RealTemp General Discussion

[burebista]

Guys just my 2 cents thrown here.
Try to rely on distance to TJMax. Why? Because your CPU thermal behavior is dictated by distance to TJMax. This is what your CPU show to OS.
This is my example which I put on every forum I visit.
E8400, Ninja rev.B with Bolt-Thru kit (now cooled by a 400 RPM fan) at 22°C room temperature in a P182 case.

No way I could have 42°C in idle with CPU frequency 1600 MHz and 0.944V Vcore so I calibrate it. Temperatures looks OK after calibration but keep an eye on distance to TJMax. It's the same. So why bother to seek perfect temperature when I have same distance to TJMax whatever I'm doing trying to calibrate my temps?
For me all is simple now: I'm looking at that distance to TJMax and I'm trying to keep it >20-30 and that's it. No brain hurt about TJMax value, no stress about high/low temperatures, I'm a happy camper now.

This is how it looks calibrated with a standard OC. Ninja was fanless when I take that screenshot.

 

[chooky]

I could be wrong, I often am. But isn't Distance to TJMax just whatever you have your TJMax set to in settings minus the temperature. So in your second image your TJMax setting is 94 which is a bit low for an E8400. Should be 100 shouldn't it?

What is wrong with Intel anyway it can make 400+ million transistors in an E8400 work together pretty well most of the time but it can't get a couple of diodes right!!! Its not as if this is bleeding edge technology we are talking about. Have they got it right in the i7?

 

[burebista]

But isn't Distance to TJMax just whatever you have your TJMax set to in settings minus the temperature.

Yep, but keep in mind that the only real value is distance to TJMax, both TJMax and temperature are arbitrary values. TJMax is a number pulled from hat by Intel and temperature is a result of that number minus what DTS reports (distance to TJMax).
So, you can impress your friends with low temperatures putting TJMax to 50 or you can scare them putting your TJMax to 150.

Have they got it right in the i7?

As far as I see all reports on different forums i7 sensors are better. Beside that TJMax is written in MSR now.

 

[F7P]

Linpack was written by Intel so it must be safe.

I learned one thing during this project. When not overclocked, Core 2 Dual and Quad core processors can reliably run at pretty much any temperature you throw at them. Intel might have cheaped out on the temperature sensors that they're using but the CPUs themselves are close to bullet proof.

I once ran Prime95 Small FFTs on my E8400 for 3 hours. I deliberately disconnected the CPU fan to try and create some excitement. The core temperature went up to 98C but my computer continued to run flawlessly at this temperature until I got bored and finally stopped this test after 3 hours. I was more afraid I was going to warp my board than hurt my CPU.

The sensors are far from accurate and as long as your computer is 100% stable then there's no reason to give your core temperature a second look. I hate to tell people not to run my program but in your case, there's no need to run it. The software you use is never going to get the core temperature up to a dangerous level even if the CPU fan fails.

For calibration I'd use TJMax = 97, 104, 100, 100 and use calibration factors of about -8.0 for cores 1, 2 and 3. For core 0, use whatever calibration factor is necessary so that the 4 cores are pretty much equal at idle.

If your thermal paste is fresh and your case was closed during testing then I think the above numbers might be pretty close. Give it a try and see how your temps look at idle and when running Prime95 Small FFTs. All 4 core temperatures should be fairly close at idle or at full load.

Thank you. That balanced them all out around 29/30 degrees.
Can i make a small suggestion though, and if its silly forgive me ! In RealTemps main window could we have Core 0 Core 1 Core 2 etc above the temps. Because i wasnt sure if they went 0,1,2,3 !! When i configured it.
I will test it later with Prime95 and see how they balance under load.
One other thing..... stuck sensors..... how do you determine if they are stuck or ok. ? ( yes i am a noob to temp testing but have 18 years computing experience)

 

[F7P]

Thank you. That balanced them all out around 29/30 degrees.
Can i make a small suggestion though, and if its silly forgive me ! In RealTemps main window could we have Core 0 Core 1 Core 2 etc above the temps. Because i wasnt sure if they went 0,1,2,3 !! When i configured it.
I will test it later with Prime95 and see how they balance under load.
One other thing..... stuck sensors..... how do you determine if they are stuck or ok. ? ( yes i am a noob to temp testing but have 18 years computing experience)

OK ran Prime95 on Large FFT for an hour. Temps hit 42 degrees and went no higher. all cores stable at around 41/42/41/41/ etc...

 

[unclewebb]

F7P: It looks like your reported core temperatures are closer to the truth now and the 4 cores should move at a similar rate when going from idle to full load.

The Distance to TJMax number displayed by RealTemp is the raw data coming from the on chip sensors. This is the only data that the processor uses to determine when it should thermal throttle itself or do a thermal shut down if the core temperature gets really out of control. A thermal shut down doesn't happen until about 125C which is difficult to get up to unless your heatsink falls off.

burebista is right. Accurate core temperatures is a guessing game full of maybes and possible errors. The only important number is Distance to TJMax. Your CPU will start to thermal throttle and slow down when your processor heats up and this number counts down to about 3.

If you are not overclocking, always leave yourself a few degrees of head room away from this number and you'll never have a problem. You won't have to concern yourself about how accurate TJMax is or how much slope error your CPU sensors have or whether your sensors are stuck at low temperatures or anything else. As long as you have some head room your CPU will run at full speed.

I can run Prime95 Small FFTs on my E8400 at 95C at the default speed of 3000 MHz or even when mildly overclocked to 3600 MHz. This processor will also run Prime stable at 4000 MHz but it can't do that at 95C. In order to run Prime stable at 4000 MHz, I need to lower the maximum temperature to about 65C or 70C or leave about 30 to 35 degrees of head room instead of 5. If you are overclocking, maximize the amount of head room you have from the thermal throttling point and you will be able to reliably overclock your processor to its maximum.

All of these sensors will become stuck at some point as they cool down. Theoretically, these sensors can read down to -27C but some of them will stop moving at 40C or 50C. Your actual CPU core temperature might drop from 40C to 30C but the sensor will get stuck and keep reporting a constant number like 40C. If you're lucky, the sticking point will be low enough that in normal use you will never notice this problem. About the only time I notice my E8400 sensors getting stuck is when I drag my computer out to the backyard during the winter to do some overclocking.

The CPU Cool Down Test is a good way to find a stuck sensor. At each step of the test, you should see a higher Distance to TJMax number or a cooler temperature. When a sensor is stuck, the last 4 or 5 steps of this test will show the exact same number. Some 45nm sensors are truly horrible and every step of this test will show the exact same number from top to bottom but that's not very common.

If you go into the Settings Window, you'll see the buttons for color selection labeled Core 0, Core 1, Core 2, Core 3. I like keeping the main interface clean so you'll have to do some thinking to understand the Intel method for numbering cores.

 

[F7P]

Thank you unclewebb,
your help on here and every other contributors help is great, especially as the program is free and you spend so much time explaining it as well ! It is appreciated.

Finally my cool down test. Noting the last 5 steps, was i one of the very few lucky Q9300 owners who doesnt seem to have a stuck sensor ?

 

[unclewebb]

You're a lucky 45nm Quad owner. Your sensors are far better than most. I think I posted somewhere in this thread what a Cool Down Test looks like when you have sticking sensors. If I can find the pic I'll repost it here. My usual XS Forum is down for maintenance for the last day where most of my good pics are located.

Here's a good pic from page 3 of this thread:

It's pretty obvious to see core 0 getting stuck at a Distance to TJMax of 55 while core 1 continues to move as the CPU gets cooler. You definitely don't have any problems like that in your normal temperature range. Above the sticking point, these two sensors look like they're very well balanced which is normal when two cores are running the exact same code and sitting a couple of millimeters away from each other under the same heatsink and fan.

You'll probably need to bring your computer to a meat locker before your sensors will start sticking.

 

[boucay]

Hello unclewebb,

I was wondering if you can help me out?

I'm haveing problems with the temp. readings for my E8400 ver. E0; it seems the sensors are stuck at 41C. I used your beta 3.20 RC1 (with the case open) I ran the sensor test/CPU cooldown test, and the temperature max. reached was 44C.

Screenshot (sorry, it's not too clear, but it should be clear enough, I hope):

The only time the temperature actually moved was when I ran LinPack using OCCT, it went as high as 49C.

Screenshot (again, not to clear, sorry):

The ambient temperature was about 20C; it's spring right now, so it's not too hot at the moment. Also, my CPU temperature in BIOS fluctuates between 28C-30C, and the system temperature fluctuates between 37C-38C. There is no way the cores could be at 41C, at least I don't think it's correct.

Can you help me out please?

Thanks
boucay

 

[unclewebb]

boucay: There's not much I can do with your sensors. They're stuck at a Distance to TJMax of 59 or at a temperature of 41C. It's not possible for either sensor to report a temperature lower than 41C.

Above this temperature point, these sensors will be reasonably accurate.

Intel designed these sensors to control thermal throttling and thermal shut down and doesn't recommend using them to report accurate core temperatures. Some users get lucky and can use these sensors to report reasonably accurate idle temperatures and other users like yourself can only depend on them for reasonably accurate, full load temperatures as long as you get up over 41C.

 

[boucay]

Man, that sucks Anyway, thanks for your help.

Do you think it would be ok if I use the CPU case temp. (this is the reading I see in BIOS, right?) as a guage instead since the core sensors stuck at 41C?

I also meant to ask, based on the info. I provided, and the readings in BIOS, can you take an educated guess as to what the temps. of the cores are at idle?

 

[unclewebb]

Idle temperatures are completely unimportant. It doesn't matter whether you're idling at 30C or 40C. Even users with working sensors rarely take the time to test how well their sensors are calibrated. An error of +/- 10C at idle is typical so most of the temperature numbers you see on the internet are just meaningless random numbers.

You CPU bios TCase temperatue might be very accurate or it might be out to lunch like it is on my board. I think when my board came out Asus used inaccurate information from the core sensors of these chips to calibrate TCase which would be wrong. At full load running a stress test like Prime you should see TCase temperatures about 20C to 25C less than the core temperature but on my board, the TCase reading is almost 10C higher than the core temperature.

Above 41C, your core sensors work fine. They might be off by a few degrees but they're usually pretty close. The Distance to TJMax number is the only important data to keep an eye on. These CPUs typically start to thermal throttle and slow down when this number counts down to 3. As long as you leave yourself a little bit of headroom from this number and your computer is stable then you don't have to worry about your CPU temperature. These chips run great.

 

[boucay]

Alright, thanks. I will keep my eye on those TJMax numbers instead.

 

[lulwut^]

Hi. Can i please have so assistance in calibrating real temp before I start overclocking this system. I hope the picture below of the Cool Down will help you trying it help me in workings this out, Any help would be great.

I have no idea what i am looking at

 

[unclewebb]

Don't let the random temperature numbers keep you from overclocking your Quad. My best guess is that TJMax is not consistent across your 4 cores. That's pretty common with these.

For core 0 and core 1 there is a 5C offset so setting TJMax to 95C / 100C might be OK for those two. I would need to know more about your room temperature near your computer, type of cooling you're using, whether your case is open, graphic card or cards being used, etc. to come up with some more accurate calibration suggestions. Try reading the RealTemp docs here on TechPowerUp about calibration for some more information. When you start overclocking and get some more heat into your cores, the sensors will get up into a temperature range where they are more accurate and it will be easier to try and calibrate them then.

With 45nm Core processors you don't have to worry too much about core temperatures. It doesn't matter what temperature number it says. If you are Prime stable then you're good and if you're not stable then you need to reduce your overclock until you are. These CPUs let you know when you are pushing them too hard by becoming unstable at full load so as long as you're stable, no worries. Maximize your Distance to TJMax and you will also maximize your ability to overclock your CPU reliably.

 

[F7P]

Hello hows everyone doing ?

Ok.. quick question. I have my core voltage set to auto but wondered if setting it to 1.256 volts would be enough to run my system stably . I want to set it lower than auto but i am not sure how low i can go without upsetting anything. Lower means cooler i think.

Thank you.

SYS:
Running ....
Windows XP Pro SP3
Intel Core 2 Quad Q9300 1333FSB 6MB Cache Quad Core Core 64 Bit Processor
ASUS P5N-T Deluxe Motherboard
4GB DDR2 800MHz Dual Channel Ram
640GB (2 x 320GB) Serial ATA II Hard drive with 16MB Buffer
Dual Nvidia 8800GTX 1024MB PCI-E Graphics Running In SLI Mode

 

[unclewebb]

Prime95 Small FFTs is a good enough stability test for me. When overclocking I set my MHz to a number I'd like to run at and then add enough core voltage so that it's stable for a few hours while running Prime. No other day to day software I use creates that much heat so that test is good enough for my purposes. Others like using IBT / LinX / Linpack for testing but I don't run any software that creates anywhere near as much heat as this so I find it's a little overkill.

My board on the AUTO setting really adds a lot of extra core voltage when overclocking so I avoid using AUTO and always set the core voltage manually. Run as much voltage as you need to maintain stability. You don't need to use more than that. You won't be able to maximize your overclock with only 1.256 volts but you will create less heat with that setting.

The only time I'd use AUTO is if I wasn't overclocking at all. If you have Speedstep / C1E enabled, on most boards at default MHz, the AUTO setting will allow your core voltage to drop down to about 1.10 to 1.15 volts at idle. Check it out with CPU-Z. Vista users might have to wait a few minutes for their computer to settle down before it drops down to the lower voltage setting.

 

[danielkza]

What I specially like about using Linpack is that it usually fails much faster than Prime95 would: Just tested with a previously-checked failed overclock, 3.8 GHz @ 1.24v on my E7300. Takes only 2 iterations of Linpack using 1800MB of memory, aproximately 5 minutes, while Prime95 takes more than 20min to start showing errors. This can be a real time saver when guessing and tweaking settings for the first time, with no good references for working options, and when messing with memory and CPU at the same time, for instance.

 

[F7P]

Prime95 Small FFTs is a good enough stability test for me. When overclocking I set my MHz to a number I'd like to run at and then add enough core voltage so that it's stable for a few hours while running Prime. No other day to day software I use creates that much heat so that test is good enough for my purposes. Others like using IBT / LinX / Linpack for testing but I don't run any software that creates anywhere near as much heat as this so I find it's a little overkill.

My board on the AUTO setting really adds a lot of extra core voltage when overclocking so I avoid using AUTO and always set the core voltage manually. Run as much voltage as you need to maintain stability. You don't need to use more than that. You won't be able to maximize your overclock with only 1.256 volts but you will create less heat with that setting.

The only time I'd use AUTO is if I wasn't overclocking at all. If you have Speedstep / C1E enabled, on most boards at default MHz, the AUTO setting will allow your core voltage to drop down to about 1.10 to 1.15 volts at idle. Check it out with CPU-Z. Vista users might have to wait a few minutes for their computer to settle down before it drops down to the lower voltage setting.

I enabled Speedstep and C1E and now in RealTemp it tells me ..

333.34 x 6.0 occassionally jumping to 7.0 and 7.55 .... is this normal safe behaviour ?

Thanks again.

 

[unclewebb]

danielkza: You're definitely right that when trying to find some stable settings, Linpack will fail sooner than Prime95 or anything else I can think of. I use Linpack too when testing. I just don't use any software that fully works my 4 cores for an extended period of time like Linpack does. I like a quiet computer so I lock my CPU fan speed to a minimum and it's perfectly stable, even if it's not Linpack stable at that fan speed.

For me, Linpack is like dropping a brick on the gas pedal of your car and then waiting to see if your engine blows up. My car might not be brick stable but it's good enough to get me to the store to buy some groceries.

F7P: Speedstep is designed to drop your CPU core voltage and CPU multiplier down to 6.0 when your CPU is idle. It sounds like your computer is working the way Intel intended it to. In theory, this saves some energy. As soon as you apply a load to your computer, it is designed to instantly jump up to the maximum multiplier and it also raises the core voltage back up a millisecond or two before that so your computer will be perfectly stable when you are using Speedstep. It works good. Some users like this and some don't. Now you know what to turn on or off in the bios to enable this feature. A CPU can transition like this hundreds of times a second so RealTemp reports the average multiplier during every 1 second sample interval.

 

[stasio]

Notice that beta version 3.20 RC3 is out!
Btw,XS forum on line again

 

[lulwut^]

Try reading the RealTemp docs here on TechPowerUp about calibration for some more information.

I had a look at the write up and i did every thing he said to do and the temp with my case open was 26.5c and room temp was 26c. I have got a TRUE 120 with 1 fan, So my real idle is around 33/34 on the cores?

Also in Everest it is showing my CPU idle temp is 24c and the cores are 41/45/39/36

 

[unclewebb]

If your room temp is 26C then it's obvious that your CPU temp as reported by Everest can't be 24C or less than your air temperature. All temperature sensors have some error in them.

It also looks like your core temperatures are reported a little too high. That's pretty common with 45nm CPUs. My E8400 reports about the same amount too high as well.

In your previous screen shot you were still a good 15C degrees away from where these sensors start to become more accurate. If you start overclocking with additional core voltage then your peak numbers while running Prime95 Small FFTs will give us some more information about how much TJMax differs from core to core.

You could also forget about all this calibration nonsense and just overclock and be happy. With a True 120, you don't have to worry too much about core temperatures.

 

[lulwut^]

In your previous screen shot you were still a good 15C degrees away from where these sensors start to become more accurate. If you start overclocking with additional core voltage then your peak numbers while running Prime95 Small FFTs will give us some more information about how much TJMax differs from core to core.

You could also forget about all this calibration nonsense and just overclock and be happy. With a True 120, you don't have to worry too much about core temperatures.

Is this what you after? I really want to fined my true tamps of the CPU.

 

[unclewebb]

lulwut^: In your second CPU Cool Down Test your full load is at 90.2% while during your first test it was at 97.1%. Either you didn't run Prime Small FFTs or something else was going on. Just before the Idle stage you are at 9.2% in your second test while during your first test your CPU was at 2.8%. That's usually a sign that you've got too much junk running in the background or you're working on something else while you're testing.

You've got 4 sensors that work and you've given me enough data so I'll come up with a calibration attempt for you later today and then you can do some more testing after that to see how it looks. Your sensors don't have a lot of difference in slope from one to the other so it will ber easy to make some improvements in the accuracy of your reported temperatures. The majority of the error looks like differences in TJMax from one core to the next.