Tim Sweeney comments about PS3 architecture

[Procrastinato]

Kynes said:

Procrastinato said:

Soleron said:

Cypher1980 said:
Can the Cell be used to render ?

Why doesnt the PS3 make more use of this function and bring up graphical performance ?

Because it isn't very fast at that compared to the GPU.

That's entirely dependant on what portion of the graphics pipeline you're talking about.  The Cell is awe-inspiring at vertex ops.  It just can't be used to do pixel/fragment ops efficiently.  The more vertex ops you offload onto the Cell, the more breathing room the pixel pipelines have during the frame to do their work as well.

That isn't completely true, because the RSX is a non unified architecture, so the number of pixel pipelines is independent of the use or not of SPUs for vertex rendering. You can generate models of a higher vertex complexity, but the pixel fillrate is where the bottleneck is in PS3's case, SPUs used or not. If RSX were an unified architecture, the gains in performance could be much higher, because you could almost use all the power to do pixel work, but it isn't the case.

let me put it this way... Do you think the RSX spends any time waiting on its 8 vert pipes, since its basically doing no skinning, no morph targets, no vertex anims, no nothing, other than projecting the vertex into screen space?  I might add that say.. doubling the number of verts in a model goes a lot further towards making it look good (esp. with good lighting) than adding some minute fraction of extra texture resolution does.

If you add skinned meshes to a game, and have a decent vertex count, are doing facial anims, etc... don't you think you might want to devote a fair number of the Xenos' pipelines to doing that (vertex) work, or are you claiming the "awesome" shared-cache-resource Xenon CPU can cover such work efficiently?