Intel: best days are ahead for servers
Xeons everywhere, virt no cause for alarm
As it comes off yet another global recession, and with a severe freeze in IT spending starting to thaw, Intel's server-chip lineup is perhaps the best
Chipzilla The Continuum has ever put into the field - including its much-delayed and much-maligned quad-core Itanium 9300s.
What a difference seven years makes.
Today, rival Advanced Micro Devices doesn't have all the right ideas as it did back in 2003 with its Opterons. Or rather, Intel now has its own versions of those same right ideas while AMD is having a tougher time getting what it deems is its fair share of the microprocessor and chipset sales in the server racket.
In addition, here in 2010 the Sparc and Sparc64 processors from Oracle and Fujitsu, respectively, seem to be frozen in time. And everyone knows that except for three operating systems controlled by HP - OpenVMS, HP-UX, and NonStop - and a smattering of mainframe operating systems in Europe and Japan, Itanium is not a contender. Red Hat and Microsoft have each pulled the plug on Itanium for future development, as IBM and Sun did for AIX and Solaris a decade ago, before Itanium even came to market.
The way, it seems, is clear for the increasing dominance of Intel's Xeon server chips.
But that's not the only thing that has Kirk Skaugen, general manager of Intel's Data Center Group, stoked. The Data Center Group was created this year not just to chase the server-chip market, but also to get Xeons into storage arrays and networking gear, replacing Power and other processors that have long had dominance there.
Skaugen says that a few years back, Intel was getting about 20 per cent of the design wins from storage vendors opting to use its Xeons; this year that share will hit 70 per cent and next year it will be 80 per cent or so, he said. The same kinds of trends are holding in networking gear, which uses Power, MIPS, and other specialized processors.
"This is growth for us," Skaugen said at a meeting yesterday that Intel hosted for Wall Street analysts. "But you won't see it immediately because it takes time for these companies to transition their products."
Think of it as Xeon money in the bank - just like Intel earns in the two-socket workstation space, which currently accounts for about 15 percent of the Xeon CPU's total addressable market and is expected to be about the same in 2014. Intel basically owns the two-socket workstation market - a market that helped put RISC/Unix on the map twenty years ago in commercial computing but one that IBM, Sun, HP, and others could not hold against the onslaught of Windows (and then Linux) coupled with cheap-but-powerful x64 iron.
Another target that Intel has had its eye on for the past decade, Skaugen explained to Wall Street, is the high-end RISC and mainframe server market, a $16bn market Intel felt confident it could finally take down a peg with the new eight-core Nehalem-EX Xeon 7500 processors announced at the end of March.
The way IDC carved up the server racket in 2009, the Xeon family of processors drove about $22bn in server revenues, with Intel taking in about $4.4bn in chip and chipset sales. Itanium servers accounted for $4bn in revenues, and gave Intel an undisclosed revenue share - probably something on the order of $1bn.
AMD's Opterons drove another $3bn in server sales, leaving $16bn that companies shelled out for mainframes, RISC/Unix, and a smattering of proprietary midrange gear. Skaugen estimates that the RISC and mainframe server space represents a $1bn to $1.5bn processor opportunity for Intel - that is, if it can convert those customers to Xeon 7500 systems, which have some of the mainframe-class reliability features missing from prior Xeons but in Itanium.
I know what you are thinking: "How can $22bn in x64 server sales bring Intel about $4.4bn, but converting $16bn in RISC and mainframe sales to Xeons will only bring $1bn to $1.5bn?"
Well, it's precisely because those machines are not compatible with Xeons and do have some features that Xeons are missing that made customers spend a huge amount of money on them. Once customers are willing to convert, however, they're not going to spend the same kind of money. And they don't. And unless the economy gets a lot worse, that base of Unix and proprietary machines will stand pat, out-gassing a little here and there to be sure, like a piece of dry ice. But it won't simply sublimate all at once to Xeons, even if the Xeon 7500s do have a shiny new machine check architecture.
People don't spend money disrupting systems that work, no matter how much they might save. Ask OpenVMS and AS/400 shops. Ask mainframe shops. They'll tell you. This may not make sense, but that's people for you. Messing with something has a potential opportunity cost, just like leaving it alone has an opportunity profit. The money you don't spend on changing something that works can be used to create something new. Which is what IT shops do all the time, and it's not illogical. It is just frustrating - and, in some ways of looking at it, costly.