Intel launches the "McKinley" Itanium 2 processor today. It hopes to set itself on a profitable path of dominance in the market for processors for relatively expensive midrange and enterprise server servers, much as it accomplished in the PC market by the late 1980s, and in the entry server market in the mid-1990s,Timothy Prickett Morgan writes>i>.
The first-generation "Merced" Itanium chips were years late, under-performing, and widely mocked by the RISC/Unix vendors who still control the server market - except for certain high-performance clusters and specialized web applications like data encryption.
The second generation of 64-bit Itanium processors, which are about six months later coming to market than Intel and its partners had hoped, have enough credibility in the market that just about all the RISC/Unix vendors have suddenly become more forthright about their processor roadmaps than they would normally be. This is no accident. The remaining server vendors who sell machines which do not use Intel processors want to show that they are committed to their processors even if, in many cases, these vendors also hope to make a killing selling McKinley machines.
Intel makes no bones about its intentions when it comes to the server market. Plainly and boldly put, Intel wants to drive the server makers out of the chip design and fabrication business so they will port their architectures and operating systems to the Itanium chips.
Intel already accounts, by the latest reckoning of analysts of International Data Corp, for more than 88% of total server shipments worldwide because of the fact that entry servers using one or two processors still make up the bulk of server shipments each year.
Intel also has a fairly good share of four-way servers, and vendors who use Intel's high-end Pentium III Xeon and Pentium 4 Xeon processors sell thousands of eight-way servers a year. The Intel architecture - which is not entirely under Intel's control, but which is being enhanced by various chipset vendors and server markers - has also been extended into 16-way and larger servers by Unisys Corp, IBM Corp, NEC Corp, and Bull. And others like Hewlett-Packard Co are working on their own high-end servers using Pentium 4 Xeon and Itanium 2 processors. All of the major server vendors share in Intel's vision of dominance, and are ironically (some might say fatalistically) the vehicle with which Intel will attain its dominance of the top end of the server business.
Intel can count on the economics of its high-volume chip fabs to keep the pressure on those vendors who will design and/or make their own server processors, and it can also count on the drug dealer's logic - if I don't sell Itanium machines against my own servers, someone else will - to keep all but the most stubborn and idealistic vendors from endorsing Itanium, regardless of whatever technical inferiorities Itanium might have compared to RISC/Unix or proprietary alternatives.
The server architecture that offers the best price/performance and the widest software support will win in the midrange and enterprise space over the long haul. Of the $49bn a year that companies spend on the core electronics complexes in all types and sizes of servers, IA-32 processors already account for $25bn.
Intel is gunning for that other $24bn that is dominated by RISC/Unix products and proprietary machines, and history would seem to suggest that within a decade, if Intel can improve its Itanium designs so they are consistently less expensive than RISC/Unix alternatives and deliver comparable performance, there is no reason to believe that Itanium won't relegate other architectures to the dust bin.
These are two very, very large ifs, which is why many server vendors, stung by delays in their own processors as well as those from Intel, are playing both sides of the street for as long as they can.
There's too much money to be made selling RISC/Unix products today to stop, and there is too much money to lose by not selling Itanium machines starting now and perhaps in very high volumes in a year or two to not commit to doing it now. (Unless you are Dell Computer Corp, which is bizarrely taking a wait-and-see attitude and is not announcing any products or plans for the Itanium 2 chips today at the McKinley launch.)
The good thing about McKinley machines is that future Itanium processors will plug into machines that support McKinley chips and which are rolled out over the course of the summer. This was not the case with the Merced machines, which were a separate product unto themselves. But knowing that the Madison, Deerfield, and Montecito Itanium chips will plug into McKinley boxes next year and the year after should help ease the sales process to server buyers who were correctly skeptical about the prospects for Merced machines. Intel will roll out its own "Tiger" four-way servers, based on the E8870 chipset (formerly known as the i870 chipset) alongside the McKinley chips, and ServerWorks, IBM, HP, Unisys, NEC, and others are expected to divulge their own chipsets for McKinley and future Itanium processors.
The Itanium 2 processor will be available in three different configurations. A 900MHz version of the chip with 1.5MB of on-chip L3 cache will sell for $1,338. In all the time that everyone has been talking about Itanium 2, no one has even mentioned that a 900MHz version would be available. The 1GHz version of the chip with 1.5MB of L3 cache costs nearly twice as much at $2,247. This pricing difference and the advent of the 900MHz clock speed would seem to suggest that Intel is having trouble getting decent yields on the 1GHz McKinley at this point in its product ramp up.
The 1GHz version of the chip with 3MB of L3 cache costs $4,226, which is a little bit less than RISC/Unix vendors charge in their homegrown processors for entry and midrange servers. As Intel can ramp up production quantities for Itaniums and the clock speed on those chips, those prices will inevitably drop hard and fast, like a boom.