Right, turns for the bizarre. As I've reassembled the unit, I thought of installing a back fan on the PCI-E-AGP bridge chip. It's sitting on the flipside, unlike the 6600GT I have (on which I also should perform some hocus pocus to have it function).

Best thing I came up with was a Thermaltake fan made for northbridges and small GPU's. It fit the holes and the screws of the original Sapphire cooler, so no extra work needed. The problem was finding another mainboard to test it again, as the newly installed fan made it incompatible with NB heatsinks on both my P4C800 and P5PE-VM. So I've got it onto the TUSL2-M, where it, much to my surprise, sprang to life again.

Now, it's still sure that the MLCC needs to be put back, so I can still use a hand in figuring out what cap values I need to put on that spot. As the MLCC's are close to the double power connectors, I suppose with my very limited knowledge that they are involved in power regulating. And as the card threw its figurative hands in the air during a stress test, my best guess is that the power rail only needs that caps to work under full strain.

The card boots up fine on the tualatin board. So two things to do with this card:

1. find a correct MLCC to replace the missing one

2. make sure the bridge chip gets proper cooling without hampering installation in most mainboards.

Pictures for the hardware buffs:

And yes, she is throwing her hands more or less in the air in a suppliant matter. "Eli Eli! Why fo your sake did you not properly cool my bridge chip!"

Most likely missing capacitors have the exact values as its neighbors since its looks like they all stay on the main power rail and and make a common capacitance. I'd say it makes sense to add missing components but "is this will fix freezing issue?" - probably not. It could be anything. I'd check power rail first.

The PSU is a 620W Corsair,plenty of power on either rail. The P4 is not too hungry, so I'mstill unsure what the main issue is.

I didn't mean PSU itself, but the power rail that comes out from the voltage regulator(s) (this part is under heatsink on tail side of the card) to the GPU on the graphics card pcb. You may find other missing/suspicious components after taking off the heatsink.

With the bridge chip sufficiently cooled have you tried running 3d mark again to see if it'll complete the run?

I have a Sapphire X1950GT 256MB AGP, which I guess is the wimpier brother of your card. I think it's the same PCB.
It has the same bank of MLCCs that you photographed, with 5 out of the 6 positions filled.
I can confirm that all 6 of those cap positions are in parallel, including the one that's vacant. All 5 of the ceramic caps that are installed look identical.

I was hoping I could measure the capacitance across them, and maybe you could do the same, and then we could try to guess the value based on the difference in our measurements.
But I can't get a measurement. I've tried repeatedly, pushing the probes very hard into the solder, and my meter just keeps trying to charge the caps but never resolves to a reading.

Anyway, if you do add to that bank of MLCCs you might as well install 2 of them. All the positions are parallel so I don't see how it can hurt. Sorry I can't seem to help figure out their value.

Does your card have a tiny pair of DIP switches on the back of the card near the GPU? What the heck are those for?
My card doesn't work. It booted in one motherboard when I first got it, but once I started moving it to other boards it never POSTed a display again.

The card has indeed that tiny switch at the back. On my card, it's in position 1 UP and 2 Down. I'll see whether I can check the rail on the card. I have no equipment to test the MLCC's for their values...