Buggy Loop said:
take you months after months for the same effect. That doesnt mean ps2 is underpowered.
If tecmo couldnt pull out more out of ps2 with their DOA3 version, its their problem, shouldnt take a back stab at sony cause they didnt work enough for it.
If you read the interview, he makes the statement that PS2 is also easy to program for, "at least for Team Ninja". He takes the stance that PS2 developers are spending too much time complaining about the difficulty of programming PS2 and not enough time developing on PS2.
He also says that though xbox is fairly easy to develop for, it does get more difficult in the later stages of game development, or something to that effect.
I haven't done much DX 8 D3D programming yet, but I played around with DX7's direct3d, and I do remember (this is back in NV10 days) actually making good use of the geometry accelleration required jumping through some hoops - write only indexed vertex buffers, allocated in video local memory, arranged in strips or fans. I also remember being fairly suprised that when I traced into the D3D matrix multiply functions that they wern't 3dnow or sse optimized at all, it looked like regular fpu floating point, which seemed odd to me.
Developers of course generally either write their own matrix multiply and dot product functions but if it's the case that xbox's DX8 doesn't make use of SSE and developers are blindly making use of non SSE optimized functions that there is a performance hit. Remember that even in the case that there is onboard hardware geometry accelleration, the conjugate matricies (rotation, transformation, scaleing, etc) must still be multiplied outside of the hardware geometry accelleration, in the case that you have many small particles or a high number of low-polygon count models this is a signifigant performance hit. I remember getting the best performance out of NV10 with a small number of very large models (65535 verticies), because of course you only have to do minimal matrix math on the CPU, and the hardware transform handles the matrix * vector multiplies for you.
I am aware that DX8 has changed things and I look forward to playing around with it, but the point is there are still performance pitfalls on xbox that developers will (I'm sure) watch out for.