Advanced Micro Devices Inc (AMD)

[j3pflynn]

A little more on Bechtolsheim/Opteron/Kealia

http://nwc.serverpipeline.com/trends/trends_archive/17701364

"InformationWeek: You're taking on a job as chief architect for volume systems at Sun. What will that work entail, and what are your goals?

Bechtolsheim: We hope to bring to Sun an acceleration of its expansion beyond Sparc into Opteron, and into all things Linux. This [Sparc] market has been an unbelievable success for Sun. However, it is not the entire market--there's also x86. It's definitely the Opteron angle that got us involved here.

InformationWeek: How many people worked at Kealia, and what were you working on?

Bechtolsheim: I founded that company in 2001, even before Opteron came out. We spent a year-plus evaluating potential designs. It's a design company; we never shipped a product but had prototypes of Opteron systems.

The Sun thing happened fairly quickly, but the work at Kealia had been going on for several years. We were going to focus on universities and research institutions. But it's really hard for startups to get to major market share. Once Sun came into the market, it became clear that I didn't want to compete with my own former company.

Kealia had 59 employees, including me, all engineers. The only investor was myself. It was a stealth company--the company never made any press releases or announcements. My philosophy is not to talk about products until they're shipped. Sun's a little different.

We'd been working on horizontally scalable systems. On day one, the company worked on media servers, which is a scalable kind of architecture. Everybody will come to Sun; that's the engineering team. So the deal was a function of Sun announcing they'd go in the Opteron direction.

InformationWeek: So how was Kealia's technology different from that of other computer companies that have brought products to market using Opteron chips?

Bechtolsheim: We can't say--the technology will be appearing when the products come to market. Sun has only said it will announce something in the next few quarters.

What I can say is that everything from CPU performance to I/O to networking is being upgraded a lot faster--we'll update these systems every six to nine months.

And the volume opportunity is the interesting angle. The secret to life here is efficiency. We're trying to do a whole lineup of machines here in a way that enables customers to get investment protection and upgrades. It's a whole new design style to make life more efficient in the computer business. Sun's servers have price points from a few thousand to couple of hundred thousand dollars. Not that Sun hasn't tried this in the past, but CPU upgradability is very important, of course. With PCI Express, 10-Gigabit Ethernet, and other technologies coming, there are many things customers have a rightful expectation these systems will support.

InformationWeek: What computing jobs will customers want to buy Sparc systems for then, versus x86-based computers?

Bechtolsheim: Sun dominated with technical workstations in the '90s. More recently, many customers have looked at Linux on x86 as their preferred solution in traditional engineering scientific markets. This is a market where I believe Sun will be very successful with Opteron, but more likely on the Linux front. You could call this a technical early-adopter market."

From another article:
http://www.cbronline.com/print_friendly/b887ed9d8d05924980256eb60032d191
""System design is not about cranking the clock speed any more," explained Fowler. "We can build different memory hierarchies and use different interconnections to make different kinds of servers. We really do have the freedom to do what we want. There are no restrictions on scale or pricing on what Network Systems can do. And let me tell you, we are going to do some things differently. We're going to have some fun, and I promise that we are not going to be boring."

Fowler said that while Sun was happy to be selling the Xeon-based Sun Fire V60z and V65x two-way servers, the company does not plan to roll out a four-way Xeon box. Bringing up the possibility of a Sun-branded Itanium server was laughable, given all the vitriol Sun has spewed about it in recent years. So there is no point in even bringing up the idea, unless something goes terribly wrong with the Opteron line. If anything, Sun will back the 64-bit Xeons if it has to have a fall-back position in the x86 server market. But Xeon is, from now on, going to play third fiddle to Opteron and Sparc, as far as Solaris is concerned"