R
Robert Myers
Dale Pontius wrote:
At this point, Itanium is _still_ mostly expectation. My point in
commenting on the book that started the thread is that Intel seemed to
have no interest in lowering expectations about Itanium.
Intel will do _something_ to diminish the handicap that Itanium
currently has due to in-order execution. The least painful thing that
Intel can do, as far as I understand things, is to use speculative
slices as a prefetch mechanism. That gets a big piece of the advantages
of OoO without changing the main thread control logic at all. Whether
that strategy works at an acceptable cost in transistors and power is
another question.
That single change could rewrite the rules for Itanium, because it will
take much of the heat off compilation and allow people more frequently
actually to see the kind of performance that Itanium now seems to
produce mostly only in benchmarks.
As to cost, Intel have made it clear that they are prepared to do
whatever they have to do to make the chip competitive.
As to how the big (more than 8-way) boxes behave, that's up to the
people who build the big boxes, isn't it? The future big boxes will
depend on board level interconnect and switching infrastructure, and if
anybody knows what that is going to look like in Intel's PCI Express
universe, I wish they'd tell me.
It gets harder to stick with the position all the time, but you still
have to take a deep breath when betting against Intel. The message
Intel wants you to hear is: IA-64 for mission critical big stuff, IA-32
for not-so-critical, not-so-big stuff.
No marketing baloney for you and you don't care what Intel wants you to
hear? That's reasonable and to be expected from technical people.
Itanium is where they intend to put their resources and support for high
end applications, and they apparently have no intention of backing away
from that. Feel free to ignore what they're spending so much money to
tell you. It's your nickel.
RM
The question for IA64 becomes can it bring enough to the table on future
revisions to make up for its obstacles. Will >8-way become compelling,
and a what price? At this point, AMD is trying to push its Opteron ASPs
up, but probably has more flex room than IA64 or Xeon.
At this point, Itanium is _still_ mostly expectation. My point in
commenting on the book that started the thread is that Intel seemed to
have no interest in lowering expectations about Itanium.
Intel will do _something_ to diminish the handicap that Itanium
currently has due to in-order execution. The least painful thing that
Intel can do, as far as I understand things, is to use speculative
slices as a prefetch mechanism. That gets a big piece of the advantages
of OoO without changing the main thread control logic at all. Whether
that strategy works at an acceptable cost in transistors and power is
another question.
That single change could rewrite the rules for Itanium, because it will
take much of the heat off compilation and allow people more frequently
actually to see the kind of performance that Itanium now seems to
produce mostly only in benchmarks.
As to cost, Intel have made it clear that they are prepared to do
whatever they have to do to make the chip competitive.
As to how the big (more than 8-way) boxes behave, that's up to the
people who build the big boxes, isn't it? The future big boxes will
depend on board level interconnect and switching infrastructure, and if
anybody knows what that is going to look like in Intel's PCI Express
universe, I wish they'd tell me.
It gets harder to stick with the position all the time, but you still
have to take a deep breath when betting against Intel. The message
Intel wants you to hear is: IA-64 for mission critical big stuff, IA-32
for not-so-critical, not-so-big stuff.
No marketing baloney for you and you don't care what Intel wants you to
hear? That's reasonable and to be expected from technical people.
Itanium is where they intend to put their resources and support for high
end applications, and they apparently have no intention of backing away
from that. Feel free to ignore what they're spending so much money to
tell you. It's your nickel.
RM