Could a cell-like design be implemented with x64 and SSE?
One basic x64 core with several SSE subunits connected by high bandwidth, the advantage over cell is complete x86 compatibility?
Could PS3 cell-like design be implemented with x64 and SSE?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
It's not really necessary considering you'll be seeing eight FULL FEATURED cores in 2008. Four full featured cores in a native quad core design will probably outperform a Cell in most tests anyways. You'll be able to buy something like that in a few months from AMD.
The most impressive part of the Cell, IMO, is that it uses 40 watts of power. AMD's 4 core design will use 125 watts. At least that's what's being reported now. If true, that's really really pathetic on AMD's part.
The most impressive part of the Cell, IMO, is that it uses 40 watts of power. AMD's 4 core design will use 125 watts. At least that's what's being reported now. If true, that's really really pathetic on AMD's part.
They are going to use the graphics pipelines like floating point processors when they are not in use by the graphics.
Is this the same as stream processing?
Yeah. That is my take on it at least. Sony was trying to make the Cell work like a graphics processor. It has 1 PowerPC core and 8 floating point processors. When they realized that they could not get the cell to perform graphics after a year of trying they hastily added a graphics processor to the Playstation.
The Fusion is going to be able to implement the graphics processor into the core which leaves available the stream processors on the die to do floating point work when Vista is not using them. On the notebook version, the graphics processor in the fusion will probably be doing mostly 3D, but on the desktop version, those stream processors are left free to what ever you want because the external graphics will be doing the 3D load. The stream processors, while probably not as fast as a low end add on graphics, will be free to do physics, or video decoding, or even help with the 3D load.
While Intel is trying to pack more and more regular pc cores into a single die or package, AMD thinks 2 or 4 is enough and wants to add stream processing to the cores. We already see diminishing returns with more processors so I think the stream approach is better. Lets hope it can be implemented without having to change the way programs are threaded. That would be the ultimate solution. If not, AMD might be betting on a losing strategy.
This is my take on the news and is purely speculation based on a few news stories and a basic understanding of stream processing. This is why I am long on AMD and short on Intel. Intel has the momentum and the process technology, but AMD's server and graphics implementation place them in the postion to lead the industry in performance applications. I am praying that they get past this hump of poor financial performance and can break away from the pricing pressures that Intel has placed on them. With a dependance on 90nm production and intel moving towards 45nm, AMD is in trouble if they cannot move to 65nm very quickly.
The Fusion is going to be able to implement the graphics processor into the core which leaves available the stream processors on the die to do floating point work when Vista is not using them. On the notebook version, the graphics processor in the fusion will probably be doing mostly 3D, but on the desktop version, those stream processors are left free to what ever you want because the external graphics will be doing the 3D load. The stream processors, while probably not as fast as a low end add on graphics, will be free to do physics, or video decoding, or even help with the 3D load.
While Intel is trying to pack more and more regular pc cores into a single die or package, AMD thinks 2 or 4 is enough and wants to add stream processing to the cores. We already see diminishing returns with more processors so I think the stream approach is better. Lets hope it can be implemented without having to change the way programs are threaded. That would be the ultimate solution. If not, AMD might be betting on a losing strategy.
This is my take on the news and is purely speculation based on a few news stories and a basic understanding of stream processing. This is why I am long on AMD and short on Intel. Intel has the momentum and the process technology, but AMD's server and graphics implementation place them in the postion to lead the industry in performance applications. I am praying that they get past this hump of poor financial performance and can break away from the pricing pressures that Intel has placed on them. With a dependance on 90nm production and intel moving towards 45nm, AMD is in trouble if they cannot move to 65nm very quickly.