Q9400 Quad core processor temperatures

[idata]

I recently installed a Q9400 Core 2 Quad processor in my gaming computer. I installed an aftermarket CPU cooler after finding out the heatsink and fan included with the retail packaged processor did not cool my CPU adequately. I used the thermal paste included with the aftermarket cooler.

The question I have is, before and after installing the aftermarket CPU cooler, the temperatures on each of my cores are different. They varry from 91 degrees F to 105 degrees F (33C to 40C) at idle (computer on, just browsing internet). My question is whether or not this is normal. I understand higher loads on different cores creates higher tempratures, but I figured being on one piece of silicon and so close together that the temperatures would be in a much tighter range, not the 15 degree spread that I am seeing. Is this a sign that I maybe didn't get the paste spread out properly?

Core 1 and 3 are always higher in temperature than the other two. Core 2 temp is higher than Core 4 temp. This trend continues even when playing high-end games.

I am getting these temperatures from an Nvidia program included with the motherboard and graphics card, both with Nvidia chipsets.

I have a somewhat high hardware knowledge, but my software knowledge is somewhat lacking. Thanks for your help.

[idata]

Hi,

It could mean that you didn't get the paste right. It could also mean that the software is mireading the temps. At the same time be aware that intel dosen't disclose how to read accurately teh temperatures of their cores...

What does youBIOS say regarding the CPU~ temp as a whole? as long as that it is a good temp, i wouldn't worry too much.

[idata]

My Q8200S with minimal cooling is around 40C, so I'd say you in the right area.

The Intel Core2 Quads are 2 dual cores on the same package, so it's common to see a difference in temps between the 2 sets of cores. I see a 5C spread at idle.

I think CoreTemp give good values, as they too the Intel relaesed TjMax's and used those to calc the temp's

[idata]

"I think CoreTemp give good values, as they too the Intel relaesed TjMax's and used those to calc the temp's"

True, but that does not give you accurate core temperatures... just the Tjunction.

[idata]

I am kinda worried about my Q8400 which seems to be having the same problem... my temps (CoreTemp) are:

Core # 1: 29C

Core # 2: 23C

Core # 3: 38C

Core # 4: 42C

So I'm having a whopping 19C temperature difference between cores 2 and 4, which makes me think something is really goofy here...

I'm using an all copper Zalmann CNPS9500 AT heatsink and Arctic Silver 5 thermal paste, and I'm a system building veteran so I know I didn't apply the paste wrong... I don't know what to make of it. Aren't the cores supposed to be within just a few degrees C of each other? Is something wrong with my chip?

[idata]

hello,

yeah i also got the same issue..anyoen got any idea what is that

Hardware monitor Intel Core 2 Quad Q9300

Temperature 0 50°C (122°F) [0x32] (Core # 0) Temperature 1 37°C (98°F) [0x3F] (Core # 1) Temperature 2 58°C (136°F) [0x2A] (Core # 2) Temperature 3 45°C (113°F) [0x37] (Core # 3)

really weaird.. .like 20 degrees diff

any ideas?

[RGiff]

This Temp difference is normal , it depends on which core is being used most at that time , I have seen this on every CPU from P4s to Xeons . Unless you have one core stay 50c more that the other 3 , its normal.

[idata]

I don't think, the temp difference is normal, if the pattern doesn't change regardless of particular core load. And that's what previous guys I believe reported.

I've similar issue with Intel Q9400. CoreTemp readings are measured as:

Core 1 - 36C, Core 2 - 31C, Core 3 - 21C, Core 4 - 32C at idle

Core 1 - 56C, Core 2 - 45C, Core 3 - 32C, Core 4 - 47C at full load (OCCT CPU Test)

Its true that CPU itself tends to use the Core 1 more often that other cores (which I consider CPU controller weak point, as it should try to evenly use all cores). But Temp. pattern remains similar at idle, when no cores are continuosly used. And it remains at even greater disparity at test full load, despite all cores are fully loaded at all times. It points to the fact that either sensitivity of Intel used sensors, or their positioning relative to most heated core areas vary greatly. Or may be contact btw the proc and heatsink becomes uneven due to residual deflection of motherboard PCB under uneven pressure, after the heatsink clips are pressed in. In any case, "average proc Temp" shown by some utilities doesn't seem to make any sense, as the whole processor will probably fail, if one of its cores failed due to constant overheating.

So, this persistant T discrepancy pattern btw different proc cores regardless of load calls for clear explanation from Intel engineers: why exactly this happens? Definitely they see similar pattern in their test Labs, and investigated its root cause. What a customer can possibly do to maintain more even core temp. distribution? For sure its not related to bad paste spreading, since most ppl who keep reporting this issue are experienced PC buiders & overclockers (ordinary users don't use CoreTemp or pay attention to such issues that often).

Hence, I ask Intel Customer Support to provide explanation, whether this fenomena is usually caused by sensor positioning variations in chip assembly process, or bad sensor calibration, or wide sensors sensitivity tolerance, or inaccurate methods of calculating Temp, or Motherboard deflection, or WHY EXACTLY? Be as honest as you possibly can, don't just say "Its OK, don't worry".

Another question is, can anyone point to a statistics of processor failure rate & average service life depending on core temperature variations? How long a proc will work depending on at what T it usually runs?

P.S. (1) I found this datasheet info on http://www.tomshardware.co.uk/intel-dts-specs,news-29460.html Intel digital thermal sensor (DTS) inaccuracy that explains these discrepancies at least partially:

"The digital thermal sensor (DTS) accuracy is in the order of -5°C +10°C around 90°C ; it deteriorates to ±10°C at 50°C. The DTS temperature reading saturates at some temperature below 50°C. Any DTS reading below 50°C should be considered to indicate only a temperature below 50°C and not a specific temperature."

The above document also gives Intel specified TJMax values for different processor models. Processor throttling starts when core Temp is deemed to reach TJMax based on Sensor readings.

(2) I tried to "unflex" my ASUS Motherboard pushing to opposite direction after pressing into it the heatsink clips (standard attachment method). The core temps discrepancy decreased, so now it stays at:

Core 1 - 32C, Core 2 - 31C, Core 3 - 22C, Core 4 - 29C at idle

and also differs less at full load. Core 3 Temp deviation I attrribute to its non-calibrated & inaccurate T sensor. This means that residual Mobo deflection (hard to see when installed in the PC case) caused by heatsink clips being pressed in is another factor contributing to T variations due to uneven preload surface contact btw proc and the heatsink. May try to test supporting the Mobo from back by an eraser (rubber peace) to straighten the PCB.

From these findings, Intel should definitely look at improving Sensors quality and calibration, and also finding less damaging for Mobo ways to attach the proc heatsink. Both issues lead to proc capabilities being underused and Intel customers dissatisfied due to some Core Sensors reporting erroneously reaching http://www.overclock.net/intel-cpus/476469-truth-about-temperatures-voltages.html TJMax at much lower actual Temp resulting in undue throttling of the processor and constant customer worries.

[RGiff]

This temp problem also changes depending on the quality of the CPU , Like the temp readings on a Q9550 will very much more than a Xeon simply because a Xeon is made to closer specs. than a desktop CPU. So this is a good example that you get what you pay for.I've seen guys Lapping there CPUs tring to get even cooling because they are'nt FLAT , so they lap them down tring to get a better contact between the cooler and the CPU. Most Xeons are flat and don't need lapping.

[idata]

Its hard to imaging such large surface inflatness of that small proc cap surface area that can't be compensated by a Thermal Compound layer re-distribution. Of course, it won't be a flat layer after all, but will it affect much thermal conductivity of the compound?

As to "you get what you paid for", given similar production and development costs for all proc models, and huge variations in their retail prices despite that, I don't think a customer gets what he paid for in any case. What's shocking, Intel has no desire to improve sensor quality or perform thermal sensor calibration in their chips before they're shipped out of the factory?! How its possible given high prices of this mass volume product?

[idata]

Same here with a q9300. Nvidia utilities and Core Temp show cores 1-3 varying betweeen 38 and 45 degrees celsius. Core 1 reads at 55 degrees

Load seems evenly distributed across 1-3 @ approx 7% utilization. Core 0 @ 15%.

As an experiment I went an and set the affinity for every task (i could anyway- some are system locked) to cores 1-3. After 10 minutes, processor load even across all 4 cores @10-15%, cores 1-3 @ 45 degrees (+-1degree), core 0 at 56 degrees

I can't say the sensors aren't broken... but they are consistent

I switched to the q9300 a few weeks ago. I really didn't NEED to upgrade from the q8200, but I wanted hardware virtualization, and it gave me an excuse to upgrade my fiancee's machine in the process. I haven't seen a bluescreen yet, but I've had apps closing at a startling rate since the processor switch. If this is a thermal issue, I'd sincerely like to confirm that before bothering the intel support channel

[idata]

The variations between core temperatures in multi-core CPUs is normal and nothing to worry about. I suggest you check any of the many PC hardware review web sites and look at their tests of CPU coolers, where they list the temps of CPU cores and the general CPU temperatures. The figures you'll find will be very similar to your own.

Intel does not build "temperature sensors" into their CPUs for use with programs that display those temperatures, but as safety devices that will hopefully stop the CPU from being destroyed by overheating, whatever the cause may be. Intel does not say that "CPU Temperature Monitoring" is a feature of their products.

I could describe in detail the physical reasons why the variations in core temps that are seen occur, but simply put, very small objects, like CPU cores, can change in temperature very quickly up or down. Their tiny mass will heat up quickly with the application of very little energy, but also will not retain that energy due to their small mass. We aren't considering bricks or a kilogram of metal here.

There is a simple method of testing if one or more cores of your CPU are running unusually hotter than the others. Two things are needed, a temperature monitoring program for the CPU and cores, and a CPU stress testing program.While your CPU temperature monitoring program is running, start the CPU stress testing program and load all the cores up to 100%. Let that run for a few minutes at least, and note the temperature of each core. Don't forget the margin of error in the temperature readings for the cores as stated by Intel and posted above in this thread. With all cores running at 100%, any "bad" or overheating core will stand out, since the loading on all of them is equal and constant. If you don't use a high performance cooler, you should not run the stress test very long. Even if you do and the CPU terribly overheats, the CPU will throttle itself down, which should be indicated by a reduction in one or more core's temperature.When I run this test, the range of low to high individual core temperatures in a quad core i-7 processor is 4 degrees C. At idle, I normally see core temperature differences of 5 to 7 degrees C, and at times up to 10 degrees C. I'm not at all worried about that.

[idata]

This is not about small mass heating fast, but about terrible sensor inaccuracy and lack of calibration at production despite high product price. The sensors tolerance +/- 10deg is huge, given what many folks are trying to achieve by using expensive cooling devices - just make it a few degrees cooler. The sensors reading remain consistent over time, meaning they can be calibrated at production, but Intel doesn't do it. So, if a proc Tjmax is 62C, and the tolerance at this T is +/- 8, then your proc will start throttling 16C earlier than it should - at 46C leading to slow performance without any reason. It also quite worrysome for most users to see high proc temps. On top, some sensors are "stack" (show very small or no T variations despite large load changes). "Visual" proc performance monitoring is very important for many users. The point is, Intel needs to imrove sensors quality, and calibrate them before shipping. If a sensor is "stack", the proc can't be shipped to customers, since its reliability & service life will suffer, while price remains the same. It's knowinly shipping a defective product hopping it might not manifest itself resulting in product failure within a warranty period.

[idata]

Well let's see, the person that started this thread, dieselmech86, asked about the varying temperatures of the cores in his CPU. My reply much later in this thread, in part was an answer to that question, which included considerations of the temperature variations in very small objects. The six posts following the first one were about differing core temperatures, their range of temperatures, and whether or not that was normal. Seven post's into this thread, you posted your semi-on-topic diatribe regarding the inaccuracy of the thermal sensors in Intel CPUs. I was not responding to your post, but to the first post in this thread, as well as a few of the others that were wondering about the same thing. But I'll be happy to discuss your topic too.

I do agree, in theory, that poor or inaccurate calibration of temperature sensors, which then could read significantly higher than the actual temperature, would cause the CPU to throttle down before it is necessary. The result would be as you have described, a loss of CPU performance for no real reason or benefit. Also, if the opposite occurred, where the temperature sensor read significantly below the actual temperature and the CPU did not throttle down because of that, the CPU could be destroyed due to overheating.Assuming what I have written above and what you have written is valid and correct, the real question is whether or not either of these affects occurs in practice, in the real world. Let me say first that I certainly know that the reality of why, when, and how a CPU reacts to an overheating situation is far more complex than anything you or I have described. I have read some of the Intel Core i7-900 Extreme Edition Series and Intel Core i7-900 Desktop Processor Series Datasheet, Volumes 1 & 2, which describes in detail the many different and interacting things that occur before CPU throttling occurs, as well as the accuracy range of the thermal sensors, etc. Considering how complex those descriptions are, I'll bet there is still more to it than that. Regardless, we can still try to determine if we see the negative affects you first wrote about.

I enjoy reading the reviews on PC components found in PC hardware web sites, particularly CPU coolers, my favorite PC gadget. In all the reviews I have read, where the CPUs are over-clocked, over-volted, and worked overtime to see how well a cooler cools, I've never seen anything mentioned about CPU-throttling or even CPU shutdown due to high temperatures. I'v never seen it stated that any of them destroyed a CPU during the testing. Given how much some reviewer love to hate the stock Intel CPU Cooler, I'd think they would jump at the chance to say the Intel cooler caused CPU-throttling or shutdown, but I've never seen that either. I get the feeling that the tolerance of the thermal sensors Intel states is very over-stated, meaning it's really much less.

Regarding calibrating the sensors, it's very hard to comment on that since I don't know the implementation. An on-die thermal diode would be so small and likely impossible to adjust. I have experienced what seems to be a stuck thermal sensor on a CPU (by the way, you mean stuck, not stack) according to one monitoring program, but an Intel program displayed the range remaining to maximum core temperature fine, so who knows.

So, all I'm saying is I don't see much if any evidence that CPUs are throttling down early or burning out. As I've said before, Intel has never said the thermal sensors are meant to be used as real-time CPU thermometers, although many programs use them as such, and who knows how accurate they really are?

Finally, consider this: the temperature readings you are concerned about, the individual core temperatures whose range seems to be to wide, are based on the same loose tolerance thermal sensors. Frankly, unless you or I are PhD electrical engineers specializing in CPU architecture and design, most of this discussion is futile.

[idata]

From a consumer prospective this discussion is revealing, clarifying the subject matter in simple terms (not just sharing personal observations as most such discussions do), and thus seems to be quite useful. Including those Intel Marketing guys who are expected to communicate with customers and translate their ever changing needs and expectations to particular product improvement plans.

As to your referal to tests performed by some known 3-d party sources, one seldom see Temps going close to Tjmax for tested processor models in these tests, since there purpose is to reveal advanced design cooling systems efficiency, and they are performed by well traned individuals. But in a long run real life environment (not a short test), regularly stressing the proc with high temps leads to shorter service life and performance drops, including by consumers trying to avoid full proc loading being concerned and misled by inaccurate sensor readings.

It would be interesting to find out, what variables in a proc design & production process lead to such variations in T sensor readings, and how they can be mitigated to better meet customer expectations, including by sensor calibration before product shipping. Hope Intel would be willing to disclose some reliability test data on modern proc service life expectation depending on average Temp. its used at. Or, may be you can point to a link where such data is published?