q9550 just one core +10C. bad temp sensor?

[MadMan007]

Heyas. A little behind the 1337 (or 1366 ) Core i7 people but oh well I recently got a Q9550 set up and it seems fine except one weird thing, just one of the cores is ~ +9-10C versus the others. It doesn't appear to be a stuck sensor since it moves to the same relative temp under load. Temps look something like this (default settings):

Core #0: x C
Core #1: x +/- 1C
Core #2: x ~ +10C (the questionable core)
Core #3: x ~ +2-3C
x= ~29C at idle
(after some trials at load with P95 core #1 is actually a bit cooler than core #0 by ~2-3C and core #0 is the same as core #3 like so:
Core #0: x ~ +3C
Core#1: x C
Core #2: x ~ +10C
Core #3: x ~ +3C
x= ~48C at load)


Now I figure that core 0 + 1 and core 2 + 3 should be similar to each other since those are the two dies. However I would expect a bad TIM application or spread to result in one die being hotter than the other entirely, not just one *core* and not by so much.

So the qustion is are there just 'bad sensors' (not stuck) where one temp is just wrong? I'll check the TIM some time soon after letting the TIM cycle a bit but it's a PITA that I don't like to do Also if it's still the same after redoing the TIM I'd end up confused or worried. Thanks!

 

[Randyman...]

Ditto - My Q9550's Core1 reads a good 8*-10*C hotter than the other 3 cores. HSF and TIM are solid (Lapped TRUE w/Washer Mod + AS5). This isn't terribly worrisome, but I am curious as to why this happens...

 

[MadMan007]

Well at least one other person has seen it and says he has good TIM application so that's something. One other thing I noticed is that with IBT (Intel Burn Test frontend for Linpack I believe) the difference is just 5C over the lowest core versus 9-10C for P95.

Anyhow is this normal or have more people seen it before? If so which temperature should I go by?

 

[Grnfinger]

Core 1 on my Q9650 reads 10c higher, under load they all read about teh same.
Honestly idle temps mean very little compared to load temps

 

[Randyman...]

My temps have simular offsets under load, too...

 

[MadMan007]

Core 1 on my Q9650 reads 10c higher, under load they all read about teh same.
Honestly idle temps mean very little compared to load temps

Yeah that would be a stuck sensor or one that reads poorly at low temps, both are possible. I've had stuck sensors before and they do exactly what you said - become roughly equal under load once it crosses the 'stuck' value. In this case my one high core is staying higher under load rather than equalling out.

My main question is whether I should just go by the other cores for temps because even if I just go by delta to tjunction that bad core is closer to tjunction although they're all pretty far off. I usually like to keep temps below 65C for P95 and unless I ignore that core I believe any decent oc'ing will get me above that.

 

[Randyman...]

I let the CPU decide for me. If 3 cores are in agreement, and only 1 core is the oddball, then the 3 cores win . I'm seeing approx 67*C on Core#1 under load, but the other 3 cores don't even touch 60*C. Its not like 67*C would kill a 45nm anyway, but I wouldn't want to run all 4 cores that hot. Idle temps are insanely low...

I'm not terribly worried about it TBH...

 

[Grnfinger]

I agree with randyman, take the 3 cores as correct and ignore the odd ball core.

 

[Daddyjaxx]

My core 1 is always the hottest, 2 and three are usually the same, and 4 is the coolest.

Something like this:

Core 1: 69 c
Core 2: 67 c
Core 3: 67 c
Core 4: 62 c

 

[ibitato]

DTS sensor readings on 45nm chips, below 50c are completetly faulty and pointless.

DTS sensors on 45nm chips only get accurate as the get closer to 100c......

Core 0 usually runs hotter.

On 45nm chips you want to keep the distance to tjMAX ( in the 9550 case, 100c) +/- 20c/25c

Since temp readings are not accurate, keeping a 20c distance to tjMAX it's a good way to keep your chip safe.......

stuck DTS sensors in 45nm chips are a common problem , recognised by Intel.

 

[burebista]

Nice sum up ibitato.

 

[Boulard83]

DTS sensor readings on 45nm chips, below 50c are completetly faulty and pointless.

DTS sensors on 45nm chips only get accurate as the get closer to 100c......

Core 0 usually runs hotter.

On 45nm chips you want to keep the distance to tjMAX ( in the 9550 case, 100c) +/- 20c/25c

Since temp readings are not accurate, keeping a 20c distance to tjMAX it's a good way to keep your chip safe.......

stuck DTS sensors in 45nm chips are a common problem , recognised by Intel.

My Q9550 have the Core3 stuck @ 37c and raise if its higher
My E5200 2 sensor never go under 36c but raise if its higher

 

[ravaneli]

I have a similar problem, but for me it changes all the time. Sometimes Core 0, sometimes core 3 is 10C higher than the rest. Never cared about it.

But when I try to think about it, i reach a conclusion that disagrees with yours. I don't think these are faulty censors, and even less I believe that has been recognized by Intel. If all the censors are the same, why is it always core 0 and core 3 that are stuck? THat doesn't make sense. I think that it's more likely the censors read right, and the problem is with the CPU architecture and windows design. May be the load is not even. COre 0 znd 3 anre the primary cores, i mean if there is a single thread, they are the ones that will run it. That's why when you hit full load for all cores the temps tend to even out. But when u idle the cores are not loaded evenly.

This is just my opinion though. I have never read a respectable source to confirm that, so I am not ready to fight for it : )

 

[Boulard83]

Whit a G15 keaboard you can see your load and temps all the time. My core are randomly loading to 15-20% when surfing and doing normal stuff. Core 3 isnt more loaded but is ALWAIS @ 37 or more. From everything ive read, its faulty sensor ...

Even play TF2 that load my core 0 to 100% but all the others to 15-20% is proving this faulty sensor theorie.

 

[unclewebb]

I have a similar problem, but for me it changes all the time. Sometimes Core 0, sometimes core 3 is 10C higher than the rest.

RealTemp is the only program I know of that reports the temperature from the correct core based on APIC ID information. The majority of monitoring programs including Core Temp and Everest don't. If APIC ID in the RealTemp Settings window is not in sequential order, most software will report the temperature from the 4 cores but after core 0, the temperature data won't be reported from the correct core. Even Task Manager ignores APIC ID information so keep that in mind when testing for this.

If one day at idle, core 1 is the hottest core and then a week later you boot up and core 2 is being reported as the hottest core then it is likely a problem with your monitoring software. On a Core i7 when hyper threading is enabled, Everest reports temperature data from 4 of the 8 threads. If it gets lucky and picks one thread from each core, great, but if not then it's anyone's guess what this data represents. There are times when the first 4 threads only represents the first 2 cores so the last 2 cores are ignored and not reported at all. When temperature testing, use Prime 95 Small FFTs which does an excellent job of equally loading each core. LinX / Linpack create more heat but the load per core is not as consistent as P95.

The Core i7 sensors are excellent compared to the 45nm Core 2 Dual & Quad sensors which have a variety of issues. All Core 2 CPUs can get stuck if you reduce the temperature low enough.

The next problem is that TJMax is not consistent across all 4 cores on a Core 2 Quad. Intel hinted at this at IDF but were vague about the details. There seems to be quite a few Q6600 CPUs where core 0 and core 1 are set to 100C and core 2 and core 3 are set to 105C. The spread in 45nm Quads seems to be around 10C from core to core. My opinion is that Intel does this deliberately to better control thermal throttling. By offsetting TJMax, all 4 cores won't reach the thermal throttling point at the exact same time. Core 0 tends to throttle first and if that can't control the core temperature then core 1 then core 2 and finally core 3 will also get throttled back. This gives a more gradual drop in performance so most users will never notice a problem.

I'm pretty sure that Intel has done the same thing on their Core i7 chips. Core 0 is set to 100C and core 3 looks more like 105C with the two center cores somewhere in the middle. Look at enough RealTemp screen shots while running P95 Small FFTs and that pattern will be very familiar like Daddyjaxx's results above.

The last problem that the Core 2 sensors have is slope error. That means the sensor will move at a different rate than the temperature changes at. Intel mentioned a ball park figure of +/- 10% for this error. When you combine this error with errors in TJMax and sticking sensors; it's almost like an act of God when you have two sensors that track each other closely.

The load bouncing around from core to core at idle does not seem to contribute much to the differences seen in core temperatures. It's mostly sensor error or undocumented "features" that Intel doesn't like to talk about. At full load where the amount of error decreases, these sensors are OK but there is too much error and too many unknowns for users to make accurate comparisons between their core temperature data and their friend's data. You can't even compare one core to the next on the same CPU because of this.

 

[MadMan007]

Well that's some interesting techogeek information.

Anyhow I'm not bothered by it since a few others report it. I'm pretty far away from TjMax (actually there's one about 5 miles from me! lol ) so I knew I was safe and I'll just ignore that crazy core as suggested while I'm so far away from TjMax.

So anyhow the other day I decide to oc a bit. I got 8.5x400=3.4GHz at pure default settings. 3.6GHz took a bit of work, I had to tweak CPU Vtt and NB voltage but no CPU vcore increase it was a tiny bit frustrating getting it dialed in because I had to increase the NB voltage from 1.18 set/1.15 real in BIOS to 1.245 set/1.21 real in BIOS for such a small FSB increase. At load the CPU is ~1.18V (this is what CPU-Z and BOIS says too) Mobo is a DFI P45 Jr.

Real VID at load is 1.2375, batch #L847C185 since I knew people would ask I'd never heard of 'C' batch before but googling gets a few results and some people do have nice CPUs from that batch although it's quite variable.

Here's a pic after running IBT, you can see the max temps in Realtemp. It's in an Antec NSK2480 case with a direct CPU vent cut out in the top, when I did that a few months ago my CPU temp dropped by some 10C with my E8400 so it cetainly helps here. The cooling is a compact Coolermaster Geminii S with a low speed fan so these aren't TRUE+fast fan-like temps but there isn't much room for a big cooler in the case and it's nice and quiet.
(Idle pic so Speedstep has decreased the multi but you can see the FSB)

 

[ibitato]

I have a similar problem, but for me it changes all the time. Sometimes Core 0, sometimes core 3 is 10C higher than the rest. Never cared about it.

But when I try to think about it, i reach a conclusion that disagrees with yours. I don't think these are faulty censors, and even less I believe that has been recognized by Intel. If all the censors are the same, why is it always core 0 and core 3 that are stuck? THat doesn't make sense. I think that it's more likely the censors read right, and the problem is with the CPU architecture and windows design. May be the load is not even. COre 0 znd 3 anre the primary cores, i mean if there is a single thread, they are the ones that will run it. That's why when you hit full load for all cores the temps tend to even out. But when u idle the cores are not loaded evenly.

This is just my opinion though. I have never read a respectable source to confirm that, so I am not ready to fight for it : )

My post is not my opinion, it was a short list of well known technical facts.

Here:
http://www.tomshardware.co.uk/intel-dts-specs,news-29460.html

This is a short reading but very enlightening.



It has nothing to do with windows nor the cpu architecture. It's all related to the DTS sensors and how they work.

You might have a point with the cores load, but it's not windows / architecture related. It has to do with the software you run and the type of programming used.

If the software you run (like many games) it's not optimized for multicore it will run in only 1 thread; thus using only one core.

For example, Crysis was programmed (Crysis engine, to be more precise) having in mind Core 2 Duo (2 cores, 2 real threads); so running the game on a Quad will actually use more 2 cores than the other , thus, heating more the 2 cores beign used.

On the contrary, games like GTA IV, that are really complex and CPU dependant; are programmed / optimized to make full use of all 4 cores. So the CPU gets s better use and a more even load.

Windows cannot effectively "balance" process on cores; all it does it's to assign a process to a execution space (core) and will not assing other core if the core 0 it's not almost fully loaded.

Software must be programmed in orther to use multi-core.

BTW, there is a concept called CPU "affinity". Although this is a programming concept while developing muti-threaded apps ; it can be used to manipulate how the OS assigns processes to cores / threads

http://www.techenclave.com/guides-and-tutorials/cpu-affinity-made-easy-67317.html