Gartner analysts and odds-makers George Weiss and John Enck are making some five-year forecasts about the state of server technologies.
In the face of Intel's architectures (IA), Enck and Weiss' outlook for RISC architectures was pretty glum. They predicted that, by the year 2007, the probability was about 75 percent that buyers would turn to IA rather than RISC. However, they do not expect IA-64 to displace IA-32, except in applications that depend on 64-bit addressing.
In fact, during keynote sessions at the Gartner Symposium/ITxpo in Lake Buena Vista, Fla., Enck and Weiss predicted IA-64 penetration of about 7-8 percent.
Two keynote speakers, Intel CEO Craig Barrett and HP CEO Carly Fiorina, disputed that estimate.
Fiorina is counting on IA-64 to be the platform to which HP's "heterogeneous" customers consolidate. However, outside of Windows and Linux (which are supported by other vendors), that heterogeneity is very HP-oriented; the company's own operating systems (HP-UX, NSK, and OpenVMS) are slated to be moved off their current architectures (PA-RISC, Alpha) to IA-64.
Intel's Barrett acknowledged that the ecosystem of systems maker/OS provider/applications developer must be propped up in order for IA-64 to succeed, but added that Intel will make whatever investments are necessary to see that through.
Enck and Weiss said that symmetric multiprocessing (SMP) systems have approximately a 70 percent chance of giving way to more fabric-like architectures where applications are distributed across systems connected to the same bus--a bus that could be dedicated system interconnects (aka clusters) or an IP network (aka grids). Even so, said the analysts, there will always be a place for large, vertically scalable SMP systems where other fabric-oriented solutions won't be able to match the prowess or I/O of something on the order of 64- or 128-processor configurations.
Perhaps the most interesting prediction however had to do with the analysts' outlook on 32/64-bit hybrid architectures. They talked about the chances of success for both AMD's Opteron (code-named Hammer) and Intel's Yamhill. (Intel denies Yamhill's existence. But, without naming names, the Gartner analysts said that they've spoken to several major application developers who have tested an Intel prototype.) The analysts didn't seem very optimistic about either hybrid's chances of success.
In AMD's case, the analysts don't believe the Opteron meets any of the criteria for new processor adoption--application vendor enthusiasm, server vendor commitment, server vendor enthusiasm, or and customer enthusiasm. Even with OS support from Windows and Linux, unless there's a change in those critical aspects of Opteron's ecosystem, I am standing by my previous assessment: that when it comes to enterprise computing, Opteron is going to be a dud.
The analysts were equally bearish about Yamhill, but for different reasons. Intel's entry into the hybrid space could legitimize the 32/64 hybrid approach which, based on the enthusiasm for Opteron, currently has limited legitimacy. At a time when Intel is determined to buck analysts' estimates for IA-64's market penetration, the last thing that Intel will want to do is cannibalize Itanium's chances of success with architecture. For that reason, Enck and Weiss say Yamhill will never see the light of day.
Got your own server predictions and want to share it with your fellow ZDNet readers as well as Gartner's analysts? Use TalkBack below, or write to david.berlind@cnet.com.
