Re: How much RAM can debian support?
On Sun, 11 Oct 2009 05:35:18 -0500
Mark Allums <mark@allums.com> wrote:
> Tzafrir Cohen wrote:
>
> >> Personally, I don't see a need to go 128 bit on a main cpu unless you
> >> have a desire to count and enumerate every elementary particle in the
> >> known universe, without a) running out of RAM, or b) spilling the
> >> content into a multiregister add/adc pair. :)
> >
> > You don't. It seems others do:
> > http://redmondmag.com/articles/2009/10/08/rumor-windows-8-will-be-128-bit.aspx
> >
> > The PR departments has strange hardware requirements.
>
>
> 128-bits may apply to something besides memory addresses. Although, it
> is hard to conceive of a need for 128-bit precision in arithmetic ops.
>
I know some scientists that claim that they could use 128bit arithmetic,
especially for unstable or badly scaled models (such as weather models)
Others are actually moving back to single precision with the rise of the GPU
and there are claims that for most work, with correct order of operations using
a mixed single/double precision where most of the work is in single precision
will do. Although the unavailing of the fermi architecture from nvidia is
pushing people to think of double precission only worlds again.
> Some CPUs are internally actually VLIW[0] machines, even if they present
> a standard architecture to the outside world.
>
> Microsoft may be doing infrastructure work that will not see the light
> for 20 years or more, if ever. But, better to do it now than later.
> The Y2K thing comes to mind here. Also, they may want to see Windows
> running on serious supercomputing iron. (For PR reasons, of course.)
>
>
> MArk Allums
>
> 0. Very Long Instruction Word
>
>
Reply to: