temp diffs between cores - call to quad core owners

[graysky]

Is it unusual for CPU cores in a Q6600 to range in temp by about 6C at idle to over 13 C under load?

Q6600 running @ 3.0 GHz (9x333) cooled with AS5 and Ultra-120 Extreme on a Asus P5B-Deluxe. Ballistix memory @ 4-4-4-12 (1x2 gigs), Antec P182. Room temp is about 66-67F (19 C).

From TAT idle:

Processor 0: DIGITAL TMP[ 40C ]
Processor 1: DIGITAL TMP[ 39C ]
Processor 2: DIGITAL TMP[ 34C ]
Processor 3: DIGITAL TMP[ 34C ]

And they are consistent when under load (4x orthos for about 40 min). Here the difference is about 13C. From TAT:

Processor 0: DIGITAL TMP[ 70C ]
Processor 1: DIGITAL TMP[ 66C ]
Processor 2: DIGITAL TMP[ 57C ]
Processor 3: DIGITAL TMP[ 57C ]

Here is about 20 min of x264.exe. Difference here is about 11 C. From TAT:

Processor 0: DIGITAL TMP[ 65C ]
Processor 1: DIGITAL TMP[ 61C ]
Processor 2: DIGITAL TMP[ 54C ]
Processor 3: DIGITAL TMP[ 54C ]

Does this imply a problem w/ my heatsink setup or that I didn't put enough TIM over the first two cores or that the IHS isn't quite right over the first two cores, etc., or is it normal? If you have a quad core chip, can you report your temps for the cores as per TAT or Core Temp (load temps preferably).

Thanks!

 

[graysky]

Update - I've reseated the HS a total of 3 times and I don't get a change so I'm thinking it's not that. Anyone out there care to post their numbers?

Thanks.

 

[bichi]

graysky,
- I see the same asymmetric temperature deltas across my cores too
- see pictures 50~53, which show loaded temps, both core and processor, CPU core usage, along with application generating load
- if you compare core usage against core temps, you can see core temps align with core usage (some apps vary core usage during task)

Link to pictures: http://www.esnips.com/web/ConroeReady

There is also a new "line" technique applying thermal compound on multi-cores:
http://www.arcticsilver.com/ins_route_step2intelcmq.html

 

[graysky]

This is a case of more or less even load on the cores (either with orthos where I manually set the cpu affinity or with x264.exe using auto threads which does a fantastic job maxing out all 4 cores). I'll have a look at your pics.

Thanks.

 

[bichi]

here's one, loaded with OCCT, which does a nice job smacking a Quad core, symmetrically:
http://premium1.uploadit.org/bichi007//P5WDH-1801-OCCT1109b-10min.jpg

Link to OCCT: http://www.ocbase.com/perestroika_en/

 

[tersome]

The on-die temp sensors are nowhere near 100% accurate. I know that amd's sensors are designed with a plus or minus 15c margin of error.

ultimately, it's futile to try to get all cores to read the same. The cores may actually be only 1 or 2c apart and, by remounting the heatsink to get the readings to match, you'd do more harm than good.

 

[JAG87]

4C is the most delta I have seen on my QX6700. idle temps right now are 51-51-51-49 while full load temps are 78-78-74-74

 

[Elfear]

I had to hardmount my waterblock and experiment with the tension on each side to get the temperatures fairly uniform on my QX6700. Idle and load temps are now with 1-2C at idle and load, at least according to Core Temp. With aircooling it's a little more difficult since the mount can't be adjusted on most. Not sure how the TR Ultra-120 mount works, but I suspect that's the case.

 

[lopri]

Well don't forget that not all cores are created equal. Especially for an architecture like Kentsfield. (2 separate dice) Even on single die, often times each core will have different performance/thermal characteristics.

 

[graysky]

Originally posted by: Elfear
I had to hardmount my waterblock and experiment with the tension on each side to get the temperatures fairly uniform on my QX6700. Idle and load temps are now with 1-2C at idle and load, at least according to Core Temp. With aircooling it's a little more difficult since the mount can't be adjusted on most. Not sure how the TR Ultra-120 mount works, but I suspect that's the case.

Interesting suggestion. The Ultra-120 Extreme does have 4 tension screws that can be tightened/loosened. I should mess with that.

 

[graysky]

After several iterations of seating/reseating I began to analyze the pattern left by the AS5 on the thermal spreader. It seems as though the area under the first set of cores was too thick and it didn't matter if I rotated the HS 180 degrees or not. So whatever imperfection on the HS was sort of located over one half of the thing. After I guessed at that, I decided to split the difference by rotating it 90 degrees (now it's East/West in my case) and tried again.

That brought the numbers much closer together and made me happy in the process.

Here are the TAT numbers of orthos 2x now:

Processor 0: DIGITAL TMP[ 64C ]
Processor 1: DIGITAL TMP[ 65C ]
Processor 2: DIGITAL TMP[ 60C ]
Processor 3: DIGITAL TMP[ 60C ]

And in speedfan:
Core 0: 51C
Core 1: 52C
Core 2: 48C
Core 3: 47C

 

[graysky]

Originally posted by: bichi
here's one, loaded with OCCT, which does a nice job smacking a Quad core, symmetrically:
http://premium1.uploadit.org/bichi007//P5WDH-1801-OCCT1109b-10min.jpg

Link to OCCT: http://www.ocbase.com/perestroika_en/

OCCT seems kinda cool. I don't want to install MBM or Everest but I do have speedfan installed. Have you been successful enabling it as the sensor in OCCT?

 

[bichi]

graysky,
- nope
- have tried a few times, but have concluded MBM5 with CoreTemp plugin changes temp sensor channels, where OCCT can't seem to find them
- OCCT does track and graphically log voltages, with most of the popular monitor programs

 

[Samus]

Originally posted by: JAG87
4C is the most delta I have seen on my QX6700. idle temps right now are 51-51-51-49 while full load temps are 78-78-74-74

pretty high idle temps. mine always drop to 40-42c/core with a skyth hsf.

my max core temps are around 74c, however, core 0 can hit 76c. core 1, 2, 3 never go above 74c.

using Xeon X3210

 

[graysky]

Originally posted by: Samuspretty high idle temps. mine always drop to 40-42c/core with a skyth hsf.

my max core temps are around 74c, however, core 0 can hit 76c. core 1, 2, 3 never go above 74c.

using Xeon X3210

I have since dropped them by lapping the IHS on the CPU and the HS.

Lapping the IHS thread
Lapping the HS thread