Analysis: With Sun, Intel Wins, but AMD Doesn't Lose

Now that the principal server providers have all adopted the "chipnostic gospel," it will be up to the vendors to demonstrate competitive advantages in their own unique ways. Sun has already demonstrated an ability to do that with its AMD-based servers, as Insight64's Nathan Brookwood explained: "For example, a Sun four-way box expandable to eight-way, has processors mounted on little HyperTransport modules so you can upgrade the processors even as AMD evolves the product, and keep the rest of your system intact. The other guys all basically are saying, 'Well, when AMD switches from DDR1 to DDR2, or DDR2 to DDR3, it's time for a new box.' So if you like Sun's modularity story...that might be your reason to go towards Sun, independent of Solaris."

Sun CEO Jonathan Schwartz hinted all too obviously that his company and Intel are working jointly on a 8P edition of a Sun Xeon-based platform running Solaris. As Schwartz commented, "The single biggest determinant of the success of a high-scale x86 system will be the popularity of the operating system that runs on a single-socket x86 system...No one starts their business on a 16-way system; they all start their businesses and their projects on one-way."

This explains the scalability advantage and the key to holding it: Even if an operating system displays a performance edge in high-processor count servers, it won't be adopted by customers unless that OS is perceived to hold an advantage in "one-way," single-processor servers. Solaris does not hold that advantage yet, but if the Intel deal helps Sun demonstrate what it perceives to be its scalability advantage, that could change.

"I think Solaris has a long history of being highly scalable and highly robust, so those are two characteristics that I think IT managers are fond of," Brookwood stated to BetaNews. "I think Solaris and Windows probably are better at scalability today than Linux. There are a couple of people who have pushed highly scalable Linux systems, notably Silicon Graphics, but they've done it with a bunch of really expensive hardware mods."

Is this because in the broader open-source world where Linux is dominant, developers mainly have access to systems that are already in the marketplace, whereas in the Unix realm, inside players like Sun and Intel have access to systems we don't yet? "No, I think that the people who can take a total view of the operating system and the application environment - like Microsoft can with providing its operating system and SQL Server, or Sun can by providing hardware and a whole bunch of things that run on Solaris - can take a broader view of how to optimize things," Brookwood responded. "The Linux community is really good at optimizing modules of Linux, but there's not anybody who's really driving high-level architecture for Linux the way there is people driving high-level architecture within Microsoft for Windows, and within Sun for Solaris."

Could this Intel alliance signal the beginning of the end of SPARC, Sun's own CPU architecture? Nathan Brookwood couldn't say "No!" enough times, citing announcements just last week of enhancements to its Sun Fire T2000 server with the power-saving Throughput Computing architecture, and of the formalization of the roadmap for its forthcoming "Rock" power-saving processor line, now slated for the second half of 2008.

"For a long time, SPARC sales at Sun were what I would call an installed base play," Brookwood explained. "People who already wandered down that path and got involved with SPARC-based systems, and had some success with them back in the days when there weren't useful industry-standard alternatives, continued to buy SPARC as long as Sun could deliver price/performance-competitive SPARC systems, because it was the path of least resistance. But if you hadn't gone down that path, then why in God's name would you start now? The answer, for the last year, is because those SPARC Throughput Computing configurations can deliver more performance per watt, better system performance, and live within power envelopes that neither Intel nor AMD can match. So as long as Sun continues to focus on things that they can do that the other guys can't do, that means that some customers - not a whole bunch, necessarily, but some customers - who really need the improved power efficiency of SPARC will be motivated to go down that direction even if it means using a proprietary chip architecture."

But those SPARC sales probably aren't jeopardizing Intel's position, nor working against its collaboration with Sun to any measurable extent. Sun can build on its SPARC installed base, if just moderately, while the 70% figure for Solaris installations outside of Sun processors will probably increase.

And while that's good news for Intel, obviously, Nathan Brookwood believes that AMD may stand to benefit from this as well. "I don't believe the impact on AMD is as bad as I've heard some people say today on TV and on Web blogs," he said. "As a matter of fact, I'm sure that if this move makes Sun a stronger x86 server supplier, that's going to benefit both Intel and AMD in terms of its ability to sell boxes based on their chips."

It may not be the "game changing event" that Jon Schwartz made it out to be today, though Sun's new Intel alliance may be its strongest play in several years, in a game where the rules are steadily changing. If there is a new title for this game, perhaps it's "Partner or Perish."