Kan ook hoor
ik las dit:
Some features of the Opteron are available with a 32bit OS:-
1MB Cache, SEE2, onboard mem controller, and minor other improvements.
The 64bit features (64bit ALU, PC, extra/wide GPR's, extra XMM Registers) are only available with a 64bit OS and 64bit compiled program.
The FPU is not altered. The FP units have been using 80bit registers (originally register stack) and 64bit floats since the 486DX or even earlier.
The 64bit ALU's will make it easier to handle 64bit integers but make no difference if the value is a 32bit INT. In fact, a 64bit ALU will operate slower then a 32bit ALU based on the same basic design (building blocks, IC process etc).
Opteron 64bit @ 1.8GHz = 43.2GBytes of 64bit INT adds per second (theoretical)
Opteron 32bit @ 1.8GHz = 21.6GBytes of 32bit INT adds per second (theoretical)
Athlon 32bit @ 2.2GHz = 26.4GBytes of 32bit INT adds per second (theoretical)
(oops, I used 1GB = 1x10^9)
If the 64bit ALU was double pumped for 32bit you could have two 32bit additions in the same time as one 64bit addition (with slight overhead).
The SEE2 and extra XMM registers will improve the performance of SEE2 optimised code.
En maakte toen misschien wat foute veronderstellingen