If you wanted to ship an application in 2002 and take advantage of full programablity you needed OpenGL, DX9 wouldn't cut it. Not only does OpenGL have better than shading than PS2.0, but they had it before MS released DX9!
DX9 has a lot better shading then PS 2.0 or OGL 1.5- PS 3.0. It will all be spelled out even more explicitly then it has been stated the thousands of times to date when the next gen parts hit.
For the general question, DirectX is in an entirely different league then OpenGL when it comes to gaming, Carmack uses quite a bit of DirectX for his games- what people are trying to compare is Direct3D to OpenGL- D3D is one part of DirectX. OpenAL which last I was aware is a work in progress is supposed to be a cross platform standard to take care of one of the other major DX components(audio support). DX includes video support, audio, 3D visuals, 2D acceleration, control input, networking protocols- DirectX is close to a top to bottom gaming solution, OpenGL is just a 3D API.
Limiting the comparison to OpenGL versus D3D OpenGL has a rather large 'unfair' advantage- John Carmack. Without Carmack and all of his engines OGL would have been dead as a consumer level API a long time ago. Back when Carmack started D3D was an absolute mess, code was extremely complex to do even simple tasks and performance was poor. At the time there was Glide and OGL as reasonable 3D APIs and Glide limited support to 3Dfx boards. Due to this far more then Carmack's cross platform ideals(which really aren't as strong as many people make them out to be- although he does believe in supporting Linux and Mac) he ended up going with OpenGL and stuck with it. Since that point in time we have likely seen a couple hundred titles built off of Carmack's various engines and that level of penetration assures that IHV's must support the API on relatively equal footing with D3D. Now we have Croteam et al coming up with new engines not directly related to Carmack, but still owing the possiblity for their existance to Carmack's choice to stick with OpenGL over D3D. As of right now there is no realistic way to make a game for Windows without using DirectX, titles just rely on OpenGL for 3D support.
In terms of comparing the two it is fairly difficult to do. One's strength is the other's weakness and vice versa. D3D being controlled by MS has an enormous upside- you
know what features are in and out of the API and as long as a board has hardware support you use those features in the same way(not saying performance will be optimal, but the fact that it will work as long as it's DX compliant). On the flip side of that, if any IHV's going beyond or outside of the direct specifications there isn't a good way to expose it under D3D while with OpenGL they can simply utilize a proprietary extension to enable features(talking about the IHV first, and then the developer).
Both API's are heavily influenced by politics, OpenGL just has a way to work around them if they don't come out in your favor while with D3D you are SOL if MS decides to go a different direction. Again from the other side with MS you get regular updates and significant changes to the API on a comparitively frequent basis while OpenGL's advancement in terms of API wide solutions is glacier like in comparison. This can make making a title work properly on a wide range of parts difficult. OGL 1.5 as a good example still doesn't have wide support(you could argue any real support) while DX9 is obviously supported by pretty much everyone(in terms of driver level functionality at least).
In terms of overall performance it is very hard to gauge accurately as overwhelmingly when you talk about an OpenGL game you are talking about a Carmack engine and he is the best in the business at squeezing the best performance with the most demanding visuals out of people's machines(not sure many people would argue that point either, though I'm open to debate if any are so inclined
😉 ). Looking at Quake3 now it is scaling up in to the stratosphere with framerates, pusing in the 500FPS for some people while the original Unreal has trouble hitting 200FPS despite being significantly older and having, in terms of end results, far less demanding code to be dealt with. I use Unreal mainly as I consider Sweeney to be second only to Carmack in terms of engines, and he still has a long way to go. The Unreal2 engine looks like a fairly decent D3D7 level engine with fairly decent performance. From what we have seen of DooM3 performance is comparable while D3 utterly demolishes it in terms of visuals. You simply look at that end and OpenGL looks superior however Croteam's engine fails to match up with Sweeney's IMO(not bad by any means- actually pretty d@mn good, just not as good as Sweeney's) which you could say would give a leg up to D3D in that comparison. Then you have Lithtech who has come a long way but still can't really touch the 'big guns'.
When all is said and done both are solid APIs each with their own strengths and weaknesses. I think OpenGL ultimately sees the best games in terms of performance/visuals but only due to Carmack. Not saying it's an inferior API, but I would be very interested to see what he could do on the D3D side of the fence before I would use his engines as a baseline to compare APIs.
