PC Watch (my translation): PS3 Evaluation System, and much,

I think PS2's design philosophy was a bit different. Back in those console days the CPU still did the TnL. Also the GS with all that eDRAM was already a big chip so it would've been inmpossible to add on a TnL unit that's why in the end you had the VUs on the EE.

 

Tons of questions though . How many triangles are the spe's going to output ?

Did they make the right choice ? HOwever I highly doubt they juts chopped out the vertex shaders .

The gs was being designed in 1996-1998 , the first tnl gpu came out in 1999 and i'm sure that for 1999 sony made a good choice as the tnl gpus of the time weren't all that advanced .However there are alot of other features missing out of the graphics card that evne the tnt 1 had .

I"m sure in the end the programers and developers did things with the cpu that sony never thought of . But they also had to spend alot of time reinventing the wheel so to speak .

 

Nowadays, the pixel shaders do most of the "L" part of TnL. Per-pixel lighting is replacing per-vertex. With topology processing and more complex shading algorithms coming into play, the traditional VS isn't flexible enough anyway.

 

That was my question too. Though i'm pretty sure they'll fare quite well. My worry is the bandwidth between Cell and RSX, which is bound to be much lower than the bandwidth you'd get if the geometry was calculated all inside the RSX.

I do have a feeling that keeping all dynamic geometry of a scene/frame on Cell, the geometry can interact better with animation and physics, being all on the CPU.

 

To clarify, they haven't, if RSX is the same config as G70, of course. There'd be 8 vertex shaders there. But when you look at how much processing their doing, if you look at Gflops, about 85% of the programmable power in RSX is with the pixel shaders.

I'd venture to guess that for most games, the VS in RSX would be enough. What I was addressing was the rarer situation where more would be needed.

Geometry that changes from frame to frame, due to collision detection etc. has to be passed across from CPU to GPU anyway..(and geometry has to be passed back from the GPU too for collision detection/physics).

 

http://ps3.ign.com/articles/636/636848p3.html

good summary