The cards tested are a standard 256MB 7800 GT and Gainward’s 512MB 7800 GT. A typical 7800 GT straight from Nvidia with no bios modifications made by the AIB partner will contain a delta of 40 MHz. If you have no idea what I just said visit this page. Gainward’s card does not have any delta, if the core is clocked at 400 MHz, the geometry, vertex and ROP clocks will all be 400 MHz, too. To solve this inconsistency I modified Gainward’s bios using NiBiTor, added a 40 MHz delta and flashed the new bios to the card. Obviously Gainward’s card is also clocked much higher so I brought the clocks down to standard 7800GT speeds, too. Both cards were clocked a 400/500 with a delta of 40 MHz and the memory timings are identical. With these modifications there are no distinguishable differences between the two cards other than memory capacity.

The tests consist of the usual frame rate comparisons at various resolutions and image quality settings, but they are more geared towards evaluating what happens when vast amounts of memory is requested rather than looking at playable gaming settings. I also decided to use an excellent little app called Video Memory Watcher which allowed me to watch how much texture memory is used in both the VRAM and system RAM during testing. I’m not entirely sure how reliable this data is but from my limited testing the results attained are repeatable and in my opinion reliable enough. Still, the memory usage results should be read with a pinch of salt, albeit interesting salt.

In order to keep things consistent, Video Memory Watcher was not running when the frame rate benchmarks were run just in case it causes slow down.

Games tested will be HL2: Lost Coast, F.E.A.R and Quake 4.