti=1

The new i7 Turbo based method when used on a modern Core i7 or i5 CPU with C3/C6 enabled in the bios will be able to show you the multiplier of the hardest working core which in my opinion and Intel's opinion, best represents how fast your CPU is actually working. This method also works well with Core 2 mobile CPUs that support Intel Dynamic Acceleration (IDA).

The i7 Turbo method allows Super Low Frequency Mode (SLFM) to be displayed correctly on the Core 2 mobile processors that support this feature. A multiplier of 3.0 at idle when this feature is enabled is typical.

-the elapsed timer has been fixed so it shouldn't lose time any more.

-a work around for an incorrect bios setting was found for the missing core issues that sometimes plagued Quad core owners.

-a Task Manager based load meter (TM Load) was added as an option. The original load meter is based on the percentage of time a CPU is in the C0 state which is very accurate for many CPUs. Mobile CPUs work completely different and this number does not accurately reflect the load of the CPU at idle so now you have an option.

-the XS Benchmark was changed slightly so now you can Cancel it while it is running and you shouldn't see any more, "Program isn't responding" messages when it is fully loaded testing.

-an option to display your temperatures in Fahrenheit was added. Just add this to your RealTemp.INI file.

Fahrenheit=1

-the Settings window has been reworked for better consistency in preparation for possible future upgrades.

-only maximum VID is displayed in the Settings window for Core 2 owners for less confusion. This option will display real time VID on the main screen.

-an option to Disable Turbo Mode on Core i5/i7 CPUs.

etc., etc., etc., etc.

Lots of changes so if you see any problems let me know as soon as you can so I can get things fixed up. An official release is way overdue which I'd like to do in the near future. The RivaTuner plugin is next in line for an upgrade to better support the new CPUs.

Thanks rge, burebista, Monstru, somebody on TPU and my other contacts for your infinite testing and for helping to make RealTemp better and more compatible with a wider range of CPUs.

[stasio]

Thanks for update,soon I will check with i7-860.

[unclewebb]

Thanks stasio. I'm hoping this new version will more accurately report your multiplier when running a single threaded app with C3/C6 enabled similar to what the i7 Turbo program shows. This version had the new i7-860 and i7-870 in mind as well as the i5-750. With multiple bins of turbo boost becoming more common on Intel's new chips, we need some accurate software to monitor that. CPU-Z only tells part of the story.

I might have to go do some thread crapping in those other threads to get users to give RealTemp 3.35 a try so everyone can better understand what these new CPUs can really do.

If you can, post a screen shot while running a Super PI 1M bench with C3/C6 enabled to show all the bins of turbo boost that are available.

[Fungus]

Hi UncleWebb, still using realtemp and loving it. I got another small bug for you. If you need to kill explorer.exe (you know, sometimes it hangs on weird stuff for no reason), the tray icons don't reregister. =] You have to logoff/on again or reboot to get them back. I know it's something that's strange and doesn't happen very often for most people. But I thought I would mention it anyways.

Love the program as always, cheers!

[unclewebb]

Thanks Fungus. When you kill explorer, after explorer restarts, does the Task Manager show that RealTemp is still running but it's not on screen anywhere to be found and the system tray icon(s) are also missing?

I guess I've been lucky and haven't had explorer crapping out on me for a long time so I haven't noticed this issue before. If I can recreate it I'll see if I can find a fix for it. What OS are you using, XP?

[Fungus]

RealTemp is of course still running, as I used taskman to kill explorer.exe (all instances), and then restart explorer.exe using New Task. Yes, I am using XP Professional SP3. (This is not the only program to not re-register icons =)

Some do, some don't. Riva tuner doesn't either, but nhancer does. OpenDNS doesn't, Creative Volume control doesn't, Nvidia Control panel doesn't, but many programs do, like my antivirus and some other stuff. I guess they aren't bugs that devs check for =]

[unclewebb]

Thanks for a few more details Fungus. You're right, I've never given this a second thought but now I will. If I can recreate it and come up with a simple solution, I will. I hate bugs as much as anyone.

I quietly released 3.36 today. The only change was that the elapsed timer that I just improved for better accuracy was counting while in stand by mode which I didn't like so I fixed it. The RealTemp elapsed timer should only count up while the computer is running.

[KURTZ]

here is my Xeon chip, now cooled by the stock heatsink

[unclewebb]

That's exactly what I was looking for KURTZ. Now I can explain what the new version of RealTemp is doing differently.



RealTemp 3.30 used to calculate the multiplier for each of the 8 threads then it would average them out and report that.

With the new CPUs that can have multiple bins of turbo boost available, the overall average can be a meaningless number like it is in this situation when C3/C6 is enabled. Averaging 3 cores that are spending the majority of their time asleep in C3/C6 doesn't give a user any useful information so it was time to change RealTemp.

RealTemp 3.36 looks at all of your threads, finds the one working the hardest, the one running Super PI in this example, and reports the multiplier for that thread. Intel recommends that software should display the highest multiplier and when you see an example like this, that makes sense.

When running a single thread of Super PI, the multiplier will spend most of the time cycling between 25X and 26X. As soon as two cores become active, the multiplier will drop down to 25X and after that task is completed, the second core can enter C3/C6 and become inactive again which allows the CPU to resume using the 26X multiplier again. Obviously, the more background activity you can eliminate, the larger percentage of time that it will be able to stay at the full 26X. Turning off programs like CPU-Z, i7 Turbo and that side bar fluff should help get the multiplier higher.

By the looks of your testing, RealTemp 3.36 is a good tool to use when you are trying to maximize your multiplier. Less background activity equals higher multiplier equals better Super PI scores or what have you. Thanks KURTZ for sharing this info.

Here's how rge's Core i7-950 looks. The maximum multiplier on one of these is 25X. You can see that he was able to get close to the maximum when he was hitting 24.7. I think this new multiplier number that RealTemp reports is more meaningful than CPU-Z rounding things off and telling you 24.0.

[KURTZ]

Turning off programs like CPU-Z, i7 Turbo and that side bar fluff should help get the multiplier higher.

just tried, but the max multiplier is still 25.3X ... :S

[unclewebb]

Fixed VID reporting on the 45nm mobile CPUs.
Thanks somebody at TPU for bringing this to my attention.

http://www.sendspace.com/file/nsqj8n

[Monstru]

Wow, a new one allready!!!

[stasio]

uncle,
my CPU idle,but RealTemp show always Load 10-12% ?