Here is a theoretical question about a potential unified shader architecture and CELL.
If you had a lot of SPE's in CELL assigned to vertex work, wouldn't the memory pool that the USA GPU is connected to, be a lot better off if it was microthreaded like Rambus XDR2? I'm talking a lot of CELL's power dedecated to geometry.
If you had a lot of SPE's in CELL assigned to vertex work, wouldn't the memory pool that the USA GPU is connected to, be a lot better off if it was microthreaded like Rambus XDR2? I'm talking a lot of CELL's power dedecated to geometry.