The mid- to high-end Unix server market is due for a bit of a shake-up with a new HP server due -- but will it change anything?
HP is on the verge of launching new high-end Integrity servers -- based not on the dual-core Itanium 2, codenamed Montecito, which has yet to emerge from Intel's fabs but on Madison, the single-core version. This chip is widely seen as being a stopgap until Intel finishes cooking Montecito.
Indications are that, encouraged by the a report from analyst IDC that predicted a bright future for Itanium, HP's decided not to wait for Montecito.
Despite the lack of a big, headline-winning performance lift resulting from a Montecito-based product, HP reckons that bandwidth-limited users should still see some 30 per cent performance boost. Others meanwhile remain sceptical about the future of the chip, whose sales have remained largely flat, despite the billions of dollars being pumped into development, ecosystem support and marketing -- see below.
What's likely to be new is the server's chipset. Codenamed Arches, the third generation chipset can be used for systems with up to 64 CPUs and supports faster buses such as PCI Express. We'll find out more when HP launches the system next week.
But how will the HP offering compare to those of its main rivals, IBM and Sun? Before touching on this however, it's worth a reminder that over 90 per cent of the market's needs are met by servers housing x86 processors. There's not a lot of room for expansion in the high-end Unix business, according to Sageza Group analyst Clay Ryder, especially given the both the growing power of that platform, and the problems that that increasing heat generation and power demands pose for processor power universally.
On the vendor front, IBM threw down the gauntlet earlier this year as it leaked information about the forthcoming Power6 processor, which promises higher throughput while containing power draw. Scheduled to be launched in 2007 at the International Solid-State Circuits Conference (ISSCC) in San Francisco, it's said to remain cool while running twice as fast as the 1.9GHz Itanium.
IBM remains a strong competitor, well-entrenched in this sector, and which owns a broad product offering.
Meanwhile, Sun's eight-core Niagara processor -- the UltraSPARC T1 -- is out and has been hailed as revolutionary. It offers fast memory access for each CPU due to its integrated crossbar switch, even if each individual core isn't running as fast as a single UltraSPARC 4+. Given Sun's ability to tightly integrate Solaris with the chip -- and complexity within the OS is a given since this remains a RISC chip -- and the fact that it runs cool at 72W, what's not to like?
The problem remains that it's a Solaris-only chip -- and in these days of open source, community-based development, server buyers remain wary of single-vendor dependence. Compared to Solaris, not only is there a wider range of sysadmin staff to choose from when it comes to, for example, Windows, Unix or Linux, even if Sun's hardware is cheaper, staff remain add more red ink to the bottom line and will weigh heavier in the TCO equation.
But a more insuperable problem faces HP, no matter how much faster and better the new box is. It remains a high-end machine, disconnected from the mainstream of applications. While there's life there, it isn't as dynamic a market, so the number of applications won't -- and indeed hasn't -- mushroom overnight.
HP is likely to counter that it and the Itanium Solutions Alliance are pouring vast resources into development -- in December 2004, HP promised a $3 billion investment for hardware systems design to encourage software companies over the next three years.
However, that initiative is vendor- not customer-driven. When the money runs out, what will replace it? Will customers have the confidence that this initiative will endure? This is no casual point, since high-end systems are destined to last for considerably longer than the mayflies of the x86 world -- maybe decades.
It will be interesting to see what HP makes of this conundrum. We'll see next week.