QwikSand,
This is a crucial question and, while the disparity may not be this large forever, if there's no disparity at all and no edge for Sun, Sun's goose is cooked. I would like to know your reasons for thinking it will vanish.
Opteron took a lead in number of areas: 64 bit, power consumption, on-die memory controller, glueless CPU interconnect, dual core. Intel realized that, and is trying to catch up in all areas, and eventually may.
Sun, BTW, missed the boat on first year of this opportunity (Opteron was released in 2003) completely, and to great extend for more than 2 years by coming out incredible late with the first 3 members or the galaxy. These machines are still not shipping (will soon) and there are more to be unvailed later in 2006, representing lost opportunities. Blades are supposed to be released in 2006, which is almost laughable. HP may have a 2nd or 3rd generation of Opteron blades by then.
The boxes Sun announced look good, but they should have been on the market 9 months to a year ago. Sun still completely does not get the product cycles in the industry standard servers. You come out with a design the day the new CPU, chipset enhancement or memory technology is released. Every day you are past that day equals opportunity lost. There are some products / technologies that went into these boxes that Sun needed to get from third parties. The one that is known as far as what is in is the 2100, and the components that went in have been around for about a year, and that is the opportunity lost.
Going back to things that Opteron has, or areas where it leads where Intel may catch up, Intel already has the 64 bit instruction set implemented in one of its lines, high power consumption line, and will have it in middle of next year in lower power consumption line.
So as far as power consumption, there is still a one year lead until mid 2007 when Intel will have a lower power consumption server line. So if Opteron is standing still, Intel may catch up.
On dual core, Opteron is into 2nd quarter with availability of good dual core implementation, Intel will have very poor one in Q4 2005, less poor one in Q1 2006, probably a good one in Q3 2006. (Opteron may be quad core by then, we will see).
Anyway, you see these quick cycles. Dell will be releasing fresh lines implementing them immediately. This is just to illustrate how unacceptable a 2+ year design cycle of Sun is.
On on-die memory controller, Intel will not have it until 2007, the same with glueless CPU interconnect. But Sun will blow most of this time by not having product lines taking advantage of this lead.
I waited and held my shares because of Andreas Bechtolsheim. In his particular case, I do believe the hype. I think the disparity in performance comes in large part from superior design of interconnects and memory systems and etc.
Yes, but he didn't design that part. It is part of every Opteron. So there will be small differnces between different Opteron designs. Notice that Sun was attacking mainly Dell. That's because Dell does not have Opteron. Whenever there was a comparison between Sun and HP, Sun made the comparison between Xeon or Itanium HP product. The differences between HP Opteron and Sun Opteron would be small.
There must (in the sense of "it is imperative that there be") value in selling a machine that Bechtolsheim designed when Dell is selling a Taiwanese connect-the-dots motherboard I can buy at Fry's.
There is less and less of the rocket science stuff for a boxmaker to invent. IBM and HP have spent millions upon millions designing server chipsets (for Xeon and Itanium) to distinguish them from others - namely Dell and other no-R&D companies, that can only have vanilla Intel chipsets. All of this will diminish over time, as most of this logic moves on the CPU die. It is there in Opteron now, it will be in Xeon and Itanium in roughly 2007.
So the differentiation will be stuff that Bechtolsheim has been (extremely slowly) designing - chasis design, management hardware / software, selection of the right system components, competitive cost, good reliability, RAS features. Then comes marketing and customer loyalty.
Competitive CPU is also important, but as you see, Dell still sells a lot of boxes, even though they are not very competitive, leveraging other advantages.
If your assertion is that Bechtolsheim's unique abilities as a designer were confined to cases and power supplies to support a commodity opteron-and-nvidia computer that will be quickly overtaken by future iterations of the Xeon, and you turn out to be right, then I consider my SUNW shares worthless.
Dell makes money on server hardware. Why do you thing Sun will never be able to do it? The era of high margin stuff, like the dot.com era, with expensive Unix boxes is passing. The higher margin x86 stuff is 4 way and up. Sun has the 40z machines, but the 4 way galaxy is delayed to 2006. One thing where Sun can make the big box (because there is no competition) is the planned 8 socket, 16 CPU core Galaxy box. But that too is delayed to 2006, and again, with every end of the delay, some of the opportunity is lost.
Which brings me back to the problem that I see with latest unveilings. While the boxes look good, the overall breadth of Opteron offering falls short of HP. For Sun, Opteron is the main thing, for HP it is a sight business.
Joe |
