Bimole, are you in US?

I have a dead 9800PRO that I would gladly donate to you, maybe you can get one working card from 2...

I'm also very passionate about ATI's 9xxx series but I lack your skills to get mine fixed, also I already have a functional 9800XT and a 9700PRO.

bimole wrote on 2025-01-10, 14:28:

0 ohm at Vcore node.

It's not resistance, it's voltage drop at a fixed current.
As far as I remember, at the limit of 200 Ohms it is 10mA.
And the resistance of the multimeter probes will already be 0.5-0.8 Ohm.

I don’t remember the "resistance" of the Radeon 9800, but if the numbers there are more than 2-3 Ohms at 200 Ohm range, then everything is fine

One of the converters for VRAM is powered from 12V on these cards.
As far as I remember, there are 3 of them.
Separate converter for memory chips, separate for VRAM I/O and another for reference voltage (memory voltage divided by 2)

PS. Replace all electrolytic capacitors
Insert the card into a motherboard that you don't mind loose. Connect the molex.
Turn on
Core voltage should be 1.75
Three converters for memory - something like - 2.7V, 2.7V, 1.35V.
If one is missing, then look for it.

Most Radeon 9800 cards use the reference board design. I'm not sure about bios differences between manufacturers but I suspect that they are mostly reference as well and probably all behave much the same way when the molex cable is disconnected.

I have tested my Radeon 9800 Pro with the molex cable disconnected and it displays the following message:

It’s normal to measure 0 ohm for vcore unless you have a very sensitive multimeter. (or a milliohm meter)

Yes and no...
What I measure is a pure ohmic behaviour. And this is suspicious.
Silicon die is not etched of billions of resistors!
The die is actually full of transistors which have threshold voltages (PN junctions for BJTs, channel pinch-off for FETs...)
IV curves of active devices have always horizontal tangent (Rdyn very high) around 0V (also true for transistors which have both drain and gate voltage rising TOGETHER).
I should have at least some ohms of resistance around 0V at Vcore node.

Anyway will do the measurement with a milliohmmeter next week (I use the electronic lab tools where I work!).

So far I continue to think the resistance "seen" at Vcore is far too low around 0V...
I pushed the sourced current to 15A and tuned the IR camera spectrum between 23°C and 28°C to have max sensitivity in this temperature range so that weak hot spot could be observed...
I finally managed to observe a +0.5°C hot area just under the die (on BOTTOM PCB side) whereas the whole PCB temperature remains the same.
If I measure it on TOP side, the die temperature doesn't change.
That could indicate that the short circuit is under the die, maybe between tin balls. That could also explain why a MOSFET from the dual phase buck blew up.
At 15A, the voltage on the PSU is about 1.5V... BUT almost all the voltage losses are in the cable supplying the board. I didn't measure the "real" voltage at the injection points on the board. Will do it next week...
Let's assume that the short resistance is 1mohm, at 15A the associated dissipated power would be 225mW, which could be enough to increase the local temperature by 0.5°C (about 2°C/W of thermal resistance which is VERY low for PCB thermal resistance...). Actually, I think the short resistance in even below 1mohm...

So do you think a reflow could fix it ??
It's not obvious for me to consider shorts between balls.
Open circuit OK, but not short... To melt the tin balls, die temperature should already exceed the thermal runaway tempteraure, no ?

Bimole, are you in US?

I have a dead 9800PRO that I would gladly donate to you, maybe you can get one working card from 2...

I'm also very passionate about ATI's 9xxx series but I lack your skills to get mine fixed, also I already have a functional 9800XT and a 9700PRO.

That's very kind of you. Unfortunately I live in France.
I would be interrested in measuring things on another board. PM me, maybe we could agree on a shipping...

PS. Replace all electrolytic capacitors Insert the card into a motherboard that you don't mind loose. Connect the molex. Turn on […]
Show full quote

PS. Replace all electrolytic capacitors
Insert the card into a motherboard that you don't mind loose. Connect the molex.
Turn on
Core voltage should be 1.75
Three converters for memory - something like - 2.7V, 2.7V, 1.35V.
If one is missing, then look for it.

Yes, I will try to go on on MOSFET replacements and redo the +12V, +5V, +3.3V supply test on table (before blowing a motherboard!!!).
I also removed the suspicious STS8DNF3LL MOSFET (one of the 3 VRAM buck stages), but it was fine. Maybe I have another short somewhere...

TO BE CONTINUED...

Assuming you have the time and patience, I'd:

1) Wash it. I like dishwashers without soap. I've had a few completely dead things suddenly work perfectly. I put it down to conductive dust, but it could theoretically be 'tin whiskers' which are tiny enough for the 'universal solvent' that is water to dissolve.

2) Pull the core. If the short resolves, you know it was either the balls, or core. A ballout and patience with a multimeter will confirm. If it doesn't, you'll more easily see the board problem.

@bimole, There were some earlier mentions of reballing the core instead of reflowing. Is that something that you have done before and are you able to do that on this card?

It would be interesting to see the process if you could take photos as you progress.

I personally have an FX5900 with a dead core (seems the fan stopped working and it overheated) and would like to transfer the core from another card but I don't have any experience with reballing.

The hard part isn't the a adding of the balls.
That's just a matter of patience with tweezers and tiny balls of you do it manually, and not melting them together with too much hot air.

The hard part is reliably soldering it to the board.
You can't just wave a hot air iron at it. You need a fixed station.

There were some earlier mentions of reballing the core instead of reflowing. Is that something that you have done before and are you able to do that on this card?

I have absolutely no experience with reballing nor BGA chip soldering.
The ''hardest'' task successfully done was a DSP56300 (LQFP 144) replacement on a synthesizer mainboard... And obviously a reflow of Chuck and Bruce legs on a Voodoo 2 😁
Pulling up the core is definitely something I could do but after...
There are no stencils known for these old chips?

1) Wash it. I like dishwashers without soap. I've had a few completely dead things suddenly work perfectly. I put it down to conductive dust, but it could theoretically be 'tin whiskers' which are tiny enough for the 'universal solvent' that is water to dissolve.

Very interesting 👍 I should give a try.
I'm always afraid with normal drinkable water and all ions it contains for electronics...

Thanks all for your contribution that's very appreciated!

JB

If you rinse the card with a generous amount of isopropyl alcohol after using tap water and use a blower on it, then leave to dry further for a couple of hours it will be fine.

Maybe you should replace all the components that you’ve removed and any other damaged components and then reflow the core and test.

bimole wrote on 2025-01-10, 21:41:

That's very kind of you. Unfortunately I live in France.
I would be interrested in measuring things on another board. PM me, maybe we could agree on a shipping...

I can also offer a defective 9800 pro (from France).
(You have a PM)