[OT] End-user uses for x86-64 (was: Why are still not at 64 bits)

Ric Werme ewerme at comcast.net
Fri Feb 16 14:45:20 EST 2007


Jim Kuzdrall wrote:

>    End users of portable processing may benefit from clock rate 
>reduction.  The 64-bit internal and main memory paths double the 
>processor's instruction throughput.  A given 32-bit performance can be 
>had at half the clock rate in a 64-bit processor.  (Almost.  There are 
>other factors.)

Well, there are aa lot of factors.  Early Alpha code was 30% bigger than
32-bit MIPS code, some due to with 64 bit pointers, some due to with
instruction scheduling (well placed NOPs can make Alpha faster), and some due
to with the young compilers.  Also, just because you can write an instruction
that accesses 64 bits of data doesn't mean that's what memory sees.  Alpha and
other chips read a cache line worth at a time, 64 _bytes_ on I think all
implementations of Alpha.  That helps keep the cache control logic small and
fast.

If people are doing 32 bit math, a 64 bit ALU won't provide 2X the
performance.  You can add more ALUs, barrel shifters, etc, but those are
power hungry beasts.

>    The internally consumed power is proportional to the clock rate for 
>CMOS technology.  Laptops and handhelds which are limited by battery 
>life or heat removal may find the additional PC board area and cost 
>justified.

Don't forget backlight, disk, and other substantial power consumers!

      -Ric Werme


More information about the gnhlug-discuss mailing list