David Hess (davidwhess.delete@this.gmail.com) on August 22, 2013 2:38 pm wrote:
> Doug S (foo.delete@this.bar.bar) on August 21, 2013 5:18 pm wrote:
> >
> > As for the main point of your post, I think AMD64 succeeded because it was obvious there was demand
> > for 64 bit server CPUs, and even though Intel had designed its own extension, it held it back
> > because the plan was to keep x86 at 32 bits and force 64 bit to IA64, so that eventually over
> > the course of a decade or more as everything migrated to 64 bit Intel would have a monopoly enforced
> > by patents - such a monopoly being perfectly legal to abuse according to US law!
>
> Were details of the x86 64-bit extension Intel designed every released?
>
> I thought Intel's implementation of AMD64 came awfully fast so assumed they had
> inside information and had copied it initially instead of designing their own.
>


The only thing I ever saw about it were the articles Hans De Vries wrote about the clues to its existence found in Prescott. AFAIK Intel never released details about it, but perhaps some engineers who worked on it have dropped some hints about it here and there.

No one knows for certain of course, but it makes sense that Intel would have had it ready to enable if market conditions dictated (i.e. AMD's version starting to get a bit of traction)

I think they assumed AMD would try to do their own 64 bit extension, but believed they could kill it with their own if necessary in the same way Intel's SSE killed AMD's 3DNow. I believe they were caught off guard when Microsoft announced support for it, and further caught off guard when Microsoft told them they would not support a second 64 bit x86 extension. And of course caught off guard by the realization that breaking up the P4's pipeline to expose the internal double clocking necessary to reach their 10 GHz goal was not going to perform well at all and would be even worse in terms of power consumption.

I would think the fact they already had a 64 bit extension implemented in the P4 core made adding AMD64 support fairly easy. If I had to bet, I'd guess they probably didn't add the extra registers (why make x86 perform better when they want to sell IA64?) but aside from that I'd be surprised if it was all that different than AMD's effort. Certainly whatever they had implemented gave them a big head start.

Note how much longer it took them to go 64 bits in the P6 based Core/Core2 line. Since they hadn't done any work in that core to add 64 bits, as it had been kept around for laptops which P4 proved unsuitable for, it took a lot more effort to add it.