Technical > Computer Hardware Help > i7-2600K with one core running 10C cooler

i7-2600K with one core running 10C cooler

Core #1, why so cool?

The general idea is always either the probe is reading wrong (calibration, whatever) or the IHS is not mating with the HS well.

While they are close, Intel stated the tcase max of a CPU was say 72 on the older core 2 I think because the temperature variation could be as much as 30C. Quite a bit.

I think something with in the one core being different would be sort of weird... they all should be running the same clock speed and voltage with the same number of transistors... you would think that would not matter in temp as nothing could be really different but... eh... maybe.

reply to koitsu
If it shows the same temperature differential at idle and at load I would guess that it is an offset in the temperature sensor.

The idle one is most suspicious. When looked at as deviations from the ambient temperature the difference in power consumption between cores would have to be huge (~3x) for one to be 25C and another 35C. That's assuming 20C ambient, if it were higher ambient it's even worse (ie at 25C ambient a core at 25C can't be burning any significant power).

Intel will stonewall you because this isn't really a problem, as in if you didn't look at those temperatures you'd never know the difference the CPU works fine I'm assuming.

reply to koitsu
Now you have a reason to lap your cpu. =D.

reply to koitsu
Intel's digital thermal sensors (DTSs) are not designed to give an exact temperature reading of each core. They are only designed to be the most accurate at higher temperatures as they approach Tj Max (98 degrees Celsius for the 2600K.) I think you woul find if you were to run a highly stressful program that stresses all cores evenly (such as Prime95,) you would find that the difference in temperatures within the cores would be less.

Intel's sensors are not that great at showing an accurate idle temperature without using a program that can be calibrated and can compensate for the erratic data the DTSs feed it, such as RealTemp.

More info on it, here> »www.techpowerup.com/realtemp/docs.php

By the way, idle temperature doesn't mean squat, it is absolutely useless to worry about your idle temperature. What you need to make sure of is that the CPU cores do not get above 75-80 in any of the applications that you use.

My 2600K also has one sensor that is always "lower" than the rest. Same thing occurred on my old Q6600 as well. It's completely normal, and doesn't necessarily mean that your heatsink is not on properly or that you need to lap your CPU.

reply to koitsu
nothing wrong with it I had some XEONs that had upto 8C difference. one would be -1C another core would be 7C on air cooling
inside temp was about 33F heating cost are too high so it's not worth it to keep it warm.

reply to koitsu
Your temps are absolutely normal and its perfectly fine for one core to be hotter than the others.
Those are good temps and theres no need for concern.

reply to koitsu

quote:

---

Starting with Core i7, Intel has improved this situation by including the TJ Target goal within each CPU. This information is still only TJ Target though so there is no guarantee that the actual TJMax of a core is exactly equal to the value stored within each core of these processors but at least this is a step in the right direction.

---

reply to koitsu
You know what I think, koitsu...

The footprint of a modern processor is quite small and the cores are proximal, no less. I don't know how cores so closely packed could possibly vary an actual 10 degrees C. as they are just too close to each other. I think something is wrong with the calibration or reading. We're talking just millimeters, and virtually no distance between cores.

I can see 2 degrees INSTANTANEOUS variations between cores, but not a fairly static 10 degree delta. I don't see how that would be possible, they are all made from the same chunk of wafer, after all.

The ONLY way that they could vary that much is the peltier effect, and I really doubt that phsics problem exists between cores.
--
Join Teams Helix and Discovery. Rest in Peace, Leonard David Smith, my best friend, you are missed badly! Rest in peace, Pop, glad our last years were good. Please pray for Colin, he has ependymoma, a brain cancer, donate to a children's Hospital.

reply to signmeuptoo

Re: i7-2600K with one core running 10C cooler

reply to koitsu
The cores aren't made individually and then installed, nor are complete chips made even one CPU at a time. A hundred or more complete CPU dies are made at once, on a single wafer of silicon. These days one wafer can be as large as 200 mm in diameter. Each individual die on the wafer is a complete CPU, with all cores, cache and logic circuits for that CPU contained on that single piece of silicon. Each CPU is tested while they are still part of the wafer, using incredibly tiny needle point connectors to probe each bonding wire pad on the CPU. Die that totally fail testing at this stage are automatically marked with dye and then discarded at the next stage in manufacturing, when the dies are cut apart and made ready for packaging.

Once in their packaging the CPU's move on to final test. This is where Intel determines what speed level each can achieve and marks the package accordingly. The only difference between the "slow" i7's and the "fast" i7's on a wafer is how well they performed during final test. The ones that can run at max speed get marked as such and sold at the highest price, the slower ones sell for less, yet they all came from the same silicon wafer. Minute variations in the thickness of the various metal and dielectric layers cause changes in the capacitance of the internal circuitry. It is also affected by any contaminants in the air, chemicals, evaporated metals, and other materials used in the process. Remembering RC time constants for Electronics 101, a rise in capacitance increases the time for a signal to rise or fall. At the voltage and speeds inside a modern CPU, a slight change can alter things dramatically.

The temp sensors are also just part of the circuitry that is laid down layer by layer during manufacturing. So it isn't much of a stretch for some of them to be out of spec at some temps, but if it passes the test parameters at a certain test temp, then out the door the entire CPU goes.

Lots of other things are done to maximize product yield from each wafer. For example, some AMD quad core chips have one core fail some test or another and get that core disabled. This way AMD can sell it as a tri-core chip instead of throwing it away. Intel probably has some stuff like this that they do as well. I recall that back in the early 486 days Intel was having difficulty making 486's with working floating point math circuitry. They figured out how to totally disable that circuity and sell the CPU as a 486SX. Different label, same die inside. It just didn't have some of the functions available in the 486DX.

OK, I've rambled on enough... The point I was trying to make is that they don't make the cores and then stick the ones that work into a package together. It's one piece, totally interconnected on the die. Here's a link to a pic I found online of an i7 die: »www.tweaktown.com/cms/gallery.ph···title=#0 Click the top left pic.

Yeah, there's no way that they could make something like the i7 work very well with separate dies for each core. Core 2 Quads consisted of 2 dies with 2 cores each, each set of cores sharing 4MB (or 6MB) of L2 cache for a total of 8MB (or 12MB) of L2. If core #3 needed data from the cache located near core #0, it would have to do so across the FSB. The FSB also got data from RAM, HDD, pretty much everything except the L1 and L2 cache.

On the new i-series, the memory controller is now on the same die as the cores and L1/L2 cache (plus added L3 cache.) Each core has their own dedicated L1/L2 cache, and the L3 cache is located on-die where there is no need for FSB. With the memory controller being integrated on-die as well, again, no need for FSB. Integrated DMI, means no FSB. With the Sandy Bridge CPUs now even the PCIe lanes are integrated on-die (previously handled by QPI or FSB.)

In other words, inter-process communication would suffer greatly if they made cores individually, then later combined them in a package.

reply to koitsu
I think signmeuptoo has got it.

I spoke to a friend, (an electrical engineer NOT CPU engineer), and he explained it to me this way, (sounded good to me). Hope this satisfies the *engineer* in ya.

This is assuming the temp sensors are accurate, or *close*.

"If you are familiar with air-cooled four cylinder motorcycle engines, then you are aware the inner two cylinders will run hotter than the outer two. Doesn't take an engineer, of any type to understand this. Depending on the layout of the CPU die, and how the cores are etched in the silicone chip, one or more will produce heat at different rates, and if a core is surrounded on all sides with another heat producing core, (or any heat producing item) it will register warmer than the other cores."
--
Come fold for a cure with us @ Team Helix.

reply to HarryH3

reply to koitsu

»imageshack.us/f/706/36070963.jpg/

oh and maybe... :

»www.overclock.net/t/305443/ihs-r···he-facts

quote:

---

(1.3) So wait, If the IHS spreads the heat and provides a larger surface area for the heatsink/waterblock which is a good thing, why is this thread about IHS removals?

Now we get to the good stuff, the IHS is not actually built into the die, but is soldered on at the factory by Intel on most CPUs. Solder is pretty much melted metal at a high temp, it is a heat conductive glue that won't melt unless the temp is extremely high, higher than the CPU is capable of outputting. CPUs with soldered IHS's generally have great temps out of the box as hardened solder is extremely strong and is the best "filler" you can use for such a situation, so strong that you have a high chance to damage the die upon removal of IHS. On the other hand there are many CPUs out there that do not have their IHS soldered to the die, instead using TIM between the die and IHS. TIM stands for Thermal Interface Material and acts as a filler when two objects that appear flat (the die and IHS in this case) are to make contact. The thing is the contact between the die and IHS is so close, but not perfect, there is tiny air pockets trapped between the two, and if these spaces aren't filled the heat will transfer to the IHS very slowly as air is a poor conductor.

---