YK: > Looks like Intel decided to reduce the maximum
addressable physical memory on Nocona from 40 bits
to 36 bits.
It doesn't strike me as a reduction. I think it more
likely that despite the new chipset, there are still
some 36-bit restrictions in the infrastructure, a
legacy of PAE (which is also 36-bit on Xeon).
I do agree with the assessment that Intel must have
been in a "tearing hurry" to implement AMD64, as
evidenced by the sloppy paraphrasings in the
documents, and the defective implementation in sand.
After the Illegal Instruction Erratum of some years
back, I'd expected Intel to be much more disciplined
about elementary due-diligence testing. Guess not.
Perhaps some future historian will be able to reveal
to us just what event(s) provoked Intel to adopt
AMD64, and to do it so rashly. Hints are to be found
around the web on this topic, such as:
- Long-standing Microsoft disdain for IA-64, plus
what Intel can see from IA-64 sales trends
- MS probably told Intel to completely forget about
Windows on Yamhill (unless Yamhill became AMD64)
- Clear signs of pervasive 64-bit computing years
before the Intel roadmaps showed it happening - heck
there are already AMD64 laptops
- Tier One OEM adoption of AMD64 (other than Dell)
- Another round of Dell-AMD rumors (which might
suggest Dell leaning on Intel for a 64-bit
solution they can actually sell)
- Mentor, PTC and UGS dropping support for some of
their workstation apps on IA-64 (revealed in the
last 30 days, but Intel likely had indications
earlier)
- Rumors that Longhorn might not support IA-64, yet
might be 64-bit only
Anyway, here we are 17 days after Nocona launch, and
we still have no real info on 64-bit performance. True,
the public beta of Windows XP64 won't run on it, and
apparently the NDA beta has a "no publish" restriction.
But does Linux 64 really not run at all on EM64T?
Has Intel said anything about microcode/BIOS patches
to fix this?
