"Core Temp" - monitor temps on K8s and Core/2 CPUs (updated

[soundx98]

Screenshots of Core Temp and other programs are always easy to look at with a glance.

Thanks for posting

[zarathrustra]

Screenshots of Core Temp and other programs are always easy to look at with a glance.

Thanks for posting

Good point:)

 

Here are some shots of what I posted above. Even though Core Temp shows 'Toledo', its due to my 3700+ being one of those CCBWE (same as Flower King's 170 Opty) chips with a disabled core.

 

It seems acurate to me considiring the internal diode is right in the core as apposed to under the socket, and therefore reading a better positioned, realtime temp.

[soundx98]

Some nice cooling going on there my man.

 

Prime 95 didn't really seen to get my CPU as hot as an 8M run of SuperPi/

[zarathrustra]

Some nice cooling going on there my man.

 

Prime 95 didn't really seen to get my CPU as hot as an 8M run of SuperPi/

 

All hail Danger Den, and the GTS360 with the fans barely moving:D

 

I just tryed a run of SuperPI 8M. With the same room temp, Smart Guardian read the same as with Prime95, while Core Temp read 3-4 degrees cooler (29-30) Strange, seems Prime lights my fire:)

[Praz]

I'm beginning to believe that the way temperatures are monitored with CoreTemp and similar programs are capable of much greater accuracy then what we have been using. In the AMD On-Die Thermal Diode Characterization Application Note it is stated that the thermal diode may be up to 4C cooler then the center of the cpu junction. In the Opteron processor datasheet, page 70, the following can be found.

 

Thermal solutions should be not designed and validated using the thermal diode. Thermal solutions should be designed and validated against the case temperature specification per the methodology specified in AMD AthlonTM 64 and AMD Opteron™ Processor Thermal Guide, order# 26633.

 

From this the accuracy of the thermal diode is suspect. And just how inaccurate it can be is listed in footnote #5 on the same page.

 

After correcting for the diode offset, the thermal diode has an accuracy of ±10°C. This accuracy is additive to the temperature sensor accuracy.

 

Based on the above the temperature reported in the bios or by programs such as MBM could be reading low by as much as 14C. And this is assuming that the thermal offset has been properly applied and there is no error in the onboard sensor.

 

AMD uses the thermal sensor instead of the diode to enable Thermtrip to protect the processor from excessive heat. AMD's own documentation demonstrates that the thermal sensor is the more accurate means of temperature monitoring. If the few inconsistencies with the programs that read these registers can be worked out we will all have a much more accurate means of temperature monitoring.

[soundx98]

I'm beginning to believe that the way temperatures are monitored with CoreTemp and similar programs are capable of much greater accuracy then what we have been using

 

One of the reasons I asked TheCoolest to start a thread here.

Important info on Page 71 of the Opteron PDF as well.

You know, you have to wonder if/why/why not AMD is working with sensor manufacturers such as ITE as well as the major motherboard manufacturer's?

In fact, why haven't they released a program or a white paper themselves that makes the "correct" CPU Diode temperature available to manufactuers and users?

CPU Temps and cooling seem to make up about 20% of the questions on the various forums I frequent.

 

Some observations First:

The AMD 64 pdf is from 2001 (Has to be the 754 sockets).

The Opteron from Feb 2004 (Probably written in 2003 and is definately for the 940 Pin ECC 3xHTT Models not our 939s)

I would have to believe there is something, either internal to AMD or published, that MAY give us more current information on the CPUs in our rigs.

Basic changes in die size and manufacturing processes would cause things to change I would think.

In TheCoolest's original posts at XS he stated the program was for "F" processors.

Later he stated that it was a feature that was also built into previous models and he believed the program could be used on them.

I guess I am implying that it may not work the same on the earlier models

 

Vitaliy Jungle (VVJ - NextSensor & Central Brain Identifier) looked at the P-state and temperature offset in HIS CBID program.

Most temps for 939 Opterons (In a thread at XS) were reporting an offset of -2 to -9 degrees. Initial reaction was that the correct temp could be found by simply adding the offset temp to the SG, Speedfan, MBM5 or Bios temp.

It seemed to explain why folks thought the Expert/Venus boards were reading low.

 

Then more users started submitting screenshots of the program and some had offset temps on the high (or positive side).

(So just subtract them from the temps instead of add you would think?)

One of my rigs read a +9 for offset (Ultra-D)

The other rig read a +20 for offset (Venus)

(one member PMed my that his Opty 146 had a +22 degree offset)

So in seeing swings from -9 to +22 on the same Opteron 146 processor (and yes I understand that each CPU has it's own offset and that this offset is programmed in at the factory) it seemed to me that 31 degrees was a pretty darn big difference.

Kinda makes you wonder how accurate this diode is and how accurately it is programmed.

From my understanding it's true purpose is to protect the CPU from frying.

 

I had been in contact with VVJ and was testing some beta version of NextSensor and furnished some information from my rigs as we thought adding Vdimm, LDT, CPU Diode, and verifying the other voltages/temps/fans would be a great addition to his program and he did not have a DFI NF4. (you need to go to control in NextSensor and select DFI LanParty for our boards)

 

VVJ was all set to incorporate "offset" into the NextSensor program. When I told VVJ this doesn't seem to make sense (for whatever reason) in some cases he decided to drop it but did want to continue further investigation.

(And please remember that there is always some rude MF posting on the forum that CBID or Core Temp suxorz as these guys bust their balls for free. Real encouraging)

Offset is NOT incorporated into NextSensor and I believe it is not in Core Temp either.

 

I sent this screenshot to him. This a DFI Venus at Idle with an Opteron 146 with a TR XP-120 on it on a table (no case) in March and I would imagine the room temp was between 67-70 degrees.

 

latest version 8.0 is here

 

There is just no way I can believe that the CPU temp is either 49 or 9 degrees at Idle.

It can't be below ambient temps and I would think I would have had problems with it shutting off when running Prime95. It did a 37 hour prime when I put the board inside a case and I have had the CPU Temp as high as 52 in MBM5 with rig set to shut down at 60 in bios while doing some fan testing.

 

Such a shame we live in such a secretive society.

When Alex Van Kaam stopped development of MBM5 he wrote a "good bye" post.

He basically said that he could not get information from manufacturers. They were like -

OK, here is the info (year old at this point), but before you look at it you must sign this NDA (and our lawyers will bury you if they get a chance) and you cannot ever release the source code for your program. You want info - find out on your own.

Happy_Games I am sure is under an NDA (Non Disclosure Agreement) regarding things he knows. I had a job interview the other day and they required me to sign a NDA before the interview started!

Everyone is "sign and date" happy these days.

That is why it is so important to furnish these guys Screenshots and dumps so they look into it as help from corporations just ain't there.

Maybe it's because we are looking at different sensor models.

 

 

I don't have answers and am only raising questions and sharing what I know, as I sure as hell would like to get a handle on this.

Keep your screenshots and ideas and thoughts flowing in.

[neogenesis2004]

Strange, on my RS482 Infinity ITE Smart Guardian says my cpu temp is 42 C. However, Core Temp shows core 1 at 24 C and core 2 at 36 C. Im on water cooling and my ambient right now is 21 C. Having ran more than one water cooling pc im inclined to beleive the core temp readings are more accurate than smartguardians are. On a side note, I decapped my opteron for the first time and the delta between idle and load is now only 3 C. Its unbelievable!

[zarathrustra]

Could the differences showing between core 0 and 1 with X2's be a possible reason that one core often clocks higher than the other? Anyone seen a corollation between which core fails first, and which is shown by Core Temp as the hotter?

[soundx98]

Strange, on my RS482 Infinity ITE Smart Guardian says my cpu temp is 42 C. However, Core Temp shows core 1 at 24 C and core 2 at 36 C. Im on water cooling and my ambient right now is 21 C. Having ran more than one water cooling pc im inclined to beleive the core temp readings are more accurate than smartguardians are.

 

Neo, Do you mean core temp 1 or Core Temp 2 is more believable?

I wonder what CBID is showing for an offset?

What model Opteron?

 

On a side note, I decapped my opteron for the first time and the delta between idle and load is now only 3 C. Its unbelievable!

 

Topless and water cooled or not, that does seem unbelieveable.

3 degrees? What are you using to "load"? Try some gamng and see what happens.

Are you saying only 3 degrees dif /delta with SG or with Core Temp?

[ReelFiles]

Could the differences showing between core 0 and 1 with X2's be a possible reason that one core often clocks higher than the other? Anyone seen a corollation between which core fails first, and which is shown by Core Temp as the hotter?

 

I was just thinking that, when I was looking at these screenies, since core 1 always fails on me, above 2.75GHz. I was blaming the memory controller all along. I think it's time for her to go topless, what do you guys think?

 

 

 

 

And I thought I was 8C below my max...

[neogenesis2004]

1) Something between core 1 and core 2 temp seems most believable.

2) Says offset is -6. What does that mean?

3) I have a 165 0610.

4) I measured those load temps with SG while running dual prime and atitool artifact scanner simultaneously.

[soundx98]

1) Something between core 1 and core 2 temp seems most believable.

 

I'd rather be assured that one was correct.

2) Says offset is -6. What does that mean?

Thinking was that you need to add +6 to SG or MBM5, or speedfan. Like I said, I am not sure if that is valid.

4) I measured those load temps with SG while running dual prime and atitool artifact scanner simultaneously.

I'd would be interesting to see what Core Temp/SG are reading at the same time when both programs are running. Strange but Prime95 doesn't seem to get my SC as hot as SuperPi and yet others get hotter with Prime95.

Load temps are much higher with Core Temp for me while Idle temps are very close (2-4 degrees)