Why are still not at 64 bits

Jon 'maddog' Hall maddog at li.org
Thu Feb 15 13:04:40 EST 2007


> 
>   Correction: Windows 2003 R2 x64 supports a full 64-bit address
> space, and I'm pretty sure Win XP Pro x64 does as well.  See my other
> message in this thread about how support for those sucks, though.

Right.  I think you actually made my case, didn't you?  Microsoft did
not support 64-bit virtual address space until Vista.

>   What would AMD64 (or even the Alpha's feature set) do for the
> typical end-user?  I'm talking about the people browsing the web and
> writing email and downloading music and looking at porn.  These people
> aren't doing 6-way SQL JOIN's or loading the entire US phone book into
> RAM.  Their PC is plenty fast enough, so long as you clean out all the
> adware and viruses and other badware.
> 
I agree with what you have said, but unfortunately the acceptance on
non-acceptance of 64-bits depends on all of the things that we have
stated here.  In stages. Stages sometimes take very long to happen, and
depend on lots of legacy infrastructure being retired, customer demand,
etc.  And I stand by my assertion that 64-bit programs will not become
prevalent until Microsoft give 64-bits to everyone, not just "Win XP
PROs" who happen to buy 64-bit capable machines.

Other things also affect it, such as the cost of RAM.  The 64-bit
Translation Look-aside Buffers (wow, it has been a long time since I
thought about those) for the Alpha were bigger than most main memories
of desktops in the 1980s.  And that 64 KB memory cost 100K dollars in
1968.  Even in the last versions of the Alpha they did not implement the
entire 64-bit address space, since there was no one who could afford to
buy that much disk, much less that much RAM.*

The Alpha was a RISC processor, and so could afford the trade-off of
address bits and cache for micro-code store on the die.

The Alpha was a space heater.  It took no bones about sucking down
electricity.

Lots of economics also go into how fast 64-bits comes into the
marketplace.

But our software tools and software practices go a long way to slowing
it down.  In stead of coding so code "goes everywhere", we are lazy and
tired and ignorant, so we write stuff that depends on endianism, address
size and machine type.  We also try to shoehorn existing code and design
for 19" monitors and desktop machines into PDAs, but don't get me
started on that.

Warmest regards,

maddog

*I did a calculation once that if you filled a one GigaByte disk for
every second of the day, every day of the year, it would take over 5,386
years to fill those disks with 2^64 bytes of data.**

**Yes, I did take into account leap years, and unlike Microsoft, I know
that every four hundred years we skip one.



More information about the gnhlug-discuss mailing list