Seems Jesper has caught Kebabbert out
Quoting biased benchmarks will never convince anyone - reading both the Sun reference from Kebabbert and the paper it originated from that Jesper quotes I could write my own blog post at IBM central stating one CELL BE with 1 core beats Niagra with 8 cores in this very important area of pattern matching ... blah blah blah
Not quite true as CELL is only better in small dictionaries, it has SPEs which although not cores like the PPE should obviously be mentioned. Sun ref. states CELL has 8 cores - which is as untrue as my statement, one PPE which is a cut down Power core and 8 SPEs which are by no means general purpose CPU cores and as such should not be compared to the Niagras 8. (Sun thinks the 1 PPE and 8 SPEs adds up to 8 - I assume its because only the SPEs were 'doing' the work in the academic article, in fact the PPE orchestrates everything so maybe should be included to make the CELL look worse)
Having read both I'd have to conclude the Sun article is a peice of marketing FUD, they specifically chose the particular area of this benchmark that the CELL is bad at and ignored the bit where it is a lot better. They exaggerated the number of cores in the CELL which indicates a lack of understanding of the CELL, especially as they compared an SPE to a Niagra core. Worst thing is using the academic article referred to by Jesper and not having a reference so deluded Sun fanbois like Kebab never see the full story....
All fairly irrelevant any way - I work in the real world of performance (For IBM so I must be biased, actually a lot of IBMs customers use CPUs other than Power!) tuning customer systems and the CPU technology is rarely of interest (I don't get sales commission ;-)). Real applications perform badly because of bad code - order of magnitude performance improvements are not uncommon by fixing the poor code in the app. Architectural decisions, library decisions (I choose JSF for my high volume website!) etc have far far more effect on performance. Except for that customer who decided to run everything on a 486... But thats a different story.
The rare optimised workloads that run on specialised processors like CELL, or the highly optimised cluster apps are rare and they probably do care more about the CPU tech specs - I'd guess they don't take BM Seer as a credible source. Interestingly IBM has far more workloads in the top 500 supercomputers than anyone else - so maybe they are good at that, I don't know cos I don't work in that area.