Intel - Itanium Chips

Steve Hayashi sth at
Sat Jan 6 13:32:25 PST 2001

I'm definately talking out of my rear end when I say this (since I'm a
hardware guy, not a software guy), but precision has nothing to do with it
yet.  The current IEEE implementation of Floating point numbers give the
ability to calculate both REALLY big and REALLY small numbers.  There are
bits representing the exponent and bits representing the mantissa (a sort
of fraction/decimal setup).  I don't know what the specs are for 64 bit
floating point, but I'm not sure how important that precision will become.

Now, the Itanium ISA is called EPIC, which is a fancy way of saying Very
Long Instruction Word (VLIW), whose idea actually predates the PC.  With
it, as I recall, you can have several independent instructions run at the
same time.  How it works and how fast it is supposed to be is "no concern
of the software people"  It WILL be faster, however, in serious number
crunching.  The trouble will be getting compilers to actually take
advantage of this speed.  Instructions must not have dependencies to each
other.  Even so, I remember hearing about Look-ahead
buffers/microprocessors that actually do a decent job of looking ahead in
the instructions and following branches around.

End result, don't worry about it unless you plan on helping with the
compiler.  If you do, then get ready for a nightmare.


On Sat, 6 Jan 2001, jerry wrote:

> In-Reply-To: <000901c077e8$aa53a3a0$6564a8c0 at mceddy>; from at on Sat, Jan 06, 2001 at 01:57:52PM -0000
> 32bit pentium chips have >= 64bit precision through the
> integrated math coprocessor. I have not read the specs on the itanium processor
> but it probably does not have any greater precicsion except for its default
> integer instructions which will be 64 vs 32 bit.
> jpd

Unsubscribe: send email to lfs-discuss-request at
and put unsubscribe in the subject header of the message

More information about the lfs-dev mailing list