LOSTCIRCUITS

DX9 and 3DMark'05

3DMark'05 is one of the more controversial graphics benchmarks out there, mostly because there are cheats to inflate the scores either through driver optimizations or else through relatively simple modifications to the mainboard. For our purposes, though, 3DMark05 is working just fine, after all, the majority of graphics adapters tested are running on the same drivers (ATI) anyway and second, we are not interested in generating inflated scores for bragging rights but show the bottlenecks in performance under different runtime situations.

No AA

Without AA and at lower resolution, SLI configurations, even those based on the GF 6600GT, beat the crap out of any single graphics card, regardless of how high-end it might be .

4xAA

Antialiasing requires more on-board memory and one thing that current SLI configurations cannot do is toggle memory space for on-the-fly sharing between two adapters, meaning that one card could use the on-board memory (frame buffer) of the other card. This leads to insufficient memory in ultra high-end situations for the 6600GT with its only 128 MB of frame buffer - which is reflected in the fact that the GF 6600GTs won't run 3DMark'05 above 1024x768 with 4x AA enabled. Keep in mind though that in real life gaming we have not seen the equivalent to this particular situation

The X850 XT PE is the undisputed winner, even on an "outdated" platform and interface.

Next Page:    => Single Benchmarks: Firefly Forest =>