| Kyros said: because algorithms that heavily depend on branching are really bad on CELL |
Not if you implement branch hints (more work, but better efficiency). Optimising for the Cell thus has the additional benefit of achieving better cache hits on other CPUs (including the Xenon) as well.
It's important to understand the Cell's SPUs are general purpose processors able to do anything you would do on an ordinary processor core, but they require extra care (only management of the SPEs needs to be done by the PPE). They are really fast at vecor unit, DSP and some GPU related tasks, but they are far more flexible.
For double precision (64-bit) tasks the PS3 Cell is less powerful than for single- (32-bit) and half precision (16-bit) code formats, but research documents still show results that the Cell can outperform the top Quad core x86 CPUs.
