OT Linux (RE: [rescue] OT: Stuffed Proliant?)

[Scott Newell]

>Another problem with Intel (for some tasks) is floating point performance.  
>When Intel quotes benchmarks saying how fast they are, they are almost always
>using single precision.  However, almost everything uses double, which Intel
>is no where near as good at. 

This is interesting.  Is it a memory bandwidth (because you're moving
bigger hunks around), compiler, or FPU issue? I didn't think the FPU itself
was any quicker in single precision mode.  Something else to consider is
that the IA-32 FPU performs ops internally at extended precision (10
bytes), not double.  That can give the Intel chip an advantage if
intermediate results need to be carried out to higher precision.  (Fused
mulitiple add instruction helps too.)

IA-64 has an 82 bit floating point format.  Weird.


newell