GeForce 7800GTX: EVGA's KO Edition

User Overclocking

It would seem that at least for the card that we received, the modified heat sink didn't mean an incredibly greater overclock. We were able to overclock the core to 500MHz, which is a step up from the factory overclock, and we pushed the memory clock as high as 1.35GHz. This is the highest core and memory clock that we've managed to get out of overclocking any of our cards.

The high memory clock speed that we were able to get was interesting, and even beat BFG's 1.3GHz when we overclocked it. This made sense though, considering the fact that this EVGA card is the first that we've seen with RAM sinks on the back.

We're continuing to look into how these cards overclock, and we hope to have some firm answers from NVIDIA soon. It's somewhat frustrating when you have two 7800s hypothetically clocked at 426MHz and the 445MHz, and the reality is that they perform exactly the same (give or take a few tenths of a FPS on average). To us, that seems a little irresponsible, especially when the different vendors can use this to mislead buyers. An example of this would be with three of the 7800s that we've reviewed (including this one), each of which come factory overclocked. They were the EVGA e-GeForce 7800GTX (with a core clock speed of 450MHz), the BFG 7800GTX OC (460MHz core), and the EVGA e-GeForce 7800GTX KO (490MHz ). It would be logical then to assume that because of this, the EVGA clocked at 450MHz wouldn't perform as well as the other two. In fact, that assumption would be wrong.

While the EVGA card clocked at 450MHz is actually the worst of the three in terms of performance (and only by a very slight margin), the reason has little to do with the core clock speeds. It's the fact that the BFG and EVGA 7800GTX KO both have memory clocks set higher: 1.3GHz vs. 1.2GHz. You can see why this is frustrating and can make things confusing when we start to compare numbers between different factory-overclocked 7800s. The graph above should give you a clearer representation of how performance actually increases with the core clock speed adjustments. We will continue to research the frequency scaling issue and report on anything else that we learn.