Temps uneven across cores on 9450?

D

defiant007

2[H]4U
Joined
Feb 27, 2006
Messages
3,497
Just wondering whether this a potential issue. At idle HWMonitor is reporting the following temps on my Q9450:

Core 0 - 56c
Core 1 - 45c
Core 2 - 34c
Core 3 - 41c

Running stress testing on Prime95 gives the following temps:

Core 0 - 71c
Core 1 - 62c
Core 2 - 51c
Core 3 - 47c

CPU is being cooled using a thermalright ultra 120 extreme coupled with a 120mm noctua fan. The HSF feels cool to the touch at load.

Should I be worried about these readings or is something funny going on with HWMonitor?
 
Holy shit, 22C delta between cores 0 and 2?!

If you bought this retail, I'd consider an exchange. That's just ridiculous.
 
UtopiA said:
Looks like a seating problem to me.
Click to expand...

+1... Definitely a heatsink installation problem. Since this is s a heatsink that's held on by screws then make sure they're in all the way (a quarter turn can make a huge difference!). A few degrees difference between the cores is normal, but this is cause for concern.
 
Yes.. definitely a big seating problem! Let us know how it works after you redo it.
 
Well haven't reseated it, only checked to make sure the screws were tight and they are as tight as they can go.

Now using Real Temp I get the following temps at stock (was previously running at 3.4 ghz):

Core 0 - 46c
Core 1 - 38c
Core 2 - 25c
Core 3 - 36c

Real Temp also reports the following sensor movements:

Core 0 - 11
Core 1 - 7
Core 2 - 11
Core 3 - 0

I.e. the sensor on core 3 is stuck!? Apparently this is a common issue with 45nm chips.

Will try reseating and reapplying thermal grease a bit later. These issues are really driving me nuts.
 
I don't know if this is any help, but i get the following readings idle at 3.2 Ghz (20% O/C)

Core 0 - 43c
Core 1 - 39c
Core 2 - 44c
Core 3 - 36c

Real Temp reports the following sensor movements:

Core 0 - 7
Core 1 - 7
Core 2 - 7
Core 3 - 5

I'm also using a TRUE 120 with noctua fan.
 
jasondean98 said:
I don't know if this is any help, but i get the following readings idle at 3.2 Ghz (20% O/C)

Core 0 - 43c
Core 1 - 39c
Core 2 - 44c
Core 3 - 36c

Real Temp reports the following sensor movements:

Core 0 - 7
Core 1 - 7
Core 2 - 7
Core 3 - 5

I'm also using a TRUE 120 with noctua fan.
Click to expand...

Yeah thanks, hopefully it is just a seating problem. My brother is going to come over and help me later with it as it is complete pain in the ass to remount the TRUE.
 
Well completely pulled my machine to bits, removed all the old thermal grease, reapplied new grease and remounted the HSF, but the temps are all exactly the same :mad:

Either the CPU is fucked, or its temp sensors are fucked.

In the bios the PC Health status menu reports the CPU temp as 32c.
 
CPU bios temps and the sensors that RealTemp and Core Temp are reading are different. None of these sensors are calibrated or 100% accurate from idle to full load.

The temp sensors are not f***ed. They're just not designed to give you accurate core temperatures, especially at idle. Intel designed these sensors for thermal throttling and thermal shut down control and that's about it. Any functionality beyond that is not guaranteed or even recommended. All 45nm Quad sensors do pretty much the same thing. Your sensors don't seem any worse than all of the rest of them. Actually, if you RMA your processor, your next one will probably be worse.

The latest version of RealTemp has a new sensor test that usually will give me enough information to come up with a suitable calibration for your CPU. You wasted all that time remounting your heatsink so why not go waste 10 minutes and run the CPU Cool Down Test and post your results.

http://www.fileden.com/files/2008/3/3/1794507/RealTempBeta.zip

Here's what some Q6600 sensors look like:

newcooldowntestvz2.png


They're not perfect either but at least they don't get stuck like the 45nm sensors do.
 
defiant007 said:
Well completely pulled my machine to bits, removed all the old thermal grease, reapplied new grease and remounted the HSF, but the temps are all exactly the same :mad:

Either the CPU is fucked, or its temp sensors are fucked.

In the bios the PC Health status menu reports the CPU temp as 32c.
Click to expand...

Remounting is fine, but did you check to make sure you assembled the clips on the heatsink properly?
 
unclewebb said:
The latest version of RealTemp has a new sensor test that usually will give me enough information to come up with a suitable calibration for your CPU. You wasted all that time remounting your heatsink so why not go waste 10 minutes and run the CPU Cool Down Test and post your results.
Click to expand...

9450tempsbb9.jpg
 
I smell a need for lapping, but I wouldn't do it on the CPU as it voids warranty, maybe you have a defective chip and should replace it instead.
 
Let's put the sand paper away for another minute or two!

Did you run Prime95 and use the Small FFT option for this test and run it on all 4 cores? The reason I ask is because your maximum load is only 82.3%. The screen shot for my Q6600 above shows 99.8% CPU Load and most of the screen shots I've seen show a similar number.

At the end of your test between the 0.0% stage and the Idle stage, did you start using your computer? It's very odd for the temperature to increase between these two stages. As I said, this one test can tell me a lot about your CPU but you have to let it do its thing for best accuracy.

What I see so far is that your core3 sensor is stuck at a Distance to TJMax of 59. As a CPU cools down, it is very typical for these sensors to stick at one number. It is very unlikely that this core will ever report a temperature of less than 41C even if you go to the North Pole with it.

Quads typically have two cores that have well balanced sensors and it's usually core0/core1. In your example core2/core3 are your balanced sensors but unfortunately core3 has that sticking problem. Uncorrected, as is, core2 is going to give you your most accurate temperatures and I would plan to use that as a baseline.

Core0 and core1 are a little messed up but at least they're not sticking. A messed up sensor can be calibrated to give some reasonably accurate core temperatures. A sticking sensor can not be calibrated.

Sensors that don't stick have two other issues. One is slope error where the sensor changes value at a different rate than the temperature changes at. I think Intel has stated that can amount to +/- 10% which is pretty significant but that's only part of the problem.

The other problem is that TJMax is not a fixed value across all 4 of your cores. Instead of using the term TJMax, Intel prefers the term TJ Target. The TJ Target might be 100C but TJ Max will be this number or higher. They have not specifically stated how much higher is possible but I have seen some 45nm Quads where TJ Max on one core will be around 110C while your core 2/core3 sensors are likely very close to 100C. The difference for 65nm Quads tends to be about 5C (100C / 105C).

Try running Prime95 with the Small FFTs option and see what Task Manager and RealTemp report for CPU Usage / Load. Something isn't quite right and I want to make sure it isn't a bug in RealTemp. If you can get the load up to the proper amount then exit Prime95 and try running this test again. You need to start Prime95 using RealTemp so it can be in control of it. With your cores at 99% load they should get a little hotter so the difference in TJMax will be easier to see. Let the test completely finish and turn off as much background stuff as possible while testing. This will let me calculate the slopes of each core so I can compare all of them.

I'm assuming that you know how to apply paste. Remounting your heatsink didn't change anything so most of the issue here is sensor error. You're going to be surprised how much more accurate and believable your temperatures are going to be after some calibration. Your Quad is no different than many others that I've seen.
 
Thanks unclewebb for your advice. You were right, there was something funny happening in the background when I ran the first sensor test. Theres something odd with my current XP install, CPU usage is showing 20% at idle....I think I may have to reformat.

Here are the results using my Vista drive.

q9450tempsth4.jpg
 
defiant007: That screen shot makes a lot more sense. Core3 is still stuck at a minimum of 41C.

Slope error can go either way. A core will either report at idle that it is at a higher temperature than it actually is or will report that it is at a lower temperature. As I said before, I think core2 should be used as your baseline. The easiest way to check its calibration at idle is to go into your bios and set your computer FSB to 333 MHz x 6.0 and manually lock your core voltage to as low as it will go. When I run CPU-Z it reports 1.08 volts on my board but anywhere close to that is good.

rge and I have done some testing at these settings and have found that in an open computer case with your CPU fan on high, your reported temperature should be about 8C above your air temperature near your computer at idle. By running your computer at those settings, heat output will be at a minimum and you will be able to easily compare your computer to the testing we've done. rge actually drilled a hole through the top of his IHS so he could mount a thermocouple to get some more accurate temperature readings.

Here is what he came up with for a guideline:

To calibrate temps for core 2 duo: First set vcore 1.1V bios +/-~.03v and 6x333 mhz manually or leave speedstep enabled. Then set idle so it reads listed degrees above ambient for given cooling:

COOLING............................IDLE DEGREES ABOVE AMBIENT
High end water...............................6C above ambient
High end air (true push/pull) .........6-7C
High end air (1fan)..........................7C
Mid air (zalman 9500)...................8-9C depending on fan rpm
Intel stock cooler......................10-11C

A Quad Core will be about 1C higher than these numbers. A Dual Core with 1MB of cache might be 1C less. We haven't tested a Core i7 yet but with the advanced C-States enabled, the temperature delta will be very similar to a 1MB Dual Core since some of the cores get turned off at idle.

Go do some testing and see what core2 has to say. It might be reading a little low at the above settings but will likely be pretty close.

Edit: For your XP install open up Task Manager and see what task is chewing up your CPU cycles. In Vista the other day I was at 20% while doing nothing and it turned out Media Player was doing some task in the background. I hate re-installing an OS when I don't have to. It might also be a virus scanner or similar program working in the background.

Here's a link to some videos of the type of testing rge has done:
http://www.youtube.com/user/rge42
 
i had almost the exact same problem with my q9300

http://hardforum.com/showthread.php?t=1376262

bout a 13C delta. the delta narrows at high temps so i think unclewebb is right in sensors simply not being that accurate for idle temps.

just make sure your peaks temps are in spec and you should be fine.

in retrospect ive noticed this to always be the case. 1 guy would have idle temps 10C cooler than another with the same cooler/cpu, and anything above that would be assumed to be abnormal even when everything is reseated for the umpteenth time. since they were single-core procs the variance between sensors, rather than chips/cooling setups, wouldnt be as obvious.

check your heatsink/heatspreader with a razor before you go so far as to start lapping. i reached the same point as you, but i checked the flatness thoroughly and it is flat enough to obviously not be the source of the "problem".
 
With a single or Dual Core most users barely notice the problems 45nm sensors have. When Intel has to put 4 cores under the hood, they seem to have a hell of a time finding 4 that match up reasonably well and the problems become pretty obvious. Great processors, horrible sensors. I prefer that to the opposite.
 
You must log in or register to reply here.
Back
Top