Something strange in thermal diode readings

Cooling Processors quietly

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
Old Dude
Posts: 47
Joined: Fri Oct 11, 2002 10:59 pm
Location: Finland

Something strange in thermal diode readings

Post by Old Dude » Thu Nov 14, 2002 4:08 am

I purchased a Soltek SL75DRV5 motherboard to be able to underclock my XP 1700+ processor to 1.1V and take processor temperature readings directly from the processor's thermal diode. Now I'm convinced, however, that the readings are not always correct.

The prosessor has been unlocked and thus I was able to make the following test: I run the processor at 133MHz FSB and clock ratios from 6 to 12. Vcore was a constant 1.75V, the case was open and on it's side. The heatsink was a SLK-800 with a Panaflo fan. The results are on the following table (all temps in degrees C):

clk ratio_____diode_____socket
___6_______44_________34
___7_______46_________35
___8_______48_________36
___9_______57_________38
__10_______58_________39
__11_______58_________40

As can be seen, socket readings increase regularly but there's a big jump in the diode temp from clk ratio 8 to 9. The processor's temps cannot behave that way. Now I'm uncertain which one to blame: the XP diode or the motherboard. Does anyone have an insight?

crisspy
Friend of SPCR
Posts: 228
Joined: Sun Sep 29, 2002 9:05 pm
Location: Powell River, BC, Canada

Post by crisspy » Thu Nov 14, 2002 2:10 pm

Ok, the socket temp is probably a teeny smd thermistor right on the mobo under the socket? If so, it has a very flattened response curve. Does do a little jig from 36 to 38. Maybe mobo starts to loose heat a lot faster as temps go up, more so than HSF which is already very efficient and probably fairly linear.

OTOH, maybe the diode or system monitor chip just has a bad non linearity on your system. Just takes a big swing for no sane reason. Weirder things have probably happened.

Still, a CPU can't go jumping 10 deg in such a non linear way like that. Gotta be the diode is crazed. Can't figure out why or how that could be wrong though. Check out the link to Analog about themal diodes down the 'Heat Spreaders?' thread. Maybe clues there, or elsewhere in their tech docs.

I suggest you try a flat thermistor up top beside the core. Even if you only have a multimeter to read it's resistance, you would still see a BIG spike if that diode is right.

You got a good head scratcher there Old Dude :?

Asmordean
Posts: 36
Joined: Sat Nov 09, 2002 10:12 pm
Location: Calgary - Canada

Post by Asmordean » Thu Nov 14, 2002 3:20 pm

Crisspy - The diode is inside the CPU core but that still doesn't explain the strangeness.

Does a program like Motherboard Monitor confirm that the voltage is staying at 1.75v? The reason I ask is with my system (unlocked Duron 1100) I have the BIOS set to 1.75v but MBM reports the voltage at 1.81v at 1100MHz. If I bump up to 1200MHz but leave the CPU voltage at default it reports it as 1.83v.

If I set it to 900MHz the voltage goes the otherway and reports it as 1.74v.


So, to run my PC at 600MHz I have to set the voltage to 2.0 to get it to supply 1.75v or I won't boot. I am running at 1100MHz with a voltage set to 1.7v which MBM reports as 1.76v.

Perhaps your BIOS is screwing up in the same way?

crisspy
Friend of SPCR
Posts: 228
Joined: Sun Sep 29, 2002 9:05 pm
Location: Powell River, BC, Canada

Post by crisspy » Thu Nov 14, 2002 7:19 pm

Very interesting point Asmordean. The voltage regulation has got to be a prime suspect. The on die (yup, knew that) thermal diode should not be a fault here. The SMB system monitor that reads it could be. I wouldn't be suprised if the core temp was way steeper & higher than the socket thermistor either. But the jump from 48 to 57 is way more than one clock ratio step explains, and we all know that the two things determining CPU wattage are F and V.

So get a multimeter on that Vcore Old Dude. Don't even take the mobo's word it. While no modern design should suffer from unstable Vcore, you could easily have a bad board. Also maybe compare your multi meter against MBM's 12v, 5v, and 3.3v reports.

TerryW
Posts: 85
Joined: Sun Aug 11, 2002 3:26 pm
Location: Coquitlam, B.C. Canada

Post by TerryW » Thu Nov 14, 2002 7:59 pm

Hi all,

I have a few simple questions that I cannot get a confirmed answers:

On the Athlon 1800+ CPU what is the settings and/or details for the CPU thermal diode?

How do users know how to setup a monitor utility (like MBM5) to monitor the internal diode & compare to the thermistor in the CPU socket?

Acording to AMD, the internal thermal diode is used to reduce the heat and cannot be measured. Who is correct?

This seems all too easy, but I haven't found proper answers. Sigh ....

TerryW

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Thu Nov 14, 2002 8:55 pm

A LOT of mobos don't read the thermal diode properly, they continue with the in-socket thermistor. If you have a board with just the latter, you KNOW it's unreliable. The diode is much more accurate. It is the motherboard that provides the output; MBM can only display the ouput the motherboard provides.

Here's how to tell what the motherboard reading is from: install MBM5. Figure out which output is the CPU temp by comparing with the reading in the BIOS -- they should be close.

Once you have MBM5 to give you some kind of CPU temp output, set MBM5 to monitor/refresh the reading every second.

Then try any type of operation in the system that pushes the CPU. It can be just opening a program, running a benchmark or browsing quickly.

1) If the CPU temp is being read via the diode, the temp shown will change almost immediately, going up with usage and down quickly again when the operation is done. IE, it is quick, to go up in temp as well as down.

2) If it's the socket thermistor MBM is reading, the response to CPU usage will be much more sluggish. The temp might not change for 5-10 seconds even into a stress test, and when it does, it does so slowly. The temp also comes down slowly. This delay is due to the insulating effect of the CPU casing. It takes time for the core heat to filter down to the socket.

Old Dude
Posts: 47
Joined: Fri Oct 11, 2002 10:59 pm
Location: Finland

Post by Old Dude » Thu Nov 14, 2002 9:11 pm

I took all the temps in the BIOS. I did not check the Vcore in every measurement but I've never seen anything abnormal in Vcore readings when looking it. The BIOS has always showed the voltage about 0.02V below the set value and MBM5 0.01-0.02V above.

So far no-one here has suspected that the processor's diode could be faulty but in another forum one member did so. I wonder if it's technically possible that the diode itself could cause the erratic readings.

Old Dude
Posts: 47
Joined: Fri Oct 11, 2002 10:59 pm
Location: Finland

Post by Old Dude » Thu Nov 14, 2002 9:42 pm

Here http://www.hardtecs4u.com/reviews/2002/ ... ndex47.php is a description of Soltek SL75DRV5 thermal diode (frankly in German) with a picture taken from the BIOS where you can see how the temps are shown there: Temperature ABS II = the thermal diode temp and Temperature 1 = socket temp. Quite a big difference in temps in there too, 18C!

The sensor for the diode in MBM5 is LM83-7 (Remote1), informed in the review, and it's equivalent with Temperature ABS II readings. Like MikeC described, the diode's temp changes are very quick. Once I measured temps with the heatsink fan off and got diode 76C and socket 59C. Then I connected the fan to a mobo's header, and the diode reading went so quickly down that it was at one stage lower than the socket reading!

crisspy
Friend of SPCR
Posts: 228
Joined: Sun Sep 29, 2002 9:05 pm
Location: Powell River, BC, Canada

Post by crisspy » Thu Nov 14, 2002 11:01 pm

Terry, Old Dude, check out Thermal Diode Calibration from SPCR's Unique Heatsink Testing Methodology regarding thermal diodes. Mike really gave us the lowdown on thermal diodes in there.

TerryW
Posts: 85
Joined: Sun Aug 11, 2002 3:26 pm
Location: Coquitlam, B.C. Canada

Post by TerryW » Fri Nov 15, 2002 10:41 am

Thanks to all for the helpfull hints. I'll research this weekend.

Cheerio!

TerryW 8)

Old Dude
Posts: 47
Joined: Fri Oct 11, 2002 10:59 pm
Location: Finland

Re: Something strange in thermal diode readings

Post by Old Dude » Sat Dec 07, 2002 11:24 pm

Old Dude wrote:I purchased a Soltek SL75DRV5 motherboard to be able to underclock my XP 1700+ processor to 1.1V and take processor temperature readings directly from the processor's thermal diode. Now I'm convinced, however, that the readings are not always correct.

The prosessor has been unlocked and thus I was able to make the following test: I run the processor at 133MHz FSB and clock ratios from 6 to 12. Vcore was a constant 1.75V, the case was open and on it's side. The heatsink was a SLK-800 with a Panaflo fan. The results are on the following table (all temps in degrees C):

clk ratio_____diode_____socket
___6_______44_________34
___7_______46_________35
___8_______48_________36
___9_______57_________38
__10_______58_________39
__11_______58_________40

As can be seen, socket readings increase regularly but there's a big jump in the diode temp from clk ratio 8 to 9. The processor's temps cannot behave that way. Now I'm uncertain which one to blame: the XP diode or the motherboard. Does anyone have an insight?
Update: I changed my processor from the Palomino to a Thoroughbred XP 1700+ and made the measurements anew. The test setup was almost the same, the only differences were that now the Vcore was 1.5V instead of 1.75 and the fan on the HS was a Papst 12dB model. The results I got are on the following table.

clk ratio_____diode_____socket
___6_______35_________31
___7_______36_________32
___8_______37_________33
___9_______38_________34
__10_______39_________34
__11_______40_________35
__12_______41_________35

As can be seen there are no irregularities anymore. So in addition to a cooler running processor I got a properly working CPU thermometer :D

crisspy
Friend of SPCR
Posts: 228
Joined: Sun Sep 29, 2002 9:05 pm
Location: Powell River, BC, Canada

Post by crisspy » Sat Dec 07, 2002 11:42 pm

Thanks Old Dude, very interesting info. Now I am just itchin': what voltage on the Pabst. I guess 12? The rest of your data is so precise it makes an interesting performance example.

Old Dude
Posts: 47
Joined: Fri Oct 11, 2002 10:59 pm
Location: Finland

Post by Old Dude » Sun Dec 08, 2002 12:08 am

Yes, 12V. It works then at 1500 rpm according to the specs.

Red Dawn
Posts: 169
Joined: Fri Oct 04, 2002 11:46 am
Location: Stockholm

Post by Red Dawn » Sun Dec 08, 2002 6:36 am

Very interesting. The thoroughbred xp you bought, did you have to ask for it specifically, or did you just cross your fingers and hope you'd get one? I want to buy one as well, but the possibility of receiving a palomino instead is somewhat bugging me. Also, what stepping did you get on yours?

quokked
Posts: 333
Joined: Sun Aug 11, 2002 3:26 pm
Location: Melbourne, Australia
Contact:

Post by quokked » Sun Dec 08, 2002 6:43 am

Red Dawn wrote:Very interesting. The thoroughbred xp you bought, did you have to ask for it specifically, or did you just cross your fingers and hope you'd get one? I want to buy one as well, but the possibility of receiving a palomino instead is somewhat bugging me. Also, what stepping did you get on yours?
Hey red dawn if u're worried about that u can ID the type of CPU core you're getting by the stepping marked on it
http://www.overclockers.com/tips00173/ (good starting point)
there's a few other articles on the front page currently talking about the IDing of the chips, but there are exceptions http://www.overclockers.com/tips00201/

Old Dude
Posts: 47
Joined: Fri Oct 11, 2002 10:59 pm
Location: Finland

Post by Old Dude » Sun Dec 08, 2002 6:44 am

The store I bought my processor from is selling both palomino and thoroughbred XP 1700+ models so there was no risk involved of getting the wrong model. I don't know anything about the steppings. Anyway, the default voltage is 1.5V.

Update: The processor is stepping A. I confirmed it from the order code on the store's web site.

Red Dawn
Posts: 169
Joined: Fri Oct 04, 2002 11:46 am
Location: Stockholm

Post by Red Dawn » Sun Dec 08, 2002 7:12 am

The reason behind me asking is that I've yet to see vendors here in Sweden seperate the two existing cores, they just sell them all under xp1700+ and use the old "palomino" name for all of them, so it's uncertain you'll get a thoroughbred if you ordered, it could go either way basically.

Thanks for the links quokked, but I'm not sure I can put them to good use, it only depends on the fact if I can get a look at the actual core before buying it, and if I buy online that option goes out the window, unfortunately. :|
I'll have to check and see if I can find a good local store where I can take a look at the core prior to buying it.

Edit: Sorry, I just noticed the thing about OPN's just now, and I found one web shop that lists them (the processors) by it. I'll start reading first and replying later on in the future. :)

Old Dude
Posts: 47
Joined: Fri Oct 11, 2002 10:59 pm
Location: Finland

Post by Old Dude » Sun Dec 08, 2002 8:24 am

There are web-stores here in Finland where you can send the product back for full refund. You even get back the delivery costs. So there's no risk involved.

Post Reply