XFXs Fatal1ty 8800GTS: under the hood

Like we told, every 8800GTS 320MB is NVIDIA factory made, the final video card clocks and heatsink labels is the only thing which each manufacturer has different, with the exception of Calibre that is. The PCB is still not different though, and the board components are also very likely to each other (sometimes different component manufacturers have been chosen, but their functionality remain). The voltage levels of the GPU and memory chips didn't change either, how could XFX possible sell few thousand 8800GTS video cards clocked that high knowing that in previous articles we found it very hard to get even near those clock settings with 10 different samples?

The answer lies in the GPU itself, somewhere around this summer NVIDIA took their G80 gpu to a new revision, the A3 stepping. This new GPU revision seems to clock noticeable better then the old A2 revision, average GPU core overclocks to 700MHz were no longer pure luck and this also seem to have brought XFX to the thoughts to release a monstrous clocked 8800GTS.



A direct compare with the NVIDIA reference model, clocks read with Rivatuner:



The GPU core clock gets a 136 MHz clock boost while the shader clock is upgraded with another 432 MHz. But there is more, the video memory chips also have been changed, where the original 8800GTS was equipped with Hynix branded GDDR3 chips we now found Samsung K4J55323QI-BC12 GDDR3 chips mounted on the board.



K4J55323QI-BC12 chips are Samsung GDDR3 SDRAM modules each 32MB wide and rated for 800 MHz DDR (1600 MHz) at 1,8V. The real-time voltage has been set to 1,87V, this makes those chips running slightly above reference values:



This 10th generation of Samsung DDR3 chips increases the memory bus bandwidth noticeable with the clock adjusted by 207 MHz DDR (414 MHz). This also might influence the total card memory overclock, but let us first have a look at the Fatal1ty's out-of-the-box performance ->