Itanium: A cautionary tale

TalkBack 6 of 27:: Next »; « Previous

Thread View
Flat View

History lesson: Sun became the number one UNIX vendor when they came out with SPARC. One thing they did was to have a catalog of 10000 applications that were SPARC-ready, BEFORE they started selling SPARC machines. This approach worked, as customers were much less leery of taking the plunge from their 68000-based Sun3's.

When DEC came out with Alpha, they did no such thing. DEC customers were faced with giving up on Ultrix (BSD Unix) on MIPS, and moving to OSF (SysV Unix) on Alpha. Never mind that the Alpha chip was (and still is) one of the best designs ever for a processor. OSF was to be a mach-kernel based UNIX, and the direction of the industry (who BAILED as soon as DEC came out with it). But with few applications and a complicated upgrade (replacement) path, DEC's customers rebelled and went to other vendors (such as Sun!). The rest is history as DEC disappeared almost overnight.

HP made a nice, decent performing chip called PA-RISC. Over the years they came to the realization that "keeping up with the Joneses" in chips meant billion-dollar fabs, and even then they would still lag behind the industry leader (InHell). After DECs demise (partly due to billions spent on the Alpha fab), HP decided to partner with InHell and let THEM make the processor. It was (and still is) a good decision. Like the Alpha, the Itanium was a LEAP in technology (not an evolutionary step). VLIW was the mantra of the college computer departments of the time, so EPIC was born.

EPIC is the natural progression of the CPU. CISC processors (like x86) were "evolved" into RISC chips (like SPARC, PPC, PA-RISC, MIPS), by removing instructions in hardware and migrating them to software. But over the years, complex pipelining and branch prediction routines have "crept" back into hardware - making chips more complex. The EPIC architecture removes those hardware "instructions" and PRESUMABLY migrates them to software. This makes for a nice, relatively simple design - which also maximizes chip "real estate". When Moore's Law dies, and chips become "finite" in size - EPIC will make the best use of it.

The problem with Itanium was InHell. They THINK they are expert chip designers, but in reality they are just good at manufacturing chips. First InHell demanded to put "something" into the Itanic that would allow it to run x86 programs. At first it was an entire P4 core, but it was "whittled down" to a ALU that performed x86 processing, while using the Itanic registers and other parts. This design took many iterations and time and money to create - and the results are disappointing. HP finally had enough of InHell "meddling", so at first they split off their design team to "work with" InHell designers (literally in the same building - on opposite sides, and never the two would meet!). They arraingement didn't work and wasted MORE time. Eventually HP said "the HELL with it", and spun off their own designers to InHell. This hurt HP, as they were forced to design more PA-RISC chips while the Itanic was sinking.

The rest is history. InHell SLOWLY makes the Itanic better - pumping out more water than is rushing in, but it sure shows how inept InHell is at chip design (their Xscale chip was designed by ARM). Their other forays into RISC chips (i700, i800, i900) were mostly failures. In fact, x86 seems to be the ONLY CPU they are capable of designing (and it has been evolutionary).

Just what was Steve Jobs thinking? GEEZ, he made the same decision that HP made a decade ago - go with the chip "experts". Time will tell if this decision was as fateful as HP's was . . .; Posted by: Roger Ramjet Posted on: 12/07/05 You are currently: a Guest | Members login | Terms of Use