function said:
Anyway, just a general thought going round my head is this: if the PS3's CPU were to be several times more powerful than Xenon's CPU, where would this power come form?
Since neither the details of the Xenon or the PS3 is out in the wild, this is an impossible question to answer authoratively.
But there are two areas that can make large differences.
* The memory subsystem/hierarchy.
* Computational resource balancing. (Catch all bullet point.)
Augmented by a trickier third
* Internal communication overhead.
OK then, point by point, the PS3 has elected to go with a memory type that has advantages and problems. The console environment however emphasizes the benefits, which is very high potential bandwidth, (while avoiding the problems incurred when you try to package this memory into DIMMs of arbitrary size and in arbitrary numbers).
Depending on implementation details, this could give the PS3 a huge bandwidth advantage.
As far as computational resource balancing goes, the PS3 Cell processor is a vector floating point monster. Comparing it for a moment with an x86 processor (or the PPC 970) points towards a PE that is less sophisticated in terms of extracting maximum single thread performance as far as superscalarity and particularly OOO execution goes. However, it conversely has easily an order of magnitude greater resources for vector floating point math. And that's per PE.
There was no way in hell any x86 processor could come close to that, without ditching SSE for something better, and then multiplying it, adding control logic for these vector processors, trying to squeeze all this onto a die, and then writing new tools for accessing the new capabilities since existing x86 tools wouldn't be useful, invalidating any rational reason to stick with x86. So Microsoft did the reasonable thing and looked for something better geared towards the needs of this kind of product. We still don't know what IBM offered them, only that it
was sufficiently better that it made sense to abandon x86. It may well be that the PS3 Cell is still far more capable in terms of vector math, but that this will be somewhat countered by the Xenon GPU taking over some of the tasks that the Cell APUs will handle, and that offloading this to the GPU will allow other types of code (AI and other typically branch heavy stuff) to be better handled by the Xenon CPU.
The third point is about how hard it is to apply parallell execution resources to a particular problem. This depends strongly on both software tools, and basic hardware capabilities. Latencies for accessing data at various levels and locations. About this, we know next to nothing. It would make sense to assume that the Cell processor is exceedingly good at this, since parallell execution is a fundamental idea behind the Cell concept. Implementation is everything though. It could be speculated that the Xenon CPU will be best harnessed using rather coarsely parallell codes, something that might or might not map well to game codes.
Those are some issues that could make a big difference. But even knowing a lot more details than we do, it would be very very difficult to go from schematic description to real world performance projection. The proof of these puddings will be in the eating.
And perhaps both of these consoles will be sidestepped in the marketplace by a small, cheap, and quiet powermiser of a Nintendo designed and prized such that people other than the tech geeks would like to have it in their living room.
