Internal diode vs. socket thermistor vs. calibrated probe

[klosters64a]

Arrrgh!! I thought that ~5C spread took care of the guesswork between socket thermistor or internal diode and actual CPU temperature. OCer.com's recent front page articles about inaccuracies in CPU temp measurement are making me crazy.

It looks like there is no way, short of drilling a hole in your HS--in the EXACT right spot, inserting a National Bureau of Standards certified and calibrated thermal probe and touching the tip of it to the CPU EXACTLY right to get accurate temp readings. Aarrgh!!!

This just isn't right, folks. The need to corral metaphors about stinking Firestone tires and Explorers in aid of CPU thermal reality explanations....

I wouldn't be so annoyed about this if I didn't buy a Duron yesterday!

I suppose the practical way to sidestep this great gulf of thermal doubt is to pick a HSF that tested as a winner here on Ocer's.com. And then ignore any and all reported CPU temps. Oh, well.

 

[Phil]

using a probe touching the edge of the cpu is pretty accurate. I know it isn't as good as the oc.com method but were talking about us mere mortals here. The best way is to work out your expected temp from how many watts the cpu is pumping and then take that and multiply it by your heatsinks c/w (most are around 0.3-4, the really good copper sinks are around 0.2 or less) and add this number to you ambient case temp. As long as this isn't over -+5oC of the read out from the temp probe then you are ok.

 

[wild_andy_c]

I use the ribbon type probe touching the edhge of the die - wholly satisified with it too. Have reviewed a fair few coolers in this fashion.

 

[proze]

phil and andy r on the button, imho.

we can only do so much, to get the right temp.. a compunurse taped properly against the die is good enough in my book!

 

[Shadow ÒÓ]

what's the possible +/- of the different thermal sensors? 5c I'd say woud be a bit much, but any less I could live with. I mean if my processor is 50c it wouldn't matter to me if it were 45c or 55c, it's too hot!

 

[Daniel ~]

Moved Topic to "Cooling topic"

 

[GERRY136]

cmon,
not here long, but ive read and drawn my own conclusions... buy a sensor probe, watch it and get as much as u can out of ur chip.
if ur reading this, u have an internet connection or at least a way to ge to one.
not cheap including the price of ur puter.
having said that, ur not losing a child, or any other family member, so lets not get too overly dramatic.
the most expensive part in ur puter other than the monitor is probably ur processor. as far as i know, 250.00 is the most yu will spend for the best chip.
i think 99.999% of us overclockers could make that in 6 or 7 days with determination. yes, that might mean some type of manual labor, but im sure the majority of the folks here could handle that if need be.
take ur chances with the best cooling u can afford and listen to the vets... beyond that, be prepared to back up ur incorrect decisions, think about the fun u had, and take the necessary steps to acquire the part u fried.
no time to cry, only time for elbow grease and a paychek, not too hard to figure out.

 

[William]

I agree, for use a thermal probe at the edge of the cpu is the best we can do without drilling and would still be better that a software program.

 

[Hoot]

All of my HSFs are drilled for thermocouples ala Joes method. I use a type K thermocouple and Fluke Digital Temp meter to measure temps during testing. Once satisfied, I substitute a thermistor for the thermocouple, connected to a small panel meter. If I get any difference, I note it and move on. I'm not going to get into the fray, but after repositioning my in-socket thermistor, it reads within .5C of the thru-HSF thermocouple. I still do not use it though. I recently bought one of those Enermax/Compu-Nurse type strip thermistors and modified it to fit between the HSF base and CPU ceramic so it can insert all the way up to the core and nudge against it. I wrote an article and it is still available from the front page. Since the article, I went one step further and glued the thermistor to the core with AS epoxy, then restrained it to the ceramic with a thin strip of Capton tape, so it won't get stressed when the connecting cord is moved. It works great. It cost $14.99 or so and there is also a dual model for $20.00, so you can monitor the intake to the HSF fan.
I don't want to start another "feeding frenzy", but the strip thermistor reads about 1.5C higher than the drilled one. This is probably due to it not touching the HSF base and not having a small amount of the heat at the junction wicked away. This, in my opinion, though not per the AMD recommendation, gives a better indication of core temp. The nice thing about it is you don't need to be booted up to see the temp. In the time it takes to boot a PC, you can fry a CPU several times over, if something is amiss.

Hoot

 

[PDL]

I hardily agree. A sensor on the side of the core is the next best way. I consistantly see the 3-4 degree difference that the software is reporting.
Anyone know of a way to get a ribbon sensor into the area where the motherboard sensor is located? I'd like to see if I get the same temp readings as the software!!

 

[proze]

hoot has an article up on that as well.. it involves taking a dremel to your socket, so think carefully!!

 

[Hoot]

No dremels near my socket, thank you. Just bending the thermistor with a needle nose plier. Here's a real basic graphic.

Hoot

 

[klosters64a]

I have a far simpler solution. After ascertaining that the HSF is on the Duron correctly, and that it's cooling properly, I'll NEVER check, measure or worry about the CPU or case temp afterwards. This presumes that the hardware sitch inside the box won't change enough to disturb the cooling "profile."
100 cfm throughput, ~40 cfm being sucked through the PAL-6035, basta!
I've been OCing every damn box I've had for years. This box will run at stock. NO MORE eye on the temps! If the CPU burns, I don't care!!

 

[Paul -The Mad Hatter]

HOOT : Hi, thanks for sending me the article how to put in that diode temp probe from crazypc. I made sure i installed it correctly and it is touching the side of the cpu. I think its way too sensitive because it reads 55c @ idle. i have a duron 700 with a gladiator (viahm reads 35c) any suggestios? what can i do? is this thing whack or am i in trouble? thanks

 

[Hoot]

I'd say you found the best spot for monitoring the temp. Keep that sensor right there! The in-socket thermistors can be way off, depending upon how they are positioned under the CPU and the amount of free air blowing around inside the case and near the socket. Properly prepared, the in-socket thermistor can be a satisfactory method of monitoring the CPU temp. Keep in mind that it is not a "Lab Grade" monitoring device and never will be. I'd put my money on your Duron core being closer to 55C than 35C. A mispositioned strip thermistor sensor, will at the worst, read too low, never too high. There is no software like VIA HM, MBM, ASUS Probe, etc padding the result. Think about it...

Hoot

 

[Paul -The Mad Hatter]

Thanks HOOT for the reply. right now the cpu temp is 53 and via hm is 41. Not as huge a gap. What do you think about taking the averge of the two? (i know this is not extreamly acurate, but it would be an idea where my temps are at)

 

[AMDGuy]

pauldogg (Jun 29, 2001 02:22 p.m.):
Thanks HOOT for the reply. right now the cpu temp is 53 and via hm is 41. Not as huge a gap. What do you think about taking the averge of the two? (i know this is not extreamly acurate, but it would be an idea where my temps are at)

I too have done Hoot's probe mod. My first probe was off by -6 C. I sent it back to Crazypc.com for a replacement. They promptly sent me a new one which was accurate. I checked it against 2 other digital themometers in the house and it matched to +- 1 C. I'd say this is the most accurate way short of buying expensive equipment to measure temps. My insocket thermistor ended up being off by only 1 C so I must have gotten a good one.

 

[Paul -The Mad Hatter]

AMDGuy (Jun 29, 2001 02:32 p.m.):


I too have done Hoot's probe mod. My first probe was off by -6 C. I sent it back to Crazypc.com for a replacement. They promptly sent me a new one which was accurate. I checked it against 2 other digital themometers in the house and it matched to +- 1 C. I'd say this is the most accurate way short of buying expensive equipment to measure temps. My insocket thermistor ended up being off by only 1 C so I must have gotten a good one.

Yeah when i tested it against an in-house themometers it was right on the dot. funny thing is when i put it 3 inches away from my hand the temps jumped about 5c. : confused :

 

[AMDGuy]

pauldogg (Jun 29, 2001 02:46 p.m.):

AMDGuy (Jun 29, 2001 02:32 p.m.):


I too have done Hoot's probe mod. My first probe was off by -6 C. I sent it back to Crazypc.com for a replacement. They promptly sent me a new one which was accurate. I checked it against 2 other digital themometers in the house and it matched to +- 1 C. I'd say this is the most accurate way short of buying expensive equipment to measure temps. My insocket thermistor ended up being off by only 1 C so I must have gotten a good one.

Yeah when i tested it against an in-house themometers it was right on the dot. funny thing is when i put it 3 inches away from my hand the temps jumped about 5c. : confused :

Yep, mine jumped when I put it close to my hand too. That's why I like these probes. They're accurate and quick to respond!

 

[Hoot]

pauldogg (Jun 29, 2001 02:22 p.m.):
Thanks HOOT for the reply. right now the cpu temp is 53 and via hm is 41. Not as huge a gap. What do you think about taking the averge of the two? (i know this is not extreamly acurate, but it would be an idea where my temps are at)

Paul, yes they are sensitive and fast to register a temperature variation. When I walk up to my PC and stir the air, I can see the one monitoring my cooling air intake fluctuate.
As much a you would like to see a lower temp, I would not do any averaging. Have you modified the position of the in-socket thermistor, placing it dead center under the core and touching the bottom of the CPU? Also, applying a tiny dot of AS where it touches that CPU bottom? You may be surprised that the difference between them will decrease.

Hoot