With GPU makers having decided to call every single ALU a 'processor' the race for the higher number is well underway. Currently AMDs RV770 has a comfortable lead with 800 ALUs over Nvidias GT200-GPU with only 240 of those. In our rather special benchmark we try to find out, what good higher numbers are in terms of pixel shader performance.
A different look at the new GPU-Generations of AMD and Nvidia: RV770 vs. GT200
[Source: view picture gallery]
Disclaimer: Be warned, this article is not for the fps-hungry nor the 3DMark-junkie - we will be juggling with shader-types, fillrates, ALU/TEX-ratios and the like. As this article has been published in its original german version about a month ago, please bear with us the somewhat non-recent drivers used: Catalyst 8.6 Hotfix and ge/force/whatever) 177.35. The Tool
Though it may chance at some point in the future, gaming performance of todays graphics cards is mainly determined by their ability to extract as much performance (thoughput, that is) from their shader units running pixel shader programs.
Thus, we will be using a special tool in this article, which measures fillrates achieved on vast amount of shader programs collected from various classic and recent games. They are all DirectX-9-style games though, because our Tool does not work under Microsofts latest and greatest DX10-API (nor DX10.1 for that matter). That said, let us explain a little further how the results you are going to see need to be interpreted.