[Info-vax] RealWorldTech on Poulson

[Christopher]

On Jul 2, 11:12 am, JF Mezei <jfmezei.spam... at vaxination.ca> wrote:
> Neil Rieck wrote:
> > I agree. Many industry people responsible for hyping VLIW/EPIC are no
> > longer working which means there will be no personal embarrassment
> > associated with an about-face. Now is the time for a young Intel "VP
> > of engineering" to make a statement similar to this:
>
> Admitting that EPIC did not pan out is one thing.  But transforming an
> EPIC architecture into an OOE one is another.
>
> If this were a true deep down transformation of the architecture, it
> would mean that upgrading from Tukwila to Poulson based system would be
> a massive undertaking for customers who would have to use new compilers
> to recompile their software and more importantly, because of the new
> compilers, retest/certify all of their applications.
>
> This is why I think the OOE nature will be very limited and work only
> within what the existing compilers have told the CPU can execute in
> parralel.


Those are silly statements to make.  If x86 has taught us anything, it
is that the original ISA philosphy means very little.  If you throw
enough hardware at the problem you can turn CISC into RISC.  The whole
point of OOE systems is that the processor tries to determine
instruction concurrency.  You wouldn't need to recompile your software
to benefit from hardware-level OOE.  If anything, VLIW/EPIC
architectures have the ability to surpass typical CISC and raw RISC
systems in OOE by providing the hardware with much more information
about where concurrency might be found.