Dude, he doesn't have a clue "in DirectX2 we were able to jump for the first time".. seriously I don't know anyone apart from developers who actually ever experienced DirectX1 or DirectX2.
What was originally packaged with Windows 95 was DirectX3, and for Windows 3.11 you had Win32GS + DirectX1; while DX3 lasted for about a year after that things very rapidly moved to DirectX5.
The leap between DirectX9 and DirectX10 honestly isn't anywhere near as great as he makes out either. I mean yes, you can do more in a scene past that mark you'd usually max out DirectX9; although to really show much of a difference in graphics you have to push these cards far far further than the new pipeline allows.
Yes, you have a pure programmable pipeline and virtualized memory.. however guess what so does DirectX9.0c, just because developers generally don't use something doesn't mean it wasn't present.
The real addition over Dx9 is the fact that there are no fallbacks, along with the much lower render loops. Although those render loops aren't cost free, in-fact they cost slightly more than Dx9 because the API is now controlling an aspect of development that previous a programmer would. That is sorting out all the geometry so those with identical states and shaders are rendered in the same loop rather than basically a per-object loop or basic structuring of this.
The Geometry shaders that allow you to expand and the complete shader interaction you now gain are really what are far more impressive and have led to more performance in identical tasks to DX9.
Shader 3 allows for branching code (to allow for Shadows to interact with lights and such) this was the first steps towards a more pure shader pipeline allowing the graphics to be programmed almost entirely seperate to the main application (which they should be given shaders are direct graphics control, not pre-calculated tasks that the fixed function pipeline provided) ... but I think people are expecting too much here.
Sure creating Volumetric materials, with soft particles and shadowing was fairly difficult to do at a reasonable speed in Shader 3; and the new branching/predicition abilities of Shader 4 code that now works quite literally like a C program is more down to the new architecture of the gpus not the Shader technology itself.
The X1K-Series Radeons for example had this a generation ago, and this is why on the 360 that has been optimised to take advantage of this technology you can see many of the attributes that developers say "are impossible in DirectX9" are appearing in 360 games.
Viva Pinata, Call of Duty 4, etc.. all use technology optimised that on the PC either requires a far more powerful graphics platform, or the DirectX10 api to achieve the same results.
While it might seem on the surface that this is a giant development leap, from DirectX9 to 10. The reality is a bit more stark, in that really the main thing that Microsoft have done is finally redevelop the API from the ground up in order to take better advantage of current technology. Yes, it appears that DirectX10 games are more beautiful.. but with the exception of the Geometry shader that allows you to manipulate geometry within a real-time environment; Shader 4 offers no IQ enhancements over Shader 3, which in turn offered no IQ enhancements over Shader 2.
We've moved 2 Shader generations ahead, and what we're seeing now after a focus of 2-3years of Shader development is the interaction of the shader code as opposed to expanding what the shaders can physically do.
Shader 2 can output what Shader 4 graphically can do with it's Vertex and Pixel aspects. It can't however branch the code enough to provide every shader to interact with the others to give that believeable atmosphere... Shader 3 can do this, however you still have to run Shaders as a single through loop; rather than being able to back-track.
For programmers, yes this is an extremely leap in enhancing what we can do with shaders. However this isn't really tied to DirectX10 itself more to the Shader 4 specification.
While yes DirectX10 is a great move forward, it unfortunately is one of the smallest leaps forward in technology since DirectX6->7
Honestly I don't see them really topping that anytime soon either.
They could've easily made a new version of Dx9 compatible with Shader 4, the reason they haven't is to push their Dx10 API so they can have a fresh start. Unfortunately it's hit the market too soon with far too little support from hardware, and software.
If they'd released it this month and made a big deal out of it now, it would've been able to slip on the market as seemlessly as DirectX9 did despite it's forceable upgrading to use it.
I mean something that right now is killing many developers faith in it is the fact that something in Vista or WDDM is still killing graphics card performance. Until that is fixed, DirectX10 for developer is going to marred and gamers are going to keep Vista at arms length.
