One "hyped" thing on the market is the perpetually-soon-to-be-released Intel "Merced," the first processor using Intel's IA-64 architecture. In the media, there is just the most intense "hype" over this; now that Intel is working on a 64-bit processor, people have assortedly:
Forgotten that others were there first:
Atari (the "computer game" people) had their "Jaguar" processor five years ago that was sold in many home "game console" systems.
Digital has been selling the 64 bit Alpha for on the order of ten years now.
Sun's "UltraSPARC" has been on the market for a couple of years.
MIPS/SGI have some 64 bit processors.
It is highly probable that both Alpha and SPARC will be considerably faster than Merced on into the forseeable future, thus indicating that while IA-64 might provide substantially better performance than IA-32 some time around 2001, it's not likely to provide "the fastest systems in the West" any time soon.
Ignored the kinds of fabrication technology that IA-64 requires.
The "Merced" series appears to involve so many transistors that it is almost beyond the present state of the art in microelectronics fabrication.
The point of this is that unless Intel has some earthshaking technological advances they're not using yet, it will be very difficult to get decent chip yields early on. That is, they will likely produce a lot of bad chips that will have to be repaired (which has a cost in time and money) or discarded.
The quick summary: " Merced chips are hard to build, and thus the production process is likely to have a fairly high failure rate, thus making the ones that do work rather expensive."
Compilers for IA-64 appear to be hard to write
One significant "problem" with the IA-64 is that creating fast code appears to require sophisticated instruction scheduling tools within the compilers ("EPIC"). If this ends up requiring that, for fast performance, code needs to be recompiled for each IA-64 variant, this will likely prove to be a crippling problem.
At this point, it appears that modifying GCC ) to build optimized IA-64 code will prove to be a difficult task, requiring code generation techniques that are not yet widely understood.
Furthermore, Intel has not yet released enough information outside of NDA agreements to allow work on a GCC code generator to get started. This is most likely not a "conspiracy against free software," but rather merely intent not to release internal information about the chip to the public, and thus to competitors, before it is available for sale.
There have been rumours that Intel engineers have created a GCC code generator; whether that particular set of code will be released publicly soon or ever is anyone's guess at this point.
In more encouraging news, a press release indicates that The Portland Group (PGI) plans to develop IA-64 compilers, and Linux is mentioned as one of the "target platforms." They already sell their line of "High Performance Computing" / parallel compiler tools including FORTRAN 77, Fortran 90, C, and C++ for Linux.
This does not directly translate to indicate that Linux will run on IA-64; the Linux kernel contains a fair bit of code that will only compile using GCC, and it seems unlikely to me that Linus and other kernel developers would be quick to move to a non-free compiler.
Furthermore, Merced is not yet available, and according to some press releases, won't be available in quantity until year 2001. McKinley to be available in 2001
Note that this is only the latest of several slippages:
TechWeb News, April 14, 1997, reports a slip from 1998 to 1999.
PC Week, Jan 22, 1996 reports a slip of the P7 (renamed to Merced) from 1997 to 1998.
PC Week, Jan 19, 1993 had an article that says that Intel would deliver a 64-bit chip, the P7, in 1995, with the first systems shipping in 1996. (This last item did not have its source in HP/Intel, and probably should not be classified as anything but a rumor).
It is not beyond the realm of possibility that Merced will never be released in useful form, much like Intel's "432" architecture.
A Gartner Group paper indicates that: " A high-volume market for IA-64 systems will not develop before 2003 (0.8 probability). More than 80 percent of 32-bit applications will perform at least as well or better on high-end IA-32 systems compared to initial IA-64 systems (except in applications running on a 64-bit database in the same server) through the 2002 time frame (0.7 probability). "
This is in stark contrast with other Gartner Group papers that had been riding on the "Merced is the only CPU architecture that matters at all" bandwagon.
[ Compaq kicks Intel's Merced butt while it's down - THE REGISTERd]
The relevance to Linux includes the following points:
Linux will not be available for Merced until there is a code generator for GCC for Merced.
Intel marketers have discovered that they sell enough chips for Linux that it represents a market that they need to care about.
The primary thing that they can do to help Linux to be available for Merced (thus "selling" Merced chips) is to provide the information necessary to build a decent code generator for GCC.
Linux has already been ported to two 64 bit architectures, namely Digital's "Alpha" and Sun's "UltraSPARC."
As a result of this as well as the various 32 bit ports, there is a good level of consciousness of what 64- and 32-bit dependencies exist. A Merced port should prove straightforward at least from this perspective.
Linux on IA-64 - built by a group of Silicon Valley companies including VA Linux Systems, CERN, Hewlett Packard, IBM, Intel and SGI.
The IA-64 Linux Developer's Kit is a set of installable software components that allow users to develop, execute and debug IA-64 Linux applications on IA-32 computers running the Linux operating system.
In 2003 news, Linux has apparently concluded that IA-64 isn't likely to turn out well...