Why not setup one core heavily interger optimized, and one floating point? That way you can run the FP apps like rockets, and the interger apps like lightning w/o comprimizing on either. Rather than have a long chain in the pipeline you could have paralell paths, and once an instruction is set down one path, the CPU could take the next and see if it can stick it down another path.
Err... you mean like the 8086/8087 or 80186/80187 or 80286/80287 or... I'll shut up now.
Slashdot Mirror
Why not setup one core heavily interger optimized, and one floating point? That way you can run the FP apps like rockets, and the interger apps like lightning w/o comprimizing on either. Rather than have a long chain in the pipeline you could have paralell paths, and once an instruction is set down one path, the CPU could take the next and see if it can stick it down another path.
... you mean like the 8086/8087 or 80186/80187 or 80286/80287 or ... I'll shut up now.
Err