Originally Posted by
cadaveca
Um, think what you want, but you'll never stray my belief of the truth over your words.
Anyway, it does relate. Work in 3D my butt!
Um, notice quake4 ran poorly, and development was started back when the fisrt link i posted took place(summer 2005). You see, I pay attention to 3D.
Now, megatexturing runs on a variety of platforms, and is not really API specific, because it does not simply deal with just large texture overlays, really, it deals with virtual video buffers, and taking those 4GB textures for maps and leaving 8MB of virtualization running in local vidram.
Now, video memory virtualization is one of the major advancements brougth by DX10, and it's one of the key components that makes deferred rendering possible in DX10, when it's not in DX9, due to the performance hit.
Please be advised tho, that performacne has nothing to do with a gpu's rendering power, and everything to do with the bendwidth of it's framebuffer, and the bus the card sits on. We wouldn't need Megatexturing, Deferred Rendering, OR Video Virtual Memory if we had a bus like Xenos's prorietary link to Xenon.
By the way, those three "technologies" are all things that R600 excels at, and G80 does not, and this is why we can see a very large performance gap in DX10 applications between G80 and R600 currently. BTW, Megatexturing has ALOT in common with Tessellation.
Anyway, tesselation is alot like ShaderAA, but in the opposite. ShaderAA simply draws @ a higher resolution to provide the many samples needed for MSAA(rather than dedicated hardware interpolating it)and Tesselation is just like Megatexturing, with a wee bit of deffered rendering tossed in.
BTW, DilTech, Quake4 for Xenos is not a "wrapper" port. The problem is that OpenGL at the time did not support unified shaders, so Carmack was left with no choice but to almost re-write the entire engine. But he kinda faked it, and we only get 30FPS on Xenos and iDTech4.
None of these features work well in DX9, in such a way, we might as well call them unsupported, as few applications use them(STALKER uses deferred rendering), but in DX10, they are more of a "common commodity" among the main 3D engine developers.
If we look back @ titles like FEAR, Quake4, and Unreal2.5, almost 80% of ALU power goes unused, most often due to bandwidth limitations imposed by gpu's memory buffers. Enter ShaderAA to make use of it, provided the bandwidth is still available. And, fortuantely for ATI, and thier advanced scheduler, in most isntances it is, but only in limited amount(bye-bye 6xMSAA, hello tent and box filter).
This all ties together, as when the other "technologies" that I mentioned earlier are provided the power to run together, thanks to DX10, shaderAA becomes more and more feasible, as framebuffer bandwidth requirements go down. Of course, we need a real DX10 app to take advantage of this, and sadly enough, not even 5 xbox360 titles make proper use of Xenos's unified shaders, being designed for other platforms, so it seem a long time coming for anything to hit the PC front in full force!
But they are coming, soon enough, I suppose.
Bookmarks