Real Temp - New temp program for Intel Core processors

[i43]

Hi Unclewebb
here 2 screens on 45nm quad test on oc'ed @3600 .. 2 runs - clk modulation 50% and NO clk mod. (common state) - Speedstep/EIST - Disable
all the same as You tested on your's, probably same temperature (10~12C)drop on 50% down-clock.
also noticed that only Everest can register (partly?) this clock modulation only via its CPU (W) wattage / power usage field
just as i see - that dropping power draw ~ approx 50% during 50% clk. modulation state

[DGill]

Any idea why Real Temp above 3.00 only displays 3 cores? I've tried vrs 3.05 and 3.06

Using i7 320 + gigabyte EX58 + Vista 32 bit

[randomizer]

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.

It does detect something it's hard to see what's happening. In fact it even gives you a popup telling you you're throttling. I tried it once back when I was more afraid of burning my chip up. It didn't show much but I'd say with a bit more heat it would have throttled for long enough to show something valuable:



Man, I was still using XP back then... I like this page for showing what should be happening, albeit with an older version of RMClock and using a Prescott.

[Falkentyne]

Hi Unclewebb:

I Tried the programs you mentioned and here are a few results.
At 3500 mhz:



And at 4000 mhz:


So RMclock isn't showing any throttling. MSR isn't showing anything strange either.

1 more comparison:
4000 at 50% modulation:


BTW the dips in the graph were when I kept stopping prime or messing with the modulation. (prime really needed 1.352v in cpuz to not error out).

[unclewebb]

i43: Thanks for your pics. Your Quad works like my Q6600 and seems to display accurate Load percentages. rge did a lot of testing on his Core i7 and also had believable looking results.

DGill: Version 3.05 was designed to try and figure out what was going on with darkzone's computer. I made 101 changes trying to understand what was going on with his operating system. I wouldn't recommend using version 3.05. If version 3.00 works for you then continue using that at the moment.

Version 3.06 was designed to fix darkzone's problem and in theory should work OK but obviously it doesn't work for you. Now that I have a clear understanding of what was causing his problem, I'll try making some adjustments and come up with a solution that works for everyone. Maybe 3.06 will become known as the darkzone version.

Falkentyne: RMClock and my MSR tool don't show anything unusual so it looks like either the RealTemp Load meter is borked on your CPU or it is telling you that there is a problem with your CPU when overclocking.

One thing I noticed comparing your first two pictures is that in the first picture your temps are 50, 41, 33, 48 and in your second picture the temps are 57, 51, 44, 54. The difference is 7, 10, 11, 6. That's very unusual. Uneven sensors on a 45nm Quad are not unusual but when running Prime95 Small FFTs and playing with the Clock Modulation or changing the MHz or core voltage, all 4 cores tend to change in temperature more or less equally. Yours don't.

burebista just asked me yesterday to include an option in the RealTemp / RivaTuner plug-in so the Load of each individual core can be displayed. I think I will work on that next because it might give us some more information about what's really going on inside your processor. Currently, RealTemp calculates the Load for each core individually, adds them up and divides by the number of cores to report the average so changing that code to draw 4 squiggly lines instead of one shouldn't be too hard. I'm interested to see if all 4 of your cores are reporting a lower load percentage or only a couple of them are.

You might be able to run a multi-core benchmark like wPrime and compare your results at 4000 MHz to another user with a similar 45nm Quad at the same speed. Run RealTemp at the same time and watch what Load it reports. Your results are either going to be very comparable to another 45nm Quad or if your processor is only working internally at 75% when overclocked to 4GHz then your results are going to be very obvious that something is wrong.

Another test I just tried was to run 5 instances of RealTemp. When the first version starts up, go into the Task Manager and set the Priority to Realtime. Use this one so it can monitor the other 4. Start up the other 4 instances and click on the XS Bench button of those 4, one after another, so all 4 are running at the same time. What does the load meter say on the first instance of RealTemp? The scores on all 4 should each be somewhere around 1330 when you are at 4GHz. With my Q6600 at 3 GHz, all of them are pretty close to 1000. This benchmark scales very linearly with MHz so it might show you if a core or two are not working at full speed.



At the moment either RealTemp is wrong or the timers that RealTemp reads within your CPU are not working like other Quads do under load or your CPU is internally throttling. With some more testing hopefully we can figure something out.

[Falkentyne]

Hi,
I just did a quick test at 340x10 (3.4 ghz) and the 4x xsbenches reported
1131, 1130, 1131 and 1131. Time was 12.931, 12,942, 12,940 and 12,934.

I couldn't press them all at the same time as mouse "input" seemed to freeze for a second or so after I went to press the second one.

I'm going to reboot now and try 4 ghz (333x12).

BTW about that sensor:
That third sensor barely moves.
The one that's "stuck" on 27C...
If I run quad prime at, let's say, 3 ghz, the temp doesn't even go anywhere.
However if I run linpack, the sensors get up to 69, 69, 63, 63.
In prime95, the only two sensors that seem to move properly are sensors 1 and 4.

Edit:
Just ran the 4x xsbench + 1 realtime priority realtemp, at 333x12=4ghz.
score were 1328 (11.017s), 1330 (11.003s), 1329 (11.010s), 1330 (11.001s)

Temp sensor#3 remained stuck at 27C.
#1=49C, #2=34C, #4=47C.

Going to try wprime now.
Ok wprime 32M was 10.656 secs
Trying 1024M now.

Ok i just noticed something, unclewebb.
Final score was 338.703s (lowest) and 339.516s) highest).
Cores 1 2 and 4 finished at almost the exact same time in 1024M test, but when they finished, core 3 was at 96% O_o

Um....what does that mean? O_o

RMA the CPU? >_> Or is this a motherboard issue?
I'm going to go back to 3 ghz and try it...

I will say this:
Back when the QX6700 and QX6800 was out (I had an X6800 at the time), I remember a LOT of posts about people saying that 1 "thread" in quad prime95 was lagging behind other threads, and people discussing problems with "load balancing". Now I think I know what they were talking about...But if this happened on the QX6x00's....is this a motherboard or windows problem? Or do I have to RMA the CPU?.....

1 other thing I noticed:
in RMclock,
cores 0 and 1 have a VID of 1.2375v. Cores 2 and 3 have a VID of 0.825v. WHAT? O_o....

Well went back to 3 ghz again. Realtemp showing 99-99.9% load in wprime right now, but again core3 is lagging behind the other cores slightly. So it seems that the core3 lagging problem isn't related to realtemp, as RT is showing 99% at 3 ghz?
I'm going to search XS for more posts about the lagging core problem....

BTW at 3 ghz in wprime, which is running right now, at 1.25vcore, cores 2 and 3 sensors are not moving at all (well, core 2 moved by 1C, core 3 moved by 0. Cores 1 and 4 sensors are functioning properly. Core 3 is pretty much stuck except at high vcore and load (only Linpack gets it to rise up to as much as core 4).

[DGill]

DGill: Version 3.05 was designed to try and figure out what was going on with darkzone's computer. I made 101 changes trying to understand what was going on with his operating system. I wouldn't recommend using version 3.05. If version 3.00 works for you then continue using that at the moment.


Thanks for the reply

[Falkentyne]

Unclewebb
forum was down for awhile
anyway, I found the problem with the load% on realtemp.
It's something to do with the multiplier.

At the default multiplier of 9, realtemp shows 99.9% load.
At multiplier of 12, it's 74%, regardless of CPU speed (FSB) used. both 266x12 and 333x12=74% load.
At multiplier of 15, it's 59%. (200x15).
And at 20, it's 49%.(i used 150x20=3ghz).

Benchmark scores are not affected by this (except much longer tests or bandwidth from using a lower fsb);
200x15 in wprime 32M gave 14.1s, same score as 333x9. Benchmark scores related perfectly to core speed.

I haven't tried going under multi 9.
Any ideas why this is happening?

BTW about the core 3 being slightly slower in wprime, seems that is more related to load balancing; I ran some "1 thread" tests (took a really long time) with affinity for the new wprime process that appeared, set to a separate core, and core2 seemed fastest, followed by core 4, 1 and 3. But at 1700 seconds, core 3 taking 10 seconds more than core 1, is not much (basically a 10 second difference between the cores finishing, at about 25 minutes).

In the 4 thread test though (which finished in 5 minutes 32 seconds) it was much more pronounced; core 3 finishing about 16 seconds after the first 3 (which finished at the same time). That seems to point to an issue with load balancing I guess....still very strange how that's happening though... 10 seconds slower at 25 minutes with single core/thread test, and 16 seconds at multithread/4 threads at 5 minutes... weird...

[pra]

Hi, I have ASUS P5Q Deluxe and I have read that ASUS advice that the new Bios 1805 will fix the termal sensors of the motherboard. I have already flashed my Bios with new one but I still have a problem with my Core 3 and Core 4 on my Q9550. The promblem is: it doesn`t matter if my PC is overclocked or default all the programs which indicates the CPU temp shows that the 3rd and the 4th cores have 39C and the temp doesn`t move down.

Thank you in advance!

Pls give me advice: do I need to wait for another Bios, do I need to change my CPU or do I need to change the motherboard?

It looks like you have 2 stuck sensors. Do a CPU Cool Down Test and post your results here.

I'd say not to worry, stuck sensors are harmless. But annoying.

[burebista]

Man, you forgot to launch prime95 when RealTemp asked you.
Please do it because you have 0.1% load.

[pra]

Man, you forgot to launch prime95 when RealTemp asked you.
Please do it because you have 0.1% load.

sory

[unclewebb]

Falkentyne: I've been trying to post all morning that I think I figured out what's going on with the load meter on your computer but XS was down so I guess you beat me to it.

The method and timers that I am using are somewhat documented for the Core i7. I found they also seemed to work on most Core 2 processors so that's why I included a Load meter based off of them in RealTemp. On most Core 2 Dual and Quads like my Q6600 with a fixed multiplier the RealTemp Load meter works great. On Extreme processors that have an adjustable multiplier, it looks like I need to adjust the load meter based on what multiplier you're using compared to the default multiplier. When you are using a multiplier of 12 and your default multiplier is 9 then I think the load meter will be reading about (9/12) X 100% or ~75% when running 4 threads of Prime95 Small FFTs.

The internal timer I'm using seems to get screwed up when you change the multi on the Core 2 QX processors but I think it's fine on all Core i7 processors, including the 965 Extreme.

It should be very easy to correct for this and come up with a quick fix for you if you can help me out. The Intel documentation is like looking at Swiss cheese. A lot of information is labeled as Reserved or in other words, "Good luck trying to figure out what's hiding in this MSR."

Can you read a few MSRs for me with your multi locked to 12, at its default of 9 and maybe even locked to a number less than 9.

The 3 MSRs I'm interested in are:
0x17
0x15F
0x198

You can run 3 instances of MSR Tool at the same time so you'll only need to send me or post 3 screen shots. I think there should be enough information in those MSRs to take care of this problem. A 12X multi in hex should show up as the letter C in some of those MSRs when you are using that one.

The last person to complain about the Load meter had a QX6800 but I didn't put 2 and 2 together then. Using the internal timers seems to be the most accurate way to calculate the load so I'd like to get this minor bug fixed.

DGill: I kind of like the new GPU button in version 3.06 for Nvidia owners so I'll try to go back to the old way of reading cores in the next release to get your problem fixed up. Hopefully the next fix will continue to work for darkzone as well.

pra: Stuck sensors are not unusual at all at lower temperatures. Intel only designed these sensors to work at higher temperatures to control thermal throttling and thermal shut down. As long as your CPU runs OK you don't have to worry about the temperature, especially low load or idle temperatures.

[Falkentyne]

Ok I have some screenshots.

x12:


x10:


x9:


x6:


Do these help?

[unclewebb]

Falkentyne: That's going to help a lot and will get me started with improving the load meter accuracy for QX owners. We might have to do a little bit of fine tuning but I should have a beta out by tomorrow for some initial testing.

Could you send me one more screen shot at x10 or x12 with Speedstep or C1E enabled? When these are enabled, does your multi drop down to 6.0 at idle like most Core 2 based processors do? I just want to make sure that I cover as many different situations as possible. So far it's looking like a simple fix is going to finally solve this problem.

[Falkentyne]

Speedstep/c1E don't work on the P5W Dh board with 45nm cpu's though
EIST function doesn't even appear with a 45nm.

At least I could never get it to work. I think others couldn't get it working either even though it worked with 65nm cpu's. Heck I remember even when it worked, it sometimes was buggy; with my X6800, sometimes (rarely) it put my vcore at 1.15v even when it was set to 1.45 in bios... O_O, and that was with speedstep/C1E/EIST all disabled.

I'll try again though, right now...
I'll set voltage to auto, and speedstep and c1e (or TM1?) to enabled....

Um...no dice.....
Speedstep not present in BIOS with 45nm.
EIST not present.
C1E set to auto.
Vcore set to auto...
In BIOS 2406, speedstep and EIST were available but didn't do anything at all.
On 2704 (and 2801) theyre gone if youre using 45nm....

Booted in windows with C1E set to auto and auto vcore, cpu ratio adjustment disabled, and no change in voltage, multiplier or cpu speed. (vcore was 1.216v).

[unclewebb]

No problem. If Speedstep doesn't work on your board / CPU then I guess I won't have to worry about that.

One thing I mentioned recently is that sometimes you have to go into the Power Options in the Control Panel and with XP you need to set it to a Mobile processor to get Speedstep to work properly. On some boards this over rides any bios settings and will drop your multi to 6.0 at idle. In Vista you need to set the Minimum processor state to a low number like 50% to accomplish the same thing. RMClock 2.30 usually allows you to toggle C1E when you are in Windows but Speedstep on my board is read only. I should have time today to fix this and then you can do some beta testing to see how it works.

[unclewebb]

This is a bug fix version to try and get the load meter working properly on the QX processors when different multipliers are being used.
If I did a good job then it should also be able to find all 4 cores of DGill and darkzone.
http://www.fileden.com/files/2008/3/...alTempBeta.zip

If you have mfc71u.dll installed in C:\Windows\System32 then you can use the smaller Beta2 version.
http://www.fileden.com/files/2008/3/...lTempBeta2.zip

Falkentyne: Time for some beta testing. Try running 3 threads of Prime95 Small FFTs at different multipliers. The Load should be fairly steady at about 75% if not too much else is going on in the background. Try it at the default 9.0 multi and at a multi above and below the default. I'm pretty sure it will work when above 9.0 but not 100% sure it will work correctly when a multiplier is set to less than 9.0. It might need a tweak for the low end. Thanks for your help.

You can try 4 threads as well but I think 3 threads will clearly show if there is a problem when using lower multipliers.
If any new bugs have been created then report them too.

[Falkentyne]

I Just used crystalcpuid to change the multiplier and everything was fine in realtemp. Used 6x to 10x.
I have to lower FSB to go higher ...but I think everything will be ok.
might as well try a quick 150x20....

*working fine after reboot at 150x20.
99.9% load in quad prime, 74.9% in 3 threads.

[unclewebb]

Thanks for confirming that Falkentyne.
Now I just need to transfer a few new lines of code to the RealTemp / RivaTuner plug-in to get that fixed up as well for QX owners.

[DGill]

This is a bug fix version to try and get the load meter working properly on the QX processors when different multipliers are being used.
If I did a good job then it should also be able to find all 4 cores of DGill and darkzone.

If you have mfc71u.dll installed in C:\Windows\System32 then you can use
If any new bugs have been created then report them too.

Still only 3 cores showing with 3.07 but version 3.00 works well

[darkzone]

Works for me

[unclewebb]

Still only 3 cores showing with 3.07

burebista isn't going to be happy. I'll have to delay his individual core load for RivaTuner project while I get new RealTemp running reliably like RealTemp 3.00.

[burebista]

No rush Kevin. Take your time, you know I'm patient.
Focus on main RealTemp features.

[oggie]

Quick question, on old RealTemp had to change TJ max to 95 for E8600, do i have to do same on RealTemp3.00 or just leave it at 100

[PRJ]

Did I just have a Crash when playing Warhead, or did Realtemp shut the computer down, I did'nt get BSOD.

Can Real temp shut down the computer, if GPU temps are too high...?

BTW Thanks for the program

PRJ