Gigabyte FX5600XT recap

My sis got this board, as attempt to make the Sims2 on her machine run a bit faster over Radeon 9000. The card history is long, but to make long story short, it served me as backup card well, when I killed my FX5700Ultra, and it even show great O/C potencial (under watercooling), but novadays I did not have any use for it, and sis like it.

However I noticed that in extra stresfull situations (like in 64k intros fr-019: poem to a horse there or Life Index (curiously not present in download section, but you can grab it there: ) that the image like to freeze and then re-appear (no change to music or intro run) later. This indicate that the core have a problem. Either overheating, or undervolted on load. Quick test (decreasing O/C to stock) proved it is the second case, and quick look at the card shoved Evercon caps.

Now I already suggested them to be added to the bad caps brand, based on my experience with JetWay N2PAP Ultra board, I know what went wrong.

There are five 1500uF 6.3V Evercon caps. There are also a Sanyo 510uF 4V Oscon cap too, but only one. I heard that these are excelent, so no change. Tough I planed I used there a 560uF Pannasonic FM one, I skipped this idea, because it turned out to must be a SMD one and the Panny it is not.

Now what we all waiting for - the results.

The Gigabite 5600XT card is what XT mark in nVidia. Pure crap. Come with default 235/200MHz clocks. With watercooling I manage it to unbeliable 375/268MHz. Now the question is, what happen on passive only cooling (Zalman ZM80-HP) in fanless case, yet with exchanged caps from Evercon ones to Samxon GC ones.

Why Samxons? Because Pannasonic FM ones are too big (won't fit under the cooler, even the Samxons are very close - curiously enough - the Evercon ones are even 3mm less tall that the Samxons!) and because Big Pope was very helpfull for providing this small testing sample to me free of charge :eek:

There is another reason for it, I just recapped the whole sis mainboard with Samxons and gained excelent results, but that is for other thread I preparing to create.

Anyway, the clock results. First at all, core has to get into 355MHz to be stable - yet this is still good, considering that the card is not watercooled anymore like it was:
(illustration pic, DangerDen Maze4 GPU block on FX5200)

Now the rams, well, they started crapping out a very little on 283MHz now!
That is pure 15MHz increase, just for caps exchange. Well, for stable run i backed down to 275MHz just to be sure, but then again - the demos won't stop anymore like they used to do before and 275MHz is still a 7MHz increase 🙂

Very very very good, if you ask me 😄

Of course, the card is oldie crap and nothing help it, but… The overclock from 235/200 to 355/275 is impressive. Is not it?
Any mainly - stable!   😄

And there is a valid lesson learned from this card. Most of the big caps (with exception to one), could be replaced with also SMD variants. They are usually with lower specs, so no interest from me, but… but the orientation of the caps not seems to be right. The white parts of the PCB, that says "ground" are in the wrong place. When I followed them, the caps are reversed and quickly start bulging...
So I have to replace them once more.

Therefore lesson learned - never trust the mainboard ground orientation for anything - especially when it did not seems to be right when soldered caps. Therefore measure first. Also pay attention witch caps hole has a rectangle around the hole - that is the positive hole! Always. Even in this, reversed example.

And now it is time to get back to this card to test one suggestion - stacking ceramic caps! 😄

First at all, I replaced every caps for a top-spec Nichicon ones:

Gigabyte FX5600XT
1x 1800uF 4V Nichicon CK SMD (d10) (5600XT) PCK0G182MCO1GS
3x 1500uF 2.5V Nichicon LE (d10) (5600XT) PLE0E152MDO1
2x 1200uF 4V Nichicon R5 (d10) (5600XT) RR50G122MDN1
2x 1800uF 6.3V Nichicon LG (d10) (5600XT) LG0J182MDO1
2x 180uF 6.3V Nichicon CG (d5) SMD (5600XT) PCG0J181MCO1GS
17x 10uF 6.3V Murata SMD 0805 (5600XT) GRM21BR70J106KE76L

...and then I teied something, that I never did before. And that is "stacking" the ceramic caps. The thing is, that todays are made caps at size 0805 (that means they are 2mm long and 1.25mm wide!) even at 10uF specs for 6.3V (beware of DC bias, at 6.3V the cap has -60% of his capacity, it holt it's specs up to around 2V, on 2.5V it is like -10% capacity, so...) and therefore I try support overclock of this card by this means 😉
From the top side of the card, it was easy. Just push the little cap near the existing one, holt it by small stick and solder quick and carefully from one side (it is 2 x 1.25mm big!) and then from other side and it is done:

Unfortunately from the bottom PCB side, there are troubles. There are in very close proximity, where the side-stacked ceramic cap ended up, soldering points to some PCB layers. Not that will lead to instantly soldering the cap to them also, witch in the end could kill the card, or at very least to make it not working well.
At first I got an idea, that I could add there just for the soldering some pad, that then I will remove after the soldering and I have guaranted that there is air space between the added cap and the soldering point(s). However upon second tough, when I will remove the pad, then it could damage (rip off) the original PCB and cap from it as well… Also when it get bendt back to the PCB by some manipulation, then it create instant card crash... So that was not a good idea.
Second idea looked at first as a very bold move - just place the cap on the top of existing and solder it to it. Hmmm. First problem is, that the stick is notably thickier that the caps lenght are...  ;D Yea, that is fixable 😄 Second problem is breath, because even deeper inhale or exhale move the cap and mostly knock it out of the bottom one...

But after a while trying I made it and the caps are added - stacked:

And now the best thing - the card is still working!  ;D I'm amazed by it myself… 🙂

So, what is the overclocking after these modifications? Well, there it go.Standard clock is 234/200MHz (sometimes it is reported as 235 and nVidia showing the memory clock as 2x…. in the DDR "translation") and how much I can push it?

What about from 234/200 to 400/275MHz?

Not bad, hmmm? Now this is sadly not stable with simple aircooling:

…but a whooping 390/275MHz is stable!

Pushing the clocks futher: