Got Athlon XP Thermal Diode to Read

[Hoot]

Well, it's been a busy few days. Got my KK266+ motherboard yesterday and it works much better than my old rev. KT7A or KK266 with my XP1800.

Today I got my MAXIM 6657 samples and got one hooked up tonight. If you're not familiar with that chip, quick check this article:

http://www.voidyourwarranty.net/review/review.php3?r1=xpdiode/index.php3

It works great, though my quick and dirty implementation looks worse than the one in the article. :D

Here's a few MBM screen shots.

Hoot

[Diggrr]

Excellent job Hoot!

I find it kind of funny how the overclockers and ameteurs demand more precision and function than the motherboard manufacturers do. They still must think we use them only for e-mail or something. :D

[Colin]

Yahoo! Good work Hoot! :D The load temps seem a little low?

[Neco]

Shooot I'd love temps like that.


Some of the guys on this board should start yor own MOD buisness or something, throw together water setups and stuff man you could get rich or something

[Hoot]

Originally posted by Colin
Yahoo! Good work Hoot! :D The load temps seem a little low?

You'd think so wouldn't you, but they correlate exactly to the temps I get with a type T thermocouple bonded to the edge of the core, much like people bond the Compu-Nurse probes. The neat thing is how fast the diode responds compared to any form of sensor attached to an outside surface on the CPU. It really gives you an appreciation for the time delay for heat to propagate from within the core to the outside. Where I used to let my system run Prime95 for a half hour to make sure the reading has stabilized, the Thermal Diode hits the same temperature within several seconds and stays there, while a sensor on the outside of the CPU is still slowly rising. Also, it's so fast that you can see the temperature bounce around a little, depending upon what Prime95 is doing.

If you do the math, running 1708 Mhz at 1.85 Vcore and getting 34C load at 19C ambient, that's only .15c/w, which is the typical result I've been getting with my H2O setup since I installed it a
month ago, using other measurement methods.

I did the calibration recommended at the Arctic Silver web site and the offset result was <1C, which disappears into the tolerance of the MAXIM chip anyway. I might mention that +/- .5C is as tight a tolerance as I would ever trust from either a thermistor or thermocouple based sensor. Manufacturing quality issues and meter tolerances are looser than you think. That's why I stopped losing sleep over a .5C difference in a reading taken from one day to the next. The technology is just not that tight.

For the wiring running from the bottom of the CPU socket over to where I have the IC laying, I used one of those thin, shielded 2-conductor cables you get with CD Rom drives for playing audio CDs through your sound card. I don't use that feature and the size was perfect.

If a person is comfortable with a soldering pencil in hand and has an Athlon XP in their system, I recommend picking up one of those chips. The rest of the materials can be scavenged from the junk box. It can easily be implemented in an hour, including unmounting and remounting the motherboard.

Hoot

[ButcherUK]

I'm still waiting opn a 1668 from maxim (it's on backorder), those things look interesting to solder ;)

[WarriorII]

I hate you :D That's a nice O/C. :cool: Very Nice job.

Q: How accurate is VHM comepaired to the new device?
How does it compair to the BIOS temps?


I am SOOOOO envious of you. - LOL

Good Job.
:cool:

[SickBoy]

This is way cool - I saw that article last week and I'd like to try it once I get a CPU that has the diode.

Does anyone know if the Morgan durons have the temp diode?

SB

[stepback]

Here is a Pcb layout for a 8 channel temp reader.
I will be sending this out for quote from the local fab houses.

If anyone is interested in buying a blank pcb at cost + shipping let me know. I will include a parts list later.

http://home.houston.rr.com/haroldnteri/board.jpg


Try this link ^

[Morbid]

Hey Hoot, I really admire your craftsmanship on the internal diode. ;)

I'd love to have that working in my XP 1800+ ... hey Hoot, if I send you my processor.. will you do it for me... i'll pay you! :D

hehe Good Job on the thermal diode.

[The Coolest]

stepback>> Ur pic is all black can't see anything...

*EDIT*: Went to the link getting it now. kewl

[Rack]

That's acceptable, Hoot.

Now how about some temps with a real wattage generator like CPUBurn's K7 burn?

This one was done with UD/Think, about on part with p95,
maybe a little less hot:

http://www.mit.edu/~the_rack/mbm.png

Rack, AKA Peter North

[pHaestus]

Ah now there are 3 of us. Did you check your C/W values at both speeds when you calibrated with the AS guide? I found that with long leads I would get a variation in C/W with MHz, but when I cut the leads to ~ 5" then those problems disappeared. I think my temps were lower before that little switch in my setup also.

I am waiting on my 1668 and 6659...it is gonna be a fun project to solder them all together.

stepback: All I got was a screen of the title bar at that link. Are you putting these on dummy dimms or what? What ICs will you be using?

[Christoph]

Where did you get the chip? The place I found only sells to businesses and I really want to do this.

[Hoot]

Originally posted by WarriorII
Q: How accurate is VHM comepaired to the new device?
How does it compair to the BIOS temps?


A lot depends upon how well the in-socket thermistor is coupled to the bottom of the CPU. If it touches the bottom, in the center of the core and has a small dab of thermal grease where it touches, you can get surprisingly accurate readings. The biggest factor is the propagation delay of the temperature as it passes through the substrate. It takes longer for the temperature to stabilize. Despite that and the wicking of heat factor, I have checked the ones in my KT7A and KK266 and if you give them time to stabilize, the temperatures are within +/- 1C of either a thermocouple or strip thermistor. I must be lucky though, because I read people who's opinion I respect, saying all the time how poor they are. With the advent of the XP processor and the array of chip capacitors on the bottom of the substrate, the challenge for efficient coupling is a little toughter. Thankfully, most thermistors are coated with an Epoxy shell, which is non-conductive and you can still let it touch the capacitor under the center of the core, even use some non-conductive thermal grease. Again, if you give it enough time to stabilize, you can get reasonable accuracy with the XP setup. I used VHM and Via Tray with all my systems. It is a low-overhead program that suits my needs for a quick temperature check. I do not use the in-socket thermistor during HSF evaluation, other than for comparison. Having just gotten this KK266+, I was eager to get it fired up and was disappointed to discover the in-socket thermistor had leads that were too short to allow it to touch the bottom of the CPU. I tried desoldering them and extending their reach as far as possible, but the thermistor still fell a few mils short of touching. Since I was in a hurry to get it running and since I was equally eager to test the thermal diode, I gave up messing with the thermistor, vowing to eventually get back in there and extend its reach to touch the chip capacitor. When I do, I will post a brief report on how they compare.

Hoot

[Hoot]

Originally posted by Morbid
Hey Hoot, I really admire your craftsmanship on the internal diode. ;)

I'd love to have that working in my XP 1800+ ... hey Hoot, if I send you my processor.. will you do it for me... i'll pay you! :D

hehe Good Job on the thermal diode.

Implementing the thermal diode reader does not require messing with the CPU. You just have to solder a pair of shielded or trifilar twisted set of wires to a couple of pins on the motherboard. It requires good soldering technique, a small iron and if your old like me, a jewelers loupe or magnifying ring.

Hoot

[Hoot]

Originally posted by Rack
That's acceptable, Hoot.

Now how about some temps with a real wattage generator like CPUBurn's K7 burn?

This one was done with UD/Think, about on part with p95,
maybe a little less hot:

http://www.mit.edu/~the_rack/mbm.png

Rack, AKA Peter North

I agree, you have to love BurnK6 and its successor, BurnK7, for real heat generation. They have alway kicked Prime95, F@H, SETI, etcs, Butts for pure heat generation. Unfortunately, most people look to Prime95 as a universal standard, probably due to its error checking capability in tandem to heating. As you might expect, BurnK7 produces more heat. Here's a screenshot:

Hoot

[pHaestus]

You should list the 6657 diode reader in your specs hoot :)

[Hoot]

Phaestus. I forgot to address your previous post.
BTW, I did not realize who Rack was when he PM'd me to join you both earlier.

When I ran the calibration, I used 1709 @ 1.85V and 854 for my speeds to make an easy 2.0 conversion comparison. My resulting figures were (34C-20C)/(27C-20C) which yielded 2.0, so I feel good about my diode accuracy.

The Hi-Z shielded pair cable length did not seem to have any impact upon my results. I did implement a 2200pf multi-layer chip capacitor at the input of the 6657 and the .1uf Tantalum decoupling capacitor on its VCC. I only connected the shield to the motherboard end, to create a Faraday Shield and avoid any chance of a PSU ground current loop through the SMB connector. Moving the shielded thermal diode lead arond, within my case had no effect on the reading. Last night, with my "cheap and dirty" quick setup, I did not take those precautions and saw my readings fluctuate when I moved either the diode lead or the leads going to the SMB connector.

Here's a picture of my final product:

Hoot

[Hoot]

Originally posted by IdeaMagnate
Where did you get the chip? The place I found only sells to businesses and I really want to do this.

It's been a busy evening replying to everyones posts. I went to the Dallas/Maxim Web Site and filled out a sample request form. Naturally, I stretched the truth a little bit :D

John "Hoot" Hill
Chief Engineer
Hill Laboratories

[pHaestus]

Wow hoot that is a nice setup! I am going to be forced to improve mine some I think :)

[pHaestus]

I set up my maze2 + heatercore + eheim 1250 last night on the testbed. I think my system can be improved and tweaked a bit. I only have silicon tubing between the waterblock and pump now, and some of the other areas are flattening out a bit. I think that the temps were still quite good (Prime95 load):

Intake air temp: 23, CPU (diode) 33

That is with a 1600+ at stock speed and voltage (1.75), so a C/W of ~ 0.18. Pretty good agreement with Hoot's rig (and we have similar setups).

[pudgy-duck]

FINALLY got my MAX 6657 chips last Friday. I tried using Hoot's DIP IC Socket idea. Made a mess of it!! Fortunately I ordered three chips. So, yesterday I dragged out my stuff for making Printed Wiring Boards. Made a quick trip to Rat Shack for PCB etchant, etched up a little PCB. Works great. Soldering the MAX 6657 to a Printed wiring board is so much easier than trying to solder wires, etc to it. That's like trying to solder to the legs of a tick. I thought soldering the wires to the CPU socket would be the worst part, but it really wasn't that bad. All it takes is a decent soldering pencil, and some small diameter solder. I just stripped back the wire a little, and tinned it with solder. Tinning shrinks back the insulation a little. I then trimmed the exposed tinned wire back to just a nub. Then I put some fresh solder on the CPU pins. Hold the wire fairly vertical on the CPU pin, and just touch it with the iron from the side. Done. But you do need a soldering pencil and a tip that comes to a fine point.

Hoot;
Got a couple of quick questions. In the MAX6657 data sheet, it says "At the device, connect the twisted pair to DXP and DXN and the shield to GND. Leave the shield unconnected at the remote sensor." In your thread, you said you connected the shield to GND at the sensor and didn't GND it at the device. Just the opposite. Did you experiment and find this better? Do you think it really makes any difference? Now that I am finished, (hind sight is always 20/20) I can see that if you Grounded the shield at the CPU, and soldered the GND shield to the big Ground area in the middle of the CPU, that would act as a stress relief for the other two wires. By the way, great idea using the CD ROM audio cable. Worked great! I even left the connector on the end connected to my PCB, so the 6657 board plugs on and off the cable to the CPU pins. That way if I make another one, I can test it without having to do any soldering on the MoBo.

Also, in the pictures of MBM you provided, one of the temps is labeled NorthBridge. Do you have a temp probe attached to the NorthBridge?

Any way, Thanks for inspiring me to do this Hoot.

Pudge

[Hoot]

With the shield acting as a Faraday Shield, IE only connected on one end of the path, it really does not matter which end you use. I chose the groundplane under the CPU socket for exactly the reason you observed. A strain relief.

The Northbridge is a conventional Thermistor read by the VIA 82C686B, not a thermal diode. I just wanted to put something else in the dashboard display. :D

Glad it worked out for you. I thought of etching something if anyone expressed an interest in my add in the classifieds.

Hoot

[pHaestus]

I wish I could see the classifieds; I could use a few QSOP pcbs. I finally got the 6655 soldered to wrapping wire:

http://www.procooling.com/~phaestus/max6655.jpg (jumper and dime for scale)

A pcb would be convenient though :)

[Hoot]

pHaestus

PM on the way...

Hoot