Video Card Testing
 

For those of you who have wondered how you could do video card waterblock testing, I have found a way. Atleast a way for the X800, although the same sort of idea should work with any video card that has an ondie diode for reading temps.

This idea came to me when i was looking through EVEREST and found that it was reading the temp through a chip called the National Semiconductor LM63. So i googled it and found the PDF file on it. I then noticed that it read diode temps MUCH like the way that is done on Athlon XP's. So i looked through the PDF a little more and found where the hook ups fo the anode and cathode are on the chip.

So all that needs to be done is to solder two leads to the chip on the x800 (or whatever board and chip you happen to have) and then use a higher quality diode reader. Myself, I will be using the same maxim eval board that pH uses which is very good as it has multiple hook up points so i can get accurat readings from both the cpu and gpu.

Here is a picture of the chip and the hook up points for it on the x800

At least in principle this works on the 9600XT. I soldered twisted pair to the pins of the LM63 that you circled and it indeed reports a diode temp on the MAX6655. I haven't yet cut the diode pins from the LM63; this will have to be done if one wishes to get any kind of useful numbers with the card actually powered on.

The temp readings are really jumpy atm and sometimes report nonsense values. It's a work in progress :)

My experience with the 9800XT's is that they all seem to report nonsense values at times... it will jump from like 45C to 15c and then up to 70C. I think the 6600/6800 may have a better diode reading than the ATI cards.

I think my 6800 has more accurate temps then my mobo and my old 9600xt.

My x800 gives me very stable readings. its the most consistent readings ive ever gotten in general. ill have to try hooking up my x800 to my maxim eval system.

I haven't noticed any weird temperature fluxes on the 9600XT I am playing with now. It seems to report sensible numbers to MBM via ATITool anyway. Still it'd be far preferable to read the temps from a somewhat calibrated source via the EVSYS. I'm gonna mess with it again tonight whilst I'm formatting/installing a couple of systems to get the new test box set up.

so there is still hope. pH, if you have success with this I willl have to try it on my x800.. i just dont want to remove the lm63 though... hopefully it wont adversly effect the results.

What I suggested to pHaestus yesterday (although he already had thought of it) was to isolate the LM63 chip by cutting traces and hooking it up to a dummy diode, then using a standalone diode reader for the onboard temps. So, you don't have to remove the onboard LM63. Just isolate it.

Yea it looks to me like the easiest operation would be to desolder the 2200 pF cap that's across the diode connections right next to the LM63. Then solder your diode reader wires to that and cut the traces between the LM63 and the diode wires. Should work....assuming the diode is really connected to that LM63 :p

Don't forget to be very careful when cutting traces on multi-layer boards! ;)

I'll cut back on coffee before I grab the dremel :D

good luck to you pH. i hope you dont kill the card. I just got my x700pro today and I am hapy to say that it also has the lm63 so it will report temps despite what ppl have said.

taking the dremel to a pcb. please tell me you are joking.... please...

Yes of course I was joking. I played around with the MAX6655 and the 9600XT some again today and I'm pretty sure that something more complicated than "solder wires and done" will be in order. The diode reader reports temperatures when you touch the wires to the legs of the LM63, but if you also touch the bare GPU then it quickly rises to 127C (high error) and stays there. I know the MAXIM board is outputting 10 and 100 uA pulses for low and high; this looks to be outside the spec of the LM63. Perhaps a different reader is in order? It could of course be that pulling the LM63 off the diode loop will return sensible behavior but I'm not so sure. Not sure enough to do a lot of modification to the 9600XT at this moment anyway. Give me a couple days to think about it...

2 days and 10 dollars says by thursday that you willl be holding a LM63 in one hand and a 9600XT in the other :p

if you need any ati pcb layout related stuff or a software api to get the readings into your own software or something just send me an email to w1zzard@gmx.net or contact me on instant messenger

dare i ask do you work for ATI???

Don't think so...w1zzard has just been modding and working with ATI cards for so long, he's probably got a layout of every card downloaded to his brain and his computer lol.

no i dont work for ati ;)

So is MR's comment about the LM63 correct W1zzard? That's where the diode temps come from?

yep thats correct

now to mod my x700..

The room I do all my testing in is still a bit too messy to solder in at this moment. I completely rearranged and reorganized everything to make a more practical space for all the instrumentation and button pressing that I need to do. It's ridiculous having so much equipment; should've spent the $$ on a nice NI DAQ card... The new arrangement is a lot nicer though. I'll post pictures when I get it cleaned up enough for soldering in again.