Gholbine, I'm going to assume that you have no coding experience and little real understanding of how the 3D pipeline works, because a lot of your statements made no sense.
Gholbine said:
Cell absolutely will help on the graphics front, just as the Emotion Engine did in the PS2. The system is built that way, with heavy CPU<=>GPU communication in mind. It was the entire philosophy behind the PS2, and it remains so with the PS3.
RSX is based heavily on the G70, and hence PS3 will follow the PC paradigm of 3D rendering much more closely. The main use of the CPU-GPU communication will be for fast texture loading and vertex transfer (as with AGP and PCIe on PC's), nothing more. There will be little point in doing any full-fledged vertex transformation on the CPU, because the VS on GPU's are very finely tuned to have fast, latency free vertex processing with quick rejection for backfacing and clipped triangles as soon as position data is complete.
Like I said, physics and maybe animation are the only places that Cell will indirectly impact graphics. Even then, Cell will be used to figure out bone transformations and transformation matrices, not move each vertex. As for actual rendering, general purpose CPU's, even Cell, are one or more orders of magnitude slower than graphics hardware. Furthermore, current graphics processors can do pretty much anything you'd want due to their programmability via shaders.
Even in current PCs, where CPU and GPU communication is nowhere near as important, the CPU can still help with detailed graphics. Take Doom 3 for example, the shadowing was done entirely within the CPU. The Cell could handle shadows, among other things, leaving cycles on the GPU for other activities.
Detailed graphics? Do you have any idea what the CPU did in Doom 3, and why ID did it that way?
The actual shadows, are all calculated and rendered entirely on the GPU.
The CPU is used to find silhouette edges on objects, and insert geometry there. This can only be done after transformation, though, so Carmack had to do all vertex transformation, including skinning, on the CPU because there's no way for the GPU to make its vertices accessible to the CPU after going through the vertex shader. This isn't a problem on these consoles. The other reason Carmack did it this way is that older hardware would need to multipass the lighting shader he uses, and the Z-only pass also needs to be done on all hardware. Using the CPU to transform everything only once meant no redundant work had to be done. Again, this is a non-issue for consoles.
Even so, you can do this all on the GPU if your vertex shaders are fast enough, and this is exactly what 3DMark2003 did. This is actually a better method, because GPUs always scale faster than CPUs. I'm sure that today, Doom3 would run faster if Carmack did everything via the GPU.
I think it's rather naive to say "Cell has nothing to do with graphics", because that's incorrect, plain and simple.
No it isn't, plain and simple. If you haven't done significant graphics coding for games, you really shouldn't be arguing this with me, because you don't have a clue what you're talking about.
Your lacking knowledge in graphics shows again when you talk about disk space. Resolution has very little to do with how much space you need. The only correlation is that with a higher resolution screen, you might want higher resolution textures. But even that is a very tenous relationship. For the past several years (I'd say starting with the GeForce2), many PC gamers have been playing at 1600x1200, which has twice the pixel count of 720p.
It has mostly been graphics memory that's been the cap on texture resolution, and the PS3 only has 256MB (using system memory would be a bad idea). It has framebuffer bandwidth comparable to 3 year old graphics cards (i.e. the 9700Pro), so I don't see it having the capacity for the huge texture detail jump you're implying.
Its very doubtful that XB360 (or PS3 for that matter) won't be able to handle compressed data like a PC does. The PS2 was probably asking for as much help as it could get because this game had demanding data requirements for a tiny amount of video ram.
Audio will never take up that much room, especially with the great compressors we have nowadays. 7.1 sound is put to use by the spacial location of sound sources in a game. Not every single sound sample will have 8 channels. The vast majority will have 1 or 2, and they'll come out of the correct speakers. Even 600MB will give you 20 hrs of unique sound, and games re-use sound a lot. Speech engines will get better and drastically reduce storage requirements.
Regarding my statement of impressive realtime video, you're taking it out of context. How much HD FMV will you be able to put on a BR disc, maybe 2-3 hours? If you have 20 hours of realtime FMV that would be much more impressive in my opinion, because it gives so much more variety, and enhances the replay factor of a game. I don't see load time being an issue because you'll probably use the same textures and model for cutscenes as the game itself.
Comparing to N64 is nonsense. First of all, it was a 20-1 ratio for a long time (32MB cartridges). Secondly, there are diminishing returns to more capacity. There was a drastic difference in texture detail between PS1 and N64. You'll never see anything near that in PS3 vs. XB360, unless someone does a really lazy port. The lack of urgency for disk space is readily apparent in the PC space, where DVD adoption for game distribution has been very slow despite the ubiquity of DVD drives.
Anyway, I don't think I'll ever convince you that this is a minor problem. PS3 developers that fill up Blu-Ray discs will do so because they can, not because they need to or because it'll make a much better game. They just won't worry about packing stuff together, and MS will probably provide lots of tools to make it easier anyway.
