Okay... since the pricecut we've got here a lot of threads about how PS3 graphics suck and X360 being the unmatched graphical standard of console gaming. However, I'm not going to say it is all trolling - there is some reason behind it. Even PS3 matching X360 texture-wise while using full RGB mode, it still got a worse overall graphics pack mainly due to major graphical issues that not even happened last gen.
I keep saying again and again that these issues are actually the side effect of a GPU architecture different of the nowadays PC gaming standard and X360. If you are guessing about the unified pipeline architecture, you guessed right. PS3's RSX uses a outdated - yes, it is - graphical architecture based over high-end GeForce 7 GPUs. As many of you may know, Nvidia just started using unified pipeline architecture in the GeForce 8 series.
The point is, some argue that these graphical issues in PS3 are in fact the effect of a weaker and outdated GPU, dispite the fact PS3 released a year after the X360. But, the stone-cold facts actually back-up that? The only single answer to that is: of course, no.
Now to the facts:
1- basic specs:
The Xenos is an ATI GPU built nowadays at 65nm process and its core clock is 500 MHz.
The RSX is an Nvidia GPU built (as far as I know) at 90nm process and its core clock is 550 MHz.
2- bandwith:
The Xenos has a total of 337 million transitors and dedicated 10 MiB memory embedded on the actual 65nm process. It can use the 700 MHz 512 MB unified main memory of the X360 as video memory.
The RSX has a total of +300 million transistors (the exact number isn't known) and uses a 700 MHz 256 MB memory. Also, it can use a max of 228 MB from the 3200 MHz XDR RAM.
Of course, these numbers are theoretical since both GPUs aren't going to use all the main memory.
3- architecture:
The Xenos uses a modern unified pipeline architecture. It has a total of 48 pipelines each one producing 10 FLOPS per cycle. Support for a superset of DirectX 9.0c API DirectX Xbox 360, and Shader Model 3.0. Plus, additional logic (192 parallel pixel processors) for color, alpha compositing, Z/stencil buffering, and anti-aliasing called “Intelligent Memory” (eDRAM) makes anti-aliasing an easy work.
The RSX uses the NV47 pipeline architecture who is translated by 24 pixel-shader pipes plus 8 parallel vertex pipelines. The former produces astounding 27 FLOPS per cycle while the later, 10 FLOPS per cycle. Support for OpenGL ES 2.0 and S3TC texture compression.
4- render output:
The Xenos renders up to 4 gigapixels per second (its 8 ROPs x 500 MHz). It's texel fillrate is 8 gigatexels per second (16 texture filtering units x 500 MHz, a max of 16 unfiltered texture samples per clock). It's maximum DOT product operations are 24 billion per second. Xenos' maximum anti-aliasing sample rate is amazing 16 gigasamples per second (4 AA samples x 8 ROPs x 500 MHz).
The RSX renders up to 4.4 gigapixels per second (again, 8 ROPs x 550 MHz). It's texel fillrate is 13.2 gigatexels per second (24 texture filtering units x 550 MHz, a max of 32 unfiltered texture samples per clock). It's maximum DOT product operations are 54 billion per second. RSX's maximum anti-aliasing sample rate is 8.8 gigasamples per second (2 AA samples x 8 ROPs x 550 MHz).
5- single floating-point performance:
Xenos: 240 GFLOPS (10 FLOPS x 48 unified pipelines x 500 MHz)
RSX: 400 GFLOPs (24 pixel shader pipelines *27 FLOPS + 8 parallel vertex pipelines *10 FLOPS *550 MHz)
Conclusion:
In a nutshell, while RSX beats the hell out of Xenos about rendering data and raw processing performance, Xenos actually is far easier to work with due to its modern architecture, developers having 4-sample anti-aliasing at very little performance cost, DirectX 9 and Shader Model 3.0.
Essay Pro