@ Yin, the first step i took was installing Rivatuner, what a wonderfull application. 2.06 is current and awesome.
the first thing i did in rivatuner was set the fan to direct controll 100%, and to boot with that setting, the stock cooler is decent (on an XFX at least cos we have the fan with a bigger hole and 2 more blades than early reference cards) but imo only when set to 100% constantly. its a little noisy but ive got like 8 fans anyway so who cares. not to mention id rather put up with a few more dB rather than 20 more degrees.
now ive had alot of gfx cards and o/c'ed them all pretty much so i didnt really go in small increments, the first o/c i did was bump the memory from 900mhz to 1000mhz, as i knew these chips are rated at that speed it shouldnt be a problem, and it wasnt, it ran fine.
i had read most 8800GT's can do 700mhz core, with that in mind i shot for 650 straight up, then went in 10mhz increments to 700, which ran 100% stable, i left the shaders linked to the core to get 700/1750, which works out at 702/1728 in reality, the shaders do have more headroom (to about 1830) but this card flies anyway. and in my experience with 8 series cards, the card is more stable with the shaders linked.
if your interested there is a heap of shader ocing things i can get into with ya, as ive found they make a huge impact on performance, on my 2 8600GT's (also XFX and came with zalmans on em) i got the core from 600 to 700, but the shaders went all the way up from 1310 to 1836, shooting the card from ~5000 3dmarks @ stock to ~6700, but that was an ordeal to get stable, as i said if your interested we can get more in depth.
once i had acertained that 700/1750/2000 was 100% stable i made a 1.1v bios and flashed it on, so all rivatuner has to do now is controll the fan.
my temps are roughly 44-48 idle and 58-64 load with fan at constant 100% which really isnt too bad at all for a single slot stock cooler.
so in short Yin i would say, Install Rivatuner 2.06, bump the fan speed, and then start oc-ing slowly and one thing at a time, dont shoot for 2000mem and 700 core at the same time, go one then the other, then i would highly reccommend flashing your card with a OC'ed bios, as it eliminates the card needing to be software oc'ed, and is imo more stable @ the same clocks.
just go slow, be patient, and dont shoot for unrealistic speeds, not every card will do 700 on core, and some will go further... especially with further cooling + vmods... (later....)
at any stage i can help with more rivatuner advice or bios advice, i can even make one for ya (still learn more doing it yourself). and if this was too much/too fast for you, i can dumb it down and we'll go in baby steps.
hope i have/can help m8
wolf.
ps. sorry for the long post, and the reason i thanked you is because you should never be afraid to ask for help.