Hopefully someone at Sun Microsystems can still explain why the Rock processor is still a good idea.
Today, Sun confirmed what we told you long ago - it's delaying Rock-based servers until the second half of 2009. The company made this admission at its analysts conference in San Francisco. We didn't hear a single question from the pundits about this slip - a curious performance to say the least.
The Rock-based boxes were once meant to arrive later this year, putting Sun in prime position to wallop Itanium- and Power-based systems from HP and IBM. The Rock chip boasts 16 cores and support for unreal amounts of memory. In addition, it can handle single- and multi-threaded software well, as we understand it.
Rather than shedding tears over the delay, Sun is pitching it as a victory for the company. The systems produced in conjunction with Fujitsu around the SPARC64 chips continue to sell very well. In the second half of this year, Sun will upgrade that gear with four-core versions of SPARC64. So, Sun presented the analysts with a "if it ain't broke why fix it" line to soften the blow of the Rock slip.
We see how attractive the prospect of fighting tooth and nail with IBM and HP for Unix server sales must be. Why bother giving yourself an edge by arguably jumping a generation ahead of rivals with a superstar new chip? That would be too easy.
Of course, what else is Sun supposed to say?
In the larger context, the Rock delay really might be okay.
In the first half of this year, Sun plans to release two-socket servers based on the Niagara III chip - aka Victoria Falls. Then, Sun will pop out blades based on that chip in the second half of 2008 and ship a four-socket server in the second half of 2008 as well. (We told you as much, er, two years ago, but who's counting.)
This Niagara-based gear sounds like some serious iron. You get 128 threads per box, and that adds up quickly when you think about stuffing a rack with, say, 48 Victoria Falls blades.
According to Sun's server chief John Fowler, a four-socket Victoria Falls server will match an eight-socket IBM p 570 running on 4.7GHz Power6 chips on database performance while consuming one quarter of the space and one third of the power. Sun's system will also be one-third the cost of IBM's, Fowler expects.
If Sun's figures hold up, then you have to wonder why it would even bother with Rock. Why not kill the chip and use the extra R&D money to keep the Niagara line on track? Surely, Sun could apply some of the Rock tricks to future versions of Niagara and end up in a similar spot. You really just lose out on the single-thread performance, and Sun seems well committed to a multi-core world.
But, hey, we're dumb hacks not business gurus. And we're sure an awful lot of you will appreciate the single thread splendor of Rock.
(Never mind that the US government was meant to foot part of the bill for Rock 2 before Sun lost a huge contract to Cray and IBM.)
Away from UltraSPARC land, Sun plans to ship more and more Barcelona- and Xeon-based servers in the coming year. It's already got tons of 2.0GHz four-core Opterons (Barcelona) humming away at a Texas supercomputing center. Yes, the bug patch has been applied. And, yes, someone is going to pay for an upgrade when the fixed Opterons arrive. AMD, we're looking at you.
With storage, Sun is set to release a number of "open" products in 2008 and to update its tape library gear in the first half of the year.
In addition, Sun will release a second version of its multi-host 10Gb Ethernet NIC in the second half of the year, while it will release a smaller scale version of its Mangum monster switch in the first half of the year. ®