Sign in with
Sign up | Sign in
Your question
Solved

Why is core 0 10 degrees warmer?

Tags:
Last response: in CPUs
Share
September 19, 2010 4:45:06 AM

Why is core #0 is ten degrees warmer than the rest of the cores of my CPU.

I've had this CPU almost 2 years now and noticed that core 0 has always been the warmest out of all the cores on the CPU. I'm runing the core 2 quad Q9650 at 3GHz. I want to overclock again as I'm hungry for speed. until either buldozer comes out or the Intel's new CPU in the begining of next year. Could it be a faulse reading?

please note that at Idle the temp for Core 0 is around the same as the other cores.

More about : core degrees warmer

a b à CPUs
September 19, 2010 5:44:35 AM

Seems pretty obvious to me that core 0 will be the most worked core in single thread applications.
I don't think you have anything to worry about, this is fairly normal behavior for quad core CPU's. That's just my opinion, it doesn't really look abnormal to me.
m
0
l
a b à CPUs
September 19, 2010 6:14:48 AM

Could be a number of reasons why that is occurring. The least likely reason is that Core 0 is actually noticeably hotter than the rest.
m
0
l
Related resources
a c 126 à CPUs
September 19, 2010 6:42:06 AM

buzznut said:
Seems pretty obvious to me that core 0 will be the most worked core in single thread applications.
I don't think you have anything to worry about, this is fairly normal behavior for quad core CPU's. That's just my opinion, it doesn't really look abnormal to me.


This is normally true when the core is the only one getting the work. In this case, all 4 cores are getting the same heavy work load and in Prime95s case, it should be core 0 and 1 at the same temp and core 2 and 3 at the same temp i.e. core 0/1 @ 45c and core 2/3 @ 49c.

I do have a few questions for you though.

What heatsink are you using? Stock or an after market? Also did you make sure there was proper thermal paste applied? Those temps alone seem pretty high for a Q9650 45nm based CPU. My Q6600 65nm CPU runs cooler than that when under Prime95 (55c) and its overclocked to 3GHz.

I would check the heatsink. Also if you have the pushpin type it might have gotten loose on one side which would make one core possibly hotter than the rest.
m
0
l
a b à CPUs
September 19, 2010 6:53:16 AM

Given that Core 0 will be part of the same silicon as one other core, if it was really that hot then one of the other cores should also be a similar temperature.
m
0
l
a c 126 à CPUs
September 19, 2010 6:55:21 AM

^I agree but I have seen dual cores that have one core slightly hotter since the HSF was not properly mounted to just one side or the CPU was too 'round' for the HSF to sit flat on it.

Thats why I think ti could be a HSF issue.

Or he could try a second temp monitor program like sSeedfan.
m
0
l

Best solution

a b à CPUs
September 19, 2010 7:00:35 AM

Speedfan will still be reading the same data. It's always a good idea to reseat the heat sink when temperature "problems" occur just in case, but it's important to keep in mind that accuracy was never part of the design plan for the sensors.
Share
September 19, 2010 8:08:22 AM

well I did a sensor test on Real Temp and the sensor movment is core #0 15, core #1 8, core #2 8, and core #3 11. I'm back with the vigor monsoon III LT. I lost the back plate for the CPU so I went to a hardware store and got new bolts and washers. I use AS5 for grease. The heat sink is on as tight as it gose. this is the first time sence I went used this heat sink a year ago when I thought one of my fans were dying that the temps is like this. I think the highest temp was 65c one stock setting. and I had it all the way up to 4GHz at about 75c. So now I'm wondering if the threamal diode is dead.
m
0
l
a b à CPUs
September 19, 2010 9:09:33 AM

Run the full sensor Cooldown test (it's on the same page IIRC). The sensor movement part is rather deprecated and doesn't provide nearly as meaningful information.
m
0
l
September 19, 2010 9:30:58 AM

where would I get the full sensor test?
m
0
l
a b à CPUs
September 19, 2010 12:26:33 PM

You have Real Temp 3.40 right? Open up the sensor test page and down the bottom will be a button that links you to where you can get Prime95 (which you already have so don't worry about that) and a button to start the test. It runs for about 5-10 minutes (make sure you have no active processes running while you do this) and varies the load on your CPU, monitoring the DTS readings as it goes. That way you can see your sensor reading across the whole load range, rather than just at idle or just at full load. It's very handy for finding problem sensors. Post the results when you've done it.
m
0
l
September 19, 2010 1:49:11 PM

ok I have completed the stress test and I don't get it. Core #0 is hotter while under load but cooler while at Idle.



m
0
l
a b à CPUs
September 19, 2010 1:54:00 PM

When I said post the results, I meant a screenshot of the application with the completed cooldown test. ;)  I know what I'm looking for but I can't see it without the results :) 

EDIT: Looks like I posted a few seconds too late.
m
0
l
a b à CPUs
September 19, 2010 2:19:41 PM

Note that the graph shows the delta to the maximum temperature for each core vs load, not the temperature vs load. Thus higher is better.



It's pretty easy to see what's going on here. Cores 2 and 3 track together perfectly, such that one line hides the other. You can see that Cores 0 and 1 follow a similar shape to each other but with an increasing distance between them as load increases. This is rather odd because normally you'd expect that the readings would converge as load and temperature increases (sensors are more generally accurate at higher temperatures), but here the opposite is true.

Your earlier post makes it sound like you swapped out the heatsink, but did you mean that you changed it at an earlier time or after your discovered this problem and you were suggested to reseat it? If you haven't done so, reseat it and see if your results are similar. If they are, it's possible that the heatspreader on the CPU and/or the heatsink base is not flat enough. It could also just be dodgy sensors. 45nm Core 2 CPUs are well known to have dodgy sensors, but typically the problem is at idle, not load. That doesn't rule out the plausibility of it occuring in some cases though.
m
0
l
September 19, 2010 8:06:56 PM

that is what I"m tinking. becuase I had the Stock cooler on this cpu the last time too and the same thing. Core #0 was about 5-10c higher at load than Idle. So I think it's either A CPU or Heatsink is not meeting together properly, or B it could be my BIOS even though the latest BIOS for me is an Update for the Express Gate. But I know when I got this Chip Core 0 has been the highest at load and around the same temp at Idle.

I'm wondering if I should call Intel and ask them. but they would probally not know. or I'll set my CPU's multi at 6.0 and voltage at 1.10v

btw the CPU validation below is this computer from about a year ago when Core #0 was about 76c and Core 1 70c as well as the other 2 cores.
m
0
l
September 20, 2010 12:12:17 AM

ok I also have an Intel E8400 and did some of the same tests. I'll post with comparing apples to apples I only used 2 cores on the Intel Q9650 @ 2GHz which the first E8400 picture will be the E8400 at 2GHz and the next E8400 at 3GHz

E8400 @ 2GHz


Q9650 2cores @ 2GHz


E8400 @ 3GHz


Q9650 4 cores @ 2GHz


I'm going to do a quick test on the computer I built my mom but it's a celeron @ 2.5GHz
m
0
l
a c 126 à CPUs
September 20, 2010 1:41:55 AM



So thats the layout of a Core 2 Q9650.

I would assume that Core 0 is the farthest left wich means that the CPUs IHS is probably a bit round.

But then again it could be a stuck temp sensor.
m
0
l
a b à CPUs
September 20, 2010 2:12:11 AM

A stuck temp sensor would stop the temperature from going below a certain point. Without better cooling to get the CPU even lower at idle we won't see where it sticks, but at the moment none of the sensors appear stuck.

That last screenshot looks far, far more typical than the previous one. Notice that 3 cores converge at full load. The previous run may have had additional processes interfering with the results, perhaps including task manager. Core 0 is an anomaly in this case, and there's no way that Core 1 would be no hotter than Core 2 and 3 when Core 0 is apparently 7C hotter given that they are physically the same piece of silicon. I'd put little faith in that sensor.

Oh, and on a side note, don't bother talking to Intel about it. They will simply tell you that they don't care about what temperatures you're apparently getting (in a nice, professional manner) because those sensors are not designed for reading temperatures.
m
0
l
September 20, 2010 3:35:01 AM

so would you assume that core 0's temp sensor is not calibrated probally or has a higher Tjmax? As I saw that some cores could have a higher TJmax.

Btw Randomizer that last run was at 2GHz at 1.1v.
m
0
l
a b à CPUs
September 20, 2010 4:28:44 AM

Ah, I didn't notice that before. In that case I think your first run was actually "normal" after all, because the same convergence around 50 delta to Tj Max can be seen in the graph above. Your sensors just seem to be all over the place then.

Post over in this thread. unclewebb (Real Temp dev) will be able to help you more than I can. I'm a bit rusty on this stuff now and he's far better at working out what's going on with sensors that are unusual. Maybe I'm just missing something obvious. It happens :) 
m
0
l
a b à CPUs
September 20, 2010 12:13:27 PM

This is normal for most quads, especially on Intel CPUS: Core 0 ends up doing the most work by default, and the second core on the same die is usually warmer as a result. Its normal for two cores to be 5-10C warmer then the rest of the CPU. My QX9650 has the same thing, as do most quads.
m
0
l
a b à CPUs
September 20, 2010 4:29:44 PM

From the testing I've done, my best guess is that Intel does not set TJMax equally across all 4 cores of a Quad. Intel actually invented a term called TJ target to try to explain this. There TJ target might be 100C but that does not guarantee that the actual TJ Max is that value. It might be 10C or 20C different on some CPUs. Intel has never published any information about the amount of error in their calibration process so without that, accurate core temperatures is mostly a guessing game.

That's the first problem with these sensors. The next problem is that the data they put out doesn't change linearly with changes in core temperature. The further you get away from the calibration point, the more these sensors can wander. When you have two sensors that are 50 degrees away from the calibration point, both sensors might be the exact same temperature but one might read 5 degrees too low so it would show 45C while the sensors right beside it might read 5C too high so it will report 55C. You've got a 10C difference right there and when you combine that with TJMax being set differently from core to core, no software is going to be able to report 100% accurate core temperatures from idle to TJMax.

The sensors on the original 65nm Core 2 CPUs were a lot better than this and the 45nm sensors used on the Core i7-900 series were excellent. The latest 32nm Core i sensors are showing lots of slope error again as well as variations in TJMax.

All of these sensors were only designed by Intel to trigger thermal throttling and thermal shutdown. They were never intended to be used to report accurate core temperatures and the 45nm Core 2 Duo and Quad sensors are about the worst sensors Intel has used on any CPU. They can also get completely stuck when they get too far away from the calibration point and will keep reporting the same thing even as the CPU core temperature gets cooler and cooler.

Core 0 tends to be the most accurate on most CPUs. On a lot of the Core 2 Quads that I've seen, it's almost if Intel deliberately sets TJMax slightly higher on core 2 and core 3 compared to core 0 and core 1. My theory is that they do this to better control thermal throttling. This would help prevent all 4 cores reaching the thermal throttling point at the same time and would give a more gradual reduction in performance which would be a lot less noticeable to the end user. In these Quads you can operate the first two cores with a different multiplier than the second two cores so it's possible to only throttle 2 of the 4 cores which is usually enough to keep the core temperature under control in most situations.

A lot of what you are seeing has nothing to do with differences in actual core temperature. It's mostly sensor error and the fact that these sensors are being used for a purpose that they were not designed to be used for.
m
0
l
September 21, 2010 12:17:32 AM

you see I am scared that over clocking my chip to 3.8GHz which was not done sense last year because of the fan on the heat sink was making terrible noise. I want the chip to last until Bulldozer and Intel's next Architecture is out before I redo my rig.
m
0
l
a b à CPUs
September 21, 2010 4:37:32 AM

It's pretty much impossible to kill a Core 2 with heat. Voltage, yes, but they keep themselves under control pretty well in the temp department. You could run the chip with no heatsink and it wouldn't burn up on you. It would just run very slowly, or simply shut down.
m
0
l
a b à CPUs
September 21, 2010 7:09:43 AM

I find users worry way too much about core temperature. If Intel was that concerned they would have used a hell of a lot better quality sensors that what they do.

Here's a 3 hours torture test of my E8400. I did some overclocking and over volting and disconnected the CPU fan just to make sure I created plenty of heat. The thing ran 100% reliably for 3 hours with zero errors in Prime95 Small FFTs until I finally got bored and stopped this extreme stress test. The CPU spent the majority of the time bouncing off the Intel thermal throttling point and never went beyond that. It remained 25C away from the thermal shutdown point so there were definitely no worries.

As randomizer said, voltage kills chips but high temperatures rarely do. When you overclock beyond 3.6GHz, you will find that you have to start reducing the maximum core temperature to remain 100% Prime stable. The more you want to overclock, the cooler you will have to run to remain stable. This forces you to keep your core temperature well below the thermal throttling point so overclocking is actually a good thing. It prevents your CPU from running too hot. Your CPU will either not be stable if you try to overclock it and it is too hot or it will simply shutdown, long before the Intel specified thermal shutdown temperature which isn't until 125C typically.

m
0
l
September 21, 2010 7:41:51 AM

Happens on my i7 930, Core 0 is hotter then core 1, core 1 is hotter than core 2 etc..

Ive presumed that Core 0 is the very first core and takes all the work load first to then spread it around as and when Intel Technology of using multicore kicks in (or however it all works)
m
0
l
September 21, 2010 9:09:27 AM

cool. I'll oc my Q9650 to 3.8GHz until I have reviews on both Sandy Bridge and Bulldozer.
m
0
l
September 21, 2010 9:34:43 AM

Good Luck Mate :) 
m
0
l
a b à CPUs
September 21, 2010 5:03:51 PM

reccy: The sensors on the newer Core i7-900 series tend to be extremely accurate and repeatable. When running a program like Prime95 with the Small FFTs option, the majority of these CPUs will show an almost exact 5C difference between core 0 and core 3. I've always believed that this is evidence that Intel deliberately sets TJMax 5C higher on the last core. If core 0 is 100C then core 3 is set by Intel to 105C. I believe they do this to better control thermal throttling but I have zero documentation from Intel to back this up.

When you have 4 cores sitting side by side, all running the exact same app at full load, it is impossible for the two end cores to be 5C different in their actual temperature. These CPUs are so tiny now that even when one core is doing absolutely nothing, it's core temperature will heat up very close to the same temperature as the surrounding cores. When all 4 cores are running at full load, the actual temperature of the 4 will be equal. Any reported temperature difference you are seeing is related to the undocumented calibration procedure that Intel uses.

This same sort of fixed difference was very clear on the early 65nm Core 2 Quads like the Q6600.

On the Core i7, core 0 tends to be the most accurate with the two center cores during a full load test being a degree or two less than the first core. The two center cores tend to be closer to core 0 than core 3. Whenever this relationship doesn't hold true, I'd be looking at how the thermal paste was installed and I'd be looking at how square the IHS is.
m
0
l
September 21, 2010 11:03:56 PM

I found my problem it is my heatsink I did a test. lay computer flat on the desk put some force on the Heat Sink and wham! Core 0 decied to cool down to near that of the other cores.

For some odd reason these bolts dose not want to tighten down
m
0
l
September 22, 2010 12:04:12 AM

new temps after heatsink tightening down

3.2GHz


Full temp at 3.6GHz


stress test at 3.6GHz



Finnally
m
0
l
a c 126 à CPUs
September 22, 2010 2:01:00 AM

I had a feeling. The layout of the CPU is like so:

C0/C1 C2/C3

The heatsink was not tight enough towards Core 0 but had enough thermal paste to ensure proper heat dissipation for Core 1 and probably gave plenty for Core 2/3. By making the heatsink tight enough, it put the pressure needed for Core 0 to finally get the proper temp.

If I were you though, I would probably invest in a new heatsink that comes with a backplate. I know its extra money as well as work but its worth it.

Thats a killer CPU you have, wish I had the funds to upgrade to it from my Q6600 but I guess I can wait it out till 22nm hits.
m
0
l
September 22, 2010 2:57:24 AM

you see I had a backplate for this one but I lost it so I have ruber washers and where 1 screw was not getting tighter I put paper on top of a washer so the heat sink would make better contact with the CPU.
m
0
l
a c 126 à CPUs
September 22, 2010 3:04:48 AM

That could work but paper is paper. Its weak and in heat like that can fall apart very fast. Then you will be back in the same place you were now.

Its better to either invest in a new heatsink or see if there is just a backplate for sale on ebay.
m
0
l
September 22, 2010 3:20:25 AM

thing is I don't feel like taking out my motherboard again. I had taken it out like 5 times the past two days. If I have to take it out again I may drill a hole where the CPU heatsink would be so it'll be easier to get at.

I found one at Ebay. I'll buy it friday.
m
0
l
September 23, 2010 1:33:13 AM

Best answer selected by blackpanther26.
m
0
l
!