| Final-Fan said: It's better, I think, to just say (when you think someone uninformed will be looking), "You can't directly compare two different types of processor by looking at the GHz, cores, etc. You have to look at performance. That's true of AMD vs. Intel, and extra double true of AMD/Intel vs. Cell." |
Not quite. In fact AMD and Intel architectures have n0ow more incommon.
The difference between the PPU and SPUs is much more fundamental.
The PPU is, like every normal core dedicated for Random Access Tasks. You can't know what will happen next, instead you must be able to switch tasks fast without heavy penalties.
The SPUs are optimized for streaming purposes. They work best without task switches, if they are allowed to only work on new data. That is the real rerason why can sometimes be simplified as a very complex GPU-Pipe.
Sure, you can use it as a normal core, in fact the OS of the PS-3 tries this, but it also proves that you shouldn't do it. It is rather difficult and complesx without any real advantages. It would be pretty ineffective to use the SPUs this way.
But the statement that the Cell was always designed to work closely with a high performance GPU isn't quite right also. Sony was not crazy enough to think that they would be capable to develop a high performance GPU.- NVidia and ATI had much more knowledge. Its companion chip would have been more asimplified GPU. It wouldn't nned very complex pipelines, when the Cell can already do much of the heavy lifting.
But the Companion chip would have interacted much more closely with the Cell directly. It would have probably worked directly on the system memory 6that the Cell uses., so there wouldn't be such a huge bottleneck due to the limited amount of system RAM.
