Re: Will Prescott work on Win64?

On Wed, 03 Dec 2003 00:20:53 GMT, "News sbc"
<[EMAIL PROTECTED]> wrote:
>I just had a vision, a flash:
>
>Some of the applications that most desperately need >32 bits
>of virtual address are the applications Intel needs to assemble
>and tape out a chip.

Yes, Intel and every other vendor doing this sort of design.

>I know, from when I was at Intel, that around the end of P6 (1995)
>some of these had to be rewritten to partition the database into
>chunks that fit in a 32 bit address space.But that trick doesn't
>work forever.

I don't know exactly when these tools were rewritten, but yeah, a lot
of them are designed to use all the ugly hacks that make them work on
x86 (32-bit) chips with more than 4GB of memory.

>I don't know if Willamette taped out using such kluges, or if
>it taped out on 64 bit HP-PAs or Itanics.
>
>Continuing to tape out on Itanics is probably not so good an idea,
>since these apps are often the type that run most poorly on Itanic.

I don't see what that should be the case, given a sufficiently
intelligent compiler.  SPEC CINT2000 has a couple sub-tests that are
related to this in 175.vpr and 300.twolf tests (both are targeting
FCPG Place 'n Route, not quite the same thing but should have similar
performance characteristics).  The Itanium2 isn't exactly running away
with it here, but in both tests the Itanium2 (1.5GHz/6MB L3) is
roughly equal to the Opteron 246 (2.0GHz) processor, and both are near
the top (P4's and Athlon64 FX chips are slightly ahead).  There's
nothing in taping out that I can think of which would make the code
very bad for the explicitly parallel architecture of IA-64.

Either way, you better believe that the Itanium is EXACTLY what Intel
wants the whole world to use for this sort of task.  It's also what
Intel uses internally.

>HP PA probably has not kept up. I doubt that anyone at Intel would
>risk using AMD 64 bit chips to tape out, even if they are significantly
>better price performance wise for the task than Itanium.

Hell no!  It's IA-64 all the way!

>Which leads to my flash:
>
>If, as is widely conjectured, some Intel chip has a non-EPIC 64 bit
>extension to the x86, perhaps Intel may not release that chip for
>sale to the outside world?  Perhaps they would use it just for internal
>use, as in taping out a new CPU?
>
>If Intel is using 3rd party tools, this would probably imply that Intel has
>implemented the AMD-64 instruction set --- since such 3rd party tools
>are unlikely to be available for an unreleased Intel x86-64 that is
>incompatible with AMD-64.

Intel is the #1 customer for third-party tools for chip design in the
world.  And guess what?  When your number 1 customer comes up and says
"Give us an IA-64 port ASAP!", the tool vendors listen.

>I wonder if Intel has purchased any AMD-64 licences for CAD tools?
>I suppose if they had, they would have tried to make it untraceable.

Intel likely doesn't need an AMD64 license for any reason, the
cross-licensing agreement they have with AMD should cover them.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca