A long time ago I has a GForce2Ti card, passively cooled with HUGE glued on heatsinks that ran at GPU/mem 268/240Mhz clocks stable. But one day I decide to cut a supposedly unused part of the board to gain better airflow in the case from the top right corner of the card. The card still works, but instead of 64MB of videoram I got now only 32MB and the game Return to the Castle Wolfenstain started jerking because there was suddently not enough videoram for textures... So I wanted a quick solution and bough this cheapo FX5200 card and regret that ASAP. It was notably slower, even I managed to overclock it from default 250/166 to 295/205... This disapointment lead to the buy of Radeon 9100 that lead to even bigger disapointment with speed and drivers problems in SoF2, ma favorite game at the time, but that is another story. So, that is how I got the card, lol. It was used sometimes for testing purposes, the Zalman heatpipe cooler is, however, sometimes limiting - the card did not fit everywhere, but I will work on that later (if ever, lol) Overview of the card: Now during the ECS sponsored OCW contest to reach the lowest 3DMark 2005 score ( http://my.ocworkbench.com/bbs/showthread.php?t=62120 ) I thought I could use the card to gain lowest score of all participiants, perhaps even WR when I use the crappy slow VIA C3 CPU in Abit BX133-Raid mobo I recently recapped But in serious downclocking the card become unstable. Obviously the G-Luxon caps are the culprits. There they are: The card use these caps: Gigabite FX5200 --------------- 5x 1000uF 6.3V d10 G-Luxon 1x 510uF 4V d8 Sanyo Os-con 1x 100uF 10V d5 Nichicon VR Obviously the Os-con polymer is great cap and since I have no bigger/betteer polymer ATM, it stays. It is obviously used for the GPU voltage. That is probably why I can clock the GPU as low, as t 8Mhz and it still working well in 3D Memory is another story. With 66Mhz it never pass 3DMark (took 12 - 16hours with stressed machine, so, go figure...) and at 60Mhz it can't even initialize 3D! With G-Luxons, of course. Now with Samxons GC: It can run 3DMark at 50Mhz! Any maybe lower, I did not tried the full test yet and maybe it fail still, because it is just like in half, so... We see. Neverless it is a clear change for better overclock/underclock as they are basicaly the same. Both put a lot more stress on voltage regulators I used 1500uF Samxon GX d10 caps to bump the capacity a little, because I believe it is important when downlocking a lot. But the card design is very odd. First at all, it reverse the markings on PCB. In all other cards/mainboards the white half of the circle on PCB bellow a cap mean the "-" polarity. On this board (and on the FX5600XT one, check there: http://my.ocworkbench.com/bbs/showthread.php?t=58455 ) it is REVERSED! But that is not all odd. Notice how far (first photo) are the caps form the rams! This is not very good. Caps have to support the voltage as close to the place where it is being "consumed" as possible. Having them on other side of the card is not exactly good design. Perhaps that is why Gigabite is not printed over the board anywhere? They are ashamed of the design, probably, and rightfully they are! See this? Someone realized that the rams need additional support, and get there ready a place for another cap, but for the cost reasons it was never included. That of course call for adding a nice SMD d8 polymer there to support the voltage filtering and get better the overlock/underclock At least something around 470uF will help for sure. This place is connected directly in parallel with one of the Samxon GC 1500uF (top right one), so... So at least I laped the core well And now let's try score really, really hard! This that get me score 31 points was only a start!