Quote:

---

The Cell will not, and should not be performing all rendering operations like the E3 2005 demos displayed. It should prove as very interesting that the Cell does perform well at those types of operations since rendering on a CPU offers more flexibility than vertex and pixel shader programs. It is unlikely the Cell would be processing the latter type of shader operation since it would involve the RSX processing an almost finished frame, before giving it up to the Cell, only for the Cell to send it back to continue down the graphics pipeline again with almost no work to be done.

Granted, 3D pipelines are configurable and you can speed up processing through it by disabling unnecessary features that you might have already accomplished on the CPU already. It is likely that developers will do some basic/macro level 3D operations on geometry before passing it off to the RSX to do more time consuming fine detailed processing.

The Xbox360 CPU could do the same thing too and aid in rendering task, but general purpose computing power doesn?t exactly lend itself well to the types of operations it would have to perform, and the vector processing capabilities of the Cell greatly out perform the Xenon in this respect.[/quote.]

Quote:

---

A couple of points that stick with me.

1) He overlooks the effort of software development. It's all very well saying 'create new techniques to fit with the Cell architecture' and is a philosophy I have applied to support Cell, but that still means adding cost and time to development, meaning for all Cell?s potential, much of it may go untapped most of the time.

2) He hasn't understood Unified Shaders at all, their advantages, and he swiftly fobbed them off saying 'ATi aren't using them in their latest GPU so they're not the way to go.' He's wrong on many points! In a few years I wouldn't be surprised if all GPUs are US!

3) In breaking down system bandwidth, he's right in his analysis of the system's useful totals, but he has totally overlooked the impact of framebuffer activities. Not a single mention was made, and he speaks as though RSX's 22 GB/s GDDR is just going to be used for textures and models. When you consider that the GPU needs lots of memory bandwidth in the rendering of pixels, especially with antialiasing, you see RSX is quite limited, especially compared to PCs. And in Xenos, much of that requirement is nullified by the eDRAM. Where RSX has 22 GB/s for textures, models and framebuffer, Xenos has some 21 GB/s from RAM + 256 GB/s for the same purposes. Xenos will never be bandwidth starved for framebuffer effects, just like PS2.

There were other little niggles too, like saying BRD's transfer speed is an improvement over PS2s, but not considering the RAM capacity has increased moreso. If the drive speed is increased 2x, and the memory increased 8x, your load times will go up by 4x.

As a reference for Cell's architecture, I think it's pretty good. As a fair comparison with XBox360, it's quite poor. It raises some fair points comparing to PC, and IMO doesn't give a particularly balanced view of PS3 specs as a whole. There are some aspects not so good that Ebony hasn't touched upon.

All in all, it's a shame people write these sorts of things to prove a point (PS3 is best) rather than to educate (here's a fair breakdown of PS3, XB360 and PC architectures with pro's and con's to each)

---