| MikeB said: @ selnor Well you can argue all you want with IBM. They are the figures and in real world terms the Cell as a whole gets nowhere near 200GFLOPS inside a PS3 for game usage. You can argue with IBM all you want. I have spoken to several IBM technical specialists, their perspective is the Xenon's theorectical peak is 76.8 GFLops (the equivalent of about 3 Cell PPUs) and the PS3's Cell theoretical peak is 218 GFlops, this while the Cell inside the PS3 is able to achieve much higher real world efficiency and the SPEs are much better suited at various tasks. For example in 2005 Mikael Haglund, technical specialist of IBM Sweden gave some Cell presentations at the AmiGBG fair in Sweden, which many DemoScene, game developers, OS and application developers attending. Our team did some life coverage, interviews, I wrote a show report and we had our own stand at the event. |
I have seen it all now. You argue that IBM's own website provides the cells Theoretical peak performance and you dont believe it? You seem miss informed to. 218 GFLOPS is the speculative theoretical performance of the cell. To get this number they test in a static environment 1 SPE on it's own. to get 25.12 GFLOPS. They than X that by 8. That is not theoretical it's speculative thoretical. There is a difference.
Now Theoretical performance of the Cell testing is done as follows. The entire Cell is tested in the same environment. So thats all 8 SPE's and PPE (as SPES cannot run without the PPE it is full theoretical Cell performance). So testing the Cell in the same environment provides us with 155 GFLOPS according to IBM themselves. Sony chose to take the speculative theoretical performance of which the numbers come from a single SPE test at 25.12 GFLOPS.
Now again, 2 SPE's in the PS3's Cell are NEVER used for games. That is fact. So that leaves 6 SPE's for games. So immediately 155.5GFLOPS in the PS3 Cell is not available for game programmers. Now the Xenon unfortunately the Theoretical performance peak numbers are not available from IBM's website. But many Tech sites claim that the theoretical Performance is 115 GFLOPS tested as the whole CPU (rather than testing on thread and adding them up).
We know as fact that only 3% of the xenon is used for OS so the rest is available to Programmers for games.
With this information, many people do not choose to show this info and continue to use Sony's claims (which is based off Speculative Theoretical performance).
Now go ahead and argue with IBM's official Theoretical performance of the CEll in PS3 if you want. But it makes you look extremely silly.
