But the biggest handicap on the PS3 is that it has two ISA's = PPE, PowerPC I-III control ISA using in order execution; and an SPE, SPE ISA.
the only way for it to work efficiently is to have two or more control (PPE) cores or for the primary to be able to execute code out of order.
The reason Xenon has been easier to develop for is that it consist of a single parent ISA and there three parents. If one is busy the other can pick up the cue.
Probably #5
If in future generations of Cell they do not have an out of order exe OR Dual core PPE, then they must double the Local storage (256KB to 512KB) so that the SPE can have larger instruction cues to better enable their autonomous function without the aid of the PPE or Level 2 cache.
To answer the OP. For this gen I would have gone with #2. But of course, it is designed the way it is for a reason.
Definitely #1
But the biggest handicap on the PS3 is that it has two ISA's = PPE, PowerPC I-III control ISA using in order execution; and an SPE, SPE ISA. This called a "heterogenous architecture" but the only way for it to work efficiently is to have two or more control (PPE) cores or for the primary to be able to execute code out of order. Due to this not being the case quite a bit of code optimization is needed to utilize SPE independent of PPE controls. The reason Xenon has been easier to develop for is that it consist of a single parent ISA and there three parents. If one is busy the other can pick up the cue.
Probably #5
If in future generations of Cell they do not have an out of order exe OR Dual core PPE, then they must double the Local storage (256KB to 512KB) so that the SPE can have larger instruction cues to better enable their autonomous function without the aid of the PPE or Level 2 cache.
OCing the GPU in most PC games seem to improve frame rate. But I think most PS3 games frame rate problems are CPU related and OCing Cell to 4 GHz won't solve it. It probably requires better memory management scheme and better SPU algo.
How much performance advantage for PS3 if
1. Add more XDR to 512MB
2 Add more GDDR3 to 512MB
3. Speedup Cell clock to 4.0GHz
4. Speed up RSX clock to 625MHz
Which one way better?
So a 512MB Radeon X1600 has a performance advantage over a 256MB X850 ? *sigh*
Console games are designed to run within a memory spec otherwise you would get a fault. If a game is designed to run within 512MB, adding more ram isn't going to give you any performance boost at all. Ergo, 1 and 2 do not help the performance. They would help with texture variety and texture sizes, but then you would want higher memory bandwidth instead, which is a different type of spec.
Increasing the core speed of RSX would help with the shaders, but again, memory bandwidth would become more of a bottleneck for the ROPs, blending etc.