Real Temp - New temp program for Intel Core processors

[Hoss331]

Let me know if this is right. At 1600mhz and about 1.1v my temps come down to 37 37 38 38 at 24 amb., so going by RGEs chart I should probably use TJmax 95 instead of 100? My temps come pretty close to each other at max temps but the first two cores always seem to lag behind at lower settings, mainly core 1 which is usually a few degrees lower. Ive ran the cool down test at lower settings and I know core 1 gets stuck at 63. Would setting TJmax to 95 and correcting core0 +1 and core1 +2 be right?

[unclewebb]

Intel says TJMax for your Q9550 is 100C so I'd leave that as is. Your full load temps look nicely balanced compared to many Quads.

If you've tested with your case open at 1600 to 2000 MHz and 1.10 volts for the core and your idle temps seem 5C too high then you can go into the Settings window and set some negative idle calibration factors. I try to balance out my idle temps during this test which seems to work pretty good.

Personally, if I had a Quad running over 4.1GHz, I wouldn't worry too much about trying to get the idle temps perfect. Nice OC.

[cky2k6]

With my core i7 920, real temp reads the bclock 10mhz too low. It happens at stock as well as my oc.

[unclewebb]

I just changed the MHz code to try to get better Mobile CPU support. I tried not to screw anything up but anything is possible.
Post a screen shot if you can of RealTemp and CPU-Z at full load to show the problem.

[rge]

With my core i7 920, real temp reads the bclock 10mhz too low. It happens at stock as well as my oc.

Can you post a pic/print screen up so can see?

Edit: I must be really slow typist

[unclewebb]

rge found one issue where the MHz isn't updated during the Sensor movement or CPU Cool Down Test. That was by design. I did this for more accurate results during these tests but I think I can leave this on now without it causing a problem. rge is testing this out at the moment.

The MHz code is more discriminating now when it sees a MHz number it doesn't like. This was put in there to help out the Mobile processors but it might need a little bit of fine tuning. If anyone is having problems and has some time to do some testing just let me know.

[rge]

RC5 seems to have solved the problem. It was only during the idle part of test where mhz went awry, but the new version reads correctly 4.2 whole time.

[randomizer]

I think that option disappears if you don't have EIST enabled. If you want your multiplier steady at 6.0 when idle, you need to enable C1E and EIST and set up Windows appropriately. If I have the Minimum processor state set to 100%, my 4 cores on my Q6600 will continuously dance between 6.0 and 9.0. Depending on your setup, your multiplier might not be as stable as you've always thought it was. I've decided to let RealTemp tell it like it is so users can make adjustments to their settings to get the multi they like.

Last time I had both enabled, CPU-Z reported my multi at a constant 6x. I really don't need either enabled I guess, but C1E allows higher overclocks... strangely. I guess that turns most OC guides on their head

[unclewebb]

I did some more testing today and found my board has no problem sitting stable at 6.0 when idle with EIST enabled and C1E disabled or 9.0 with EIST disabled and C1E disabled. When both C1E and EIST are enabled, then I have to go into the power options to make sure that is set appropriately or else I'll get the dancing multipliers.

Thanks rge for testing that out.

On my old P5B board, the reported MHz are a hair more stable compared to a couple of RC versions ago, especially when running SetFSB. Very accurate as well.

[rge]

I did some more testing today and found my board has no problem sitting stable at 6.0 when idle with EIST enabled and C1E disabled or 9.0 with EIST disabled and C1E disabled. When both C1E and EIST are enabled, then I have to go into the power options to make sure that is set appropriately or else I'll get the dancing multipliers.

Thanks rge for testing that out.

On my old P5B board, the reported MHz are a hair more stable compared to a couple of RC versions ago, especially when running SetFSB. Very accurate as well.

Interesting...I tried that with my core i7 and XP. With C1E enabled +/- EIST enabled I get dancing multi. With C1E disabled and EIST enabled I get 21 multi (bios setting). Seems C1E has to be enabled for my mobo, i7, XP combo to see any decrease in multi...at least in standard desktop power option.

[stasio]

It's RC4 or RC5?
Btw,Uncle must update first post

[randomizer]

It's RC4 or RC5?
Btw,Uncle must update first post

I think he needs to add a big red message in the OP saying:

"Refer to this post (which will be a link) if you have questions about which Tj Max to use". Then he wouldn't have to keep repeating himself

[bowman]

Finally got the TRUE on there..



Yep, that's nearly a 30 degree delta in load temperatures between the stock cooler and stock paste vs. the TRUE, 120mmx38mm and AS5. The stock paste looked and felt like concrete when I pulled that crap off. No CPU of mine ever going near a stock heatsink again..

[unclewebb]

Friends don't let friends use the OEM cooler on Core i7.

stasio: When people tell me about bugs they've found, I work with them one on one to get things sorted out. RC4 is the current release and RC5 was so rge could do some testing for me. I just noticed the first post. You're right, it definitely needs some updating! I guess I prefer to spend my time making RealTemp work better. The next official release is coming. I'm just waiting for some feedback about the MHz being displayed since that chunk of code was recently overhauled.

Edit: The beta link on the First page of this thread always downloads the latest beta, even when I haven't updated the version number for a while.

[bowman]

Not sure what the exact issues are, haven't been following the thread as well as I should.. But it certainly beats CPU-Z:



That's a lotta jiggamahurtz right there. It spikes like that for less than a second a few times per minute. No such thing in RealTemp that I can see..

[unclewebb]

RealTemp competing with the king of MHz, CPU-Z, is good but RealTemp finally beating CPU-Z is time for a celebration.


I guess all that development time with rge and burebista is starting to pay off.

Your screen shot is very unusual. I've never seen CPU-Z reporting anything like that before. You better post that screen shot on EBay and tell the world that you have a secret Intel CPU that can overclock to a gazillion MHz, on air!

[bowman]

RealTemp competing with the king of MHz, CPU-Z, is good but RealTemp finally beating CPU-Z is time for a celebration.


I guess all that development time with rge and burebista is starting to pay off.

Your screen shot is very unusual. I've never seen CPU-Z reporting anything like that before. You better post that screen shot on EBay and tell the world that you have a secret Intel CPU that can overclock to a gazillion MHz, on air!

Well, yeah. The number is so high it doesn't fit in the box, so I don't really know how much we're talking about. I think we're beyond the point where Doc Brown tells us we're about to see some serious , though.

It might be because I enabled a couple of the special sauce voodoo options in the BIOS to get the clock stable. I can't even remember their name. You know the ones, clock skew, load line calibration, spectrum spread, magic dust, god knows what. I enabled two of those. I'll have to check to remember what they're called.

[rge]

bowman, you should do a cpuz validation at that mhz...and submit it. You might have a world record for a couple days til someone pulls it down.

The cpuz bug worked for spyros, he had world record at cpuz with E8400 OC'ed with 9x935+ FSB, until they finally deleted it, it was another cpuz bug.
http://valid.canardpc.com/show_oc.php?id=468500

And congrats to Unclewebb for hopping quickly on core i7 compatibility, definitely shows.

[unclewebb]

And congrats to Unclewebb for hopping quickly on core i7 compatibility, definitely shows.

I couldn't have done it without you rge and that Intel Turbo white paper you sent my way. Thanks for all of your testing and help.

I'd also liked to thank burebista who has sent me 984MB of video clips of the funny things that RealTemp used to do on his unique system. He sent me an updated video this morning and RT is looking good on his Core 2, even with 101 different power management options all fighting against each other.

[rge]

What is interesting, someone was asking about calibrating realtemp with a core i7 in another thread....I just turned my computer off for a while to let water cool to room temp, then fired it up at stock/auto/loaded default settings, and let idle for 15 minutes, then started realtemp and everest to monitor watts. This is with EIST/C1E enabled, says Im at 6.75W idle (all c6 etc deeper sleep states are disabled by default), and with 25C ambient, 26C internal case ambient with IR gun, here is reading in realtemp, 8C above ambient, 7C above internal case temps, nearly same to what we figured with quads/core duos. pic 1

In pic 2 I disabled hyperthreading, and enabled c3/c5/c6 deeper sleep states, and watts down to 3W, and temps now 7C above ambient, 6C above internal case temps.

[unclewebb]

It was always my theory that there wasn't a lot of difference in temperatures at idle between a wide range of Core 2 based CPUs. From Dual Cores with 1MB of cache to 65nm Quads, the difference only seems to be about +/- 1C at ultra idle. Interesting to see that Core i7 is following that trend. With the minimal number of watts being consumed, it's difficult to get a significantly better delta than what you're seeing.

Can you try running the XS Bench a couple of times with your Core i7 locked at a common speed like 20x200. I want to include a new baseline on this screen for Core i7 since they run faster clock for clock compared to Core 2. This benchmark is single threaded and small enough so that it fits in the cache. Any Core 2 at the same frequency should score pretty much the same thing. It's just a quick test to make sure your computer is running properly without having to find Super Pi.

On my dual boot system it shows Vista running about 0.5% faster than XP. I get very consistent results from run to run when I push the Enter key quickly twice when it's sitting at this screen:



I added the MHz to this screen last night. Almost time for a new RC release.

rge: Try enabling Mobile in the power options to see if that gets your multi down to 12. You might need to do that to get all of the advanced C states working at their best.

[unclewebb]

You can always tell when a program starts getting good. Cracks for it start to appear on the internet.



It's FREE you dumb asses.

The most ridiculous thing is that it says 52931 have downloaded this file. I wanted to see what they were downloading but when I clicked on the link they told me to get out my credit card. Dumb, dumb, dumb.

Are there really this many dumb asses in the world?

[bowman]

You can always tell when a program starts getting good. Cracks for it start to appear on the internet.



It's FREE you dumb asses.

The most ridiculous thing is that it says 52931 have downloaded this file. I wanted to see what they were downloading but when I clicked on the link they told me to get out my credit card. Dumb, dumb, dumb.

Are there really this many dumb asses in the world?

Nope, it's just a bot page designed to scam credit cards.

At least it means your app is popular enough to make the bots grab the name and make a page.

[rge]

I changed laptop power and it sat on 12 multi, but watts stays same reads between 2W min and 3-4W max. Hard to catch watts reading though because lower half of sensor page in everest keeps disappearing, pops off whenever deeper sleeps are enabled...weird bug. Have to restart everest, hit sensor reading, wait 1 sec as it goes to idle watts then print screen...just before all sensor readings disappear.

I ran XS test on RT 4x at 20x200, UC 18x, mem 8X, 1600 ram, and results were 1607x3, 1606x1.

With my 24/7 settings, 21x200, UC 18x, mem 8x, 1600 ram results were 1688x3.

[unclewebb]

Nope, it's just a bot page designed to scam credit cards.

That's good to know. I thought the world was being taken over by idiots. More so than usual!

Thanks for the numbers rge.

Edit: You boosted your overclock by 5% and ended up with a 4.99% difference in performance based on your two Time values in your screen shot. How's that for a benchmark that is directly proportional to CPU performance?

At 4200 MHz your Core i7 can crunch numbers in a single threaded app 67.8% quicker compared to my Q6600 at 3000 MHz. A Core 2 would need to be running at 5 GHz to compete with that. It would definitely mop the floor with my Q6600 if this bench had 8 threads to keep your i7 busy. Do you have any Core 2 SuperPi times at about that speed to compare to your Core i7?