When we spoke about Intel cutting power requirements on current CPUs, we mentioned that others had done some speculating about this, too.
The article we were referring to was this one.
This article starts by pointing out that some of the reasons cited by Apple to switch to Intel indicate performance/power ratios far better than turning off leaky circuits.
It then goes into quite a bit of detail as to how Intel has gathered together all the pieces necessary to create a VLIW processor that can translate x86 code on the fly and thus get rid of a lot of circuitry.
It’s a good piece of detective work, however, it downplays the simplest explanation as to why Intel might want to do this:
In a word: Itanium. Well, to be more accurate, Son/Grandson of the current Itanium.
It seems like everyone has dismissed Itanium except Intel. This may not be a good idea.
People do not remember that the IA-64 design wasn’t supposed to hit desktops until 2007.
Yes, a lot has happened since then, but the reasons why Intel started up IA-64 in the first place have only intensified since then. It’s not just a matter of performance anymore, but heat.
Yes, the original Itanium was a dog in x86 emulation, but as the article notes, Intel has been doing a lot under the horizon to fix that.
Yes, current Itaniums are huge beasts, but remember that at 45nm, everything is a quarter of it’s current 90nm size. A 9Mb cache at 45nm would only be a bit bigger than a 2Mb cache today.
Nor is it to be doubted that any Itanium Reloaded would be heavily redesigned to strip out all the redundancies and mistakes of the earlier Itaniums and take advantage of whatever benefit multicore technology has to offer.
A code-morphing CPU (and in all likelihood, the morphing-part would be transitional) would let people use their x86 software, and essentially get a new recompiled version by just using it. Might not be the best recompile in the world, but it would probably do. This would eliminate the biggest problems folks have had with Itanium.
(Something like this probably didn’t hurt one little bit talking to Apple, either.)
Finally, there’s one old reason that’s the best reason of all for Intel:
Cutting AMD Out
Put simply, AMD has the legal right to use x86. It has no right to use IA-64, or any derivation thereof which isn’t x86-based.
If Intel throws out a fast code-morphing Descendant of Itanium for the desktop, AMD has a BIG problem.
This is one of the, if not the big carrot Intel has been pursuing as it has doggedly continued to develop Itanium throughout the years; getting AMD off its back, or at least paying for the ride.
Again, Apple may well have given us a clue as to when we might expect this.
When Apple announced they were switching to Intel, the timing of the switch puzzled many. The cheapy models were due to switch in 2006, but the heavy-duty items weren’t supposed to go until 2007.
That didn’t make sense at the time, but if something like this is in the cards for Intel, this staggered switch makes a lot of sense. Machines like the Mini Mac will get a 65nm Pentium M, while PowerMacs owners will be among the first (if not the first) kids on their block to get this x86 (MacOS X?) code-morphing chip.
Also keep in mind that IDF generally doesn’t talk about Coming Real Soon innovation, but rather technologies that are a year or more away.
So, if the sucker exists, it’s a) going to be a 2007 item and b) we’re likely to hear at least some hints about this next week (if for no other reason than there’s not a lot for Intel to crow about with 65nm), and probably hear all about it at the following IDF.
At least on the PC side, the initial iteration of the product may be something along the lines of the Pentium Pro, a sort-of desktop chip that would eventually migrate to the mainstream desktop line.
Yet another reason to pay close attention to what come out of IDF.