FX8120 and his temp

[RGone]

Speedfan?!?

EDIT: I apologize. I just learned it has been designed by a NASA thermodynamics engineer!

That makes him privvy to AMD NDA information and his code above reproach?

 

[Bubba-Hotepp]

That makes him privvy to AMD NDA information and his code above reproach?

He was being sarcastic and making a reference to the fact that I emailed a scientist from nasa for the thermodynamics question that we moved to another thread (it was the first name that came up in a search on google). He was being smug and it backfired on him.

 

[caddi daddi]

cloudheaven,
at 30c/85f your going to need to be sure to turn ON cool and quiet and perhaps turn off 2 cores.

My airflow issue is actually due to using the box for watercooling instead of air cooling, i pulled out my loop and installed the oem cooler and the cooling is quite good and with the $29.00 hyper212 it's just great. I am suffering a fan issue with the rad inside the box that i did not have when it was attached outside of the rear of the old box and I have ordered stuff and things to cure that and will remove the top fans when they arrive.

 

[RGone]

"C_D", what case are you using? Thought you had a jam-up one, like HAF912 or larger?

 

[caddi daddi]

I am stuffing the pig into a haf x 942.
the top 200mm fan plus the one i added were restricting the airflow out of the top grills.
all the lights on this box are red, a color i don't see so, i have to look at the front mounted pump/res with the blue led to know if it's on.

 

[RGone]

Does that case have a cut-out behind the cpu area. Some have drawn air to the outside of the case from there and temps were much more manageable then. Note some of this is just reporting for brainstorming.

I am stuffing the pig into a haf x 942.
the top 200mm fan plus the one i added were restricting the airflow out of the top grills.
all the lights on this box are red, a color i don't see so, i have to look at the front mounted pump/res with the blue led to know if it's on.

 

[caddi daddi]

I did install fans on the backside of the socket on my other boxes and this drove my socket temp readings way down. this worked as my core temps fell 1-3c so it worked but, the socket temp sensor i think is exposed back there and that is why i saw a huge reduction in socket temp and a very small drop in core temp.
this case has no fan port there but i will get it to the waterjet and it will have a 40mm fan port there.

 

[The Coolest]

Ok. I managed to get my hands on a 8150 and I'm running tests right now. So I thought to give a quick update.
Very interesting result already. When running Prime95 on all 8 threads the TCore and TCase temps are 18c apart. But when I set affinity to the last thread, TCase goes down while TCore shoots up and is actually higher than TCase.
If I set the affinity to the first thread, temperatures on TCore goes much lower than when the last thread was selected.
I'm gonna run this test over the other cores and see what the results are.
I have a feeling that this TCore sensor is located somewhere close to the last core/module, thus giving skewed temperature readings.
I'll let the pics do the talking:

Full load:

Last core:

First core:

 

[trents]

So what is the default core affinity with CoreTemp? Usually CoreTemp gives the same reading as HWMonitor in my experience and I don't think you can set the affinity manually in HWMonitor so I assume at default CoreTemp is set to the same affinity as HWMonitor.

But really, I guess I don't understand this affinity thing. Apparently CoreTemp is taking information from something other than the core temp sensor since there is only one of those on a current generation AMD CPU.

 

[The Coolest]

Here are the final readings:
aff 7 = tcase 57/tcore 61
aff 6 = tcase 56/tcore 61
aff 5 = tcase 55/tcore 58
aff 4 = tcase 55/tcore 58
aff 3 = tcase 57/tcore 52
aff 2 = tcase 57/tcore 52
aff 1 = tcase 55/tcore 49
aff 0 = tcase 54/tcore 49
All = tcase 71/tcore 52

@ trents:
I changed Prime95's affinity to force it to run on a certain core rather than have it run on all of them.
Core Temp manages affinity internally, and you shouldn't play with it, as it might give you wrong readings.
Now, there is only 1 sensor in the chip, HWMonitor lists all of the cores to prevent confusion in users. And the only reason that you sometimes catch different readings between cores is due to the delay at which each of the readings is taken.
In the case of AMD processors, affinity does not matter for temperature readings, as it is located in a mapped memory space and not in the MSRs.

 

[trents]

Oh, I see. It was in Prime95 you changed the affinity, not in CoreTemp. Yes, I see that now as I look more carefully at the picture. Sorry.

But it would seem this only has a bearing in our interpretation of temps when cores are either locked or unlocked. Cores cannot be unlocked with the FX CPUs and I'm not sure there are any to unlock even in the 4 and 6 core FXs so the sensor will always be in the same place with regard to the core positions on a given FX line. I suppose that may not be true if you're moving between the 4 core, 6 core and 8 core lines. What do you see as the significance of this discovery?

 

[Bubba-Hotepp]

Oh, I see. It was in Prime95 you changed the affinity, not in CoreTemp. Yes, I see that now as I look more carefully at the picture. Sorry.

But it would seem this only has a bearing in our interpretation of temps when cores are either locked or unlocked. Cores cannot be unlocked with the FX CPUs and I'm not sure there are any to unlock even in the 4 and 6 core FXs so the sensor will always be in the same place with regard to the core positions on a given FX line. I suppose that may not be true if you're moving between the 4 core, 6 core and 8 core lines. What do you see as the significance of this discovery?

Coolest can correct me if I'm wrong but what he's basically done is confirm what the Tech Rep explained (at least that's why I concluded from his post). AMD uses a formula to derive the Tcore temp using the the Tcase thermistors. The tech rep explained that it takes readings from the "nearest" thermistors to a particular core and then uses the formula to "guesstimate" the cores temp. With that it in mind it makes sense that those readings would vary that much as it's all dependent on how close the thermistor is to that particular core whereas the "Tcase" reading output is an average of all of the thermistors in the die package.

Edit - By using "affinity" with prime95 he's forcing it to run on each particular core (it's my understanding that even when running "1" worker thread it doesn't neccessarily stay on "1" core) and then taking a reading for that core.

 

[The Coolest]

That's exactly right.
What this means is that basically both TCase and TCore don't really reflect the actual temperature of the CPU.
In the case of Intel's TCore, you have a sensor in every core, thus you can see the temperature of each core independently and be sure that your CPU is running within the range of temperature that you want it to run in.
In the case of AMD it seems that it's really impossible and we can only depend on some average values calculated using some formula.
I'm really baffled as to why AMD decided to go this route... They had proper sensors in each core in the K8 days. They sometimes weren't calibrated really well, but at least we had a reading per core.
Intel does it, VIA even does it (which Core Temp supports for TCore as well ) but AMD doesn't, and it's just confusing.

I have a P-II X6 1090T here as well. I'll do some testing on it as well, just out of interest.
It'll also be interesting to see if other FX users can replicate my results, and also what the results would be on FX-4000 and FX-6000 series of chips.

 

[Bubba-Hotepp]

That's exactly right.
What this means is that basically both TCase and TCore don't really reflect the actual temperature of the CPU.
In the case of Intel's TCore, you have a sensor in every core, thus you can see the temperature of each core independently and be sure that your CPU is running within the range of temperature that you want it to run in.
In the case of AMD it seems that it's really impossible and we can only depend on some average values calculated using some formula.
I'm really baffled as to why AMD decided to go this route... They had proper sensors in each core in the K8 days. They sometimes weren't calibrated really well, but at least we had a reading per core.
Intel does it, VIA even does it (which Core Temp supports for TCore as well ) but AMD doesn't, and it's just confusing.

I have a P-II X6 1090T here as well. I'll do some testing on it as well, just out of interest.
It'll also be interesting to see if other FX users can replicate my results, and also what the results would be on FX-4000 and FX-6000 series of chips.

This is just purely a guess, but maybe it's because just showing "cores" can be misleading when you have a NB section on the chip with a seperate voltage that can affect temps too? So by covering the die with a series of thermistors and then using the average you basically "cover" the whole chip? What are your thoughts coolest?

 

[The Coolest]

My point was that in the current situation we have no way of knowing what is the hottest spot on the CPU, like we can on Intel and VIA chips.
With AMD we can only get an average temperature, which doesn't necessarily reflect potentially critical temperature being reached by the processor.
I guess that if AMD engineers thought that this is a good way to go, there is probably a good reason for it.
I can only assume that Intel follows a similar path when it has to report 'external' temperature, like TCase, which puts us in the same spot, but for us enthusiasts, we still have TCore temps to rely on if we so choose.

 

[Bubba-Hotepp]

My point was that in the current situation we have no way of knowing what is the hottest spot on the CPU, like we can on Intel and VIA chips.
With AMD we can only get an average temperature, which doesn't necessarily reflect potentially critical temperature being reached by the processor.
I guess that if AMD engineers thought that this is a good way to go, there is probably a good reason for it.
I can only assume that Intel follows a similar path when it has to report 'external' temperature, like TCase, which puts us in the same spot, but for us enthusiasts, we still have TCore temps to rely on if we so choose.

Probably why they made the "maximum operating temperature (Tcase Max)" a "conservative" limit. IMHO it doesn't really matter which temps you decide to use, just as long as everyone understands that when they see the "maximum" temp listed on the various sites and product descriptions by AMD that it's not referring to "core temps" otherwise they'll run the risk of running it much higher than the stated max temp.

 

[The Coolest]

Here are the readings with an X6 1090T:
aff 5 = tcase 57/tcore 47
aff 4 = tcase 57/tcore 53
aff 3 = tcase 57/tcore 43
aff 2 = tcase 56/tcore 44
aff 1 = tcase 57/tcore 41
aff 0 = tcase 56/tcore 40
All = tcase 68/tcore 59

 

[trents]

Probably why they made the "maximum operating temperature (Tcase Max)" a "conservative" limit. IMHO it doesn't really matter which temps you decide to use, just as long as everyone understands that when they see the "maximum" temp listed on the various sites and product descriptions by AMD that it's not referring to "core temps" otherwise they'll run the risk of running it much higher than the stated max temp.

Yeah, that makes sense to me and it's actually what we've been recommending around here for some time without knowing all the technical reasons why, speaking for myself. This has been a very rigorous discussion and though I wasn't able to follow it at all points I think it fair to say it has been very helpful to me to bring some clarity to these mysteries.

 

[Bubba-Hotepp]

Yeah, that makes sense to me and it's actually what we've been recommending around here for some time without knowing all the technical reasons why, speaking for myself. This has been a very rigorous discussion and though I wasn't able to follow it at all points I think it fair to say it has been very helpful to me to bring some clarity to these mysteries.

I have this OCD need of wanting to always know why things work the way they do

So one thing I'll point out that I've learned. Others have pointed to a "letter" that someone keeps posting in OCN from AMD where he talks about "core" (singular) temp and use that as a rebuttal as to why the "core temps" (plural). Nothing in the "letter" contradicts what the tech rep has taught us. In fact this same letter goes on to talk about how the individual core temps are derived by formula. So to summarize I think it's safe to say that the following is true -

TJunction = Motherboard Temp Sensor underneath the socket.
TCtl = The "sliding scale" algorithm that controls the built in overtemp failsafe (i.e. throttling) functions of the CPU.
TCore = The individual "Core" temperatures derived by formula from the sensors (thermistors) on the CPU package.
TCase = The reading taken by averaging the values from the sensors (thermistors) on the CPU package.

The only question left, which at this point I'm only "guessing" at, is which reading equates to which label assigned by the BIOS/UEFI and shown in the various programs like HWMonitor. At this point I'm going to assume (err on the side of caution) that the "CPU Temp", i.e. CPUTIN or TMPIN2 whatever the MB in question outputs is the "TCase" reading and that the individual cores are "TCore". I'll keep digging by contacting MB companies to get clarification on that as well.