Sun Microsystems (SUNW) co-founder Andy Bechtolsheim was struck by déjà vu as he surveyed the computer company's new high-performance testing center in Hillsboro, Ore. For Bechtolsheim -- who returned to Sun in 2004 after an almost decade-long absence -- seeing row after row of humming, flashing, ultra-powerful servers was like going back to Sun's roots, when the company specialized in high-priced workstations for universities and research institutions.
Sun's star in the high-performance computing universe waned as big clients relied on clusters of cheaper, low-end PCs to handle the computations and simulations previously assigned to workstations. Bechtolsheim left the company in 1995 to found several other tech outfits, but he returned last year when Kealia, a server designer he founded, was snapped up by Sun.
BUBBLE BURST. Sun's also making a return, of sorts, to an earlier era. It's putting a renewed emphasis on so-called high-performance computing, where multiple servers are linked to do simulations or complex calculations. It's part of an attempt to revive revenue growth that has been hobbled by the burst of the dot-com bubble and later decline of many of the Internet companies and other outfits that once flocked to Sun's products.
The company is still struggling to find growth and to attain consistent profitability (see BW Online, 6/3/05, "Sun's Surprising Tale of the Tape"). In the first quarter of fiscal 2006, which ended on Sept. 25, Sun's loss narrowed to $123 million on $2.7 billion in revenues, a 3.7% rise from a year earlier.
Since revamping its high-performance computing efforts in 2004, Sun is gaining traction. Auto makers are using Sun's gear to simulate car crashes. Energy companies have relied on the clusters' brain power to figure out whether a field contains oil -- without having to drill a well.
AFTERMATH. Customers are increasingly buying Sun's servers containing the popular Opteron line of chips made by Advanced Micro Devices (AMD), which form the backbone of several high-performance computing applications, says Bechtolsheim. In the first quarter, the Opteron-based Sun Fire server business more than doubled, in terms of units sold, from a year earlier, Sun said in October.
The market for high-performance computing represents at least a $1 billion opportunity for the company, says John Fowler, executive vice-president for network systems at Sun. The total worldwide market for such products should eventually reach $16 billion a year, according to Sun estimates.
Fowler and Bechtolsheim spoke with BusinessWeek Online reporter Olga Kharif on Nov. 10. They discussed Sun Microsystems' strategy in high-performance computing; why Sun favors AMD's chips over those made by Intel (INTC); and the aftermath of Sun's decision to make available to the public, or "open-source," its Solaris software. Edited excerpts of the conversation follow.
What's the significance of opening this high-performance computing center? Fowler: It's one of several steps we are making to change the high-performance computing landscape. Until now, much of high-performance computing has been custom, unique for each customer, involving a lot of special services and special technology. What we are very much focused on is making our high-performance computing a low-risk activity for customers.
We think there's a huge opportunity around being able to take high-performance computing into your company without having to do a lot of custom engineering and custom technology -- and, therefore, use it with low risk. Most high-performance computing centers are actually production centers, so it's difficult for customers to go try things.
This center is specifically built around learning, around teaching, around trials. It's not built around production. To my knowledge, it's unique in terms of the scale of what we're doing here and what customers can do. [Because customers can essentially see how their systems will work here], we recently shipped a customer a system that they deployed within 16 hours. That used to take weeks, if not a month.
Do you have any customer wins yet? Fowler: We have a collection of wins already. In fact, this particular center has played a major role in a large win we can't announce publicly yet, but we will be announcing publicly in the next couple of weeks.
Why is Sun well positioned to go after this market now? Bechtolsheim: With AMD's Opteron chip, we actually do have the best cost performance in the industry. I'm not sure we've said this publicly...but the majority of our Opteron sales so far have actually gone to technical and high-performance applications. A lot of customers we see want the support that Sun can offer and the additional services Sun can provide, vs. buying boxes in China and putting them together themselves. These are typically large installations, with hundreds and thousands of server nodes, and the reliability of these systems really matters.
How big is the high-performance computing market? Bechtolsheim: Researcher IDC looked just at the traditional high-performance computing markets, and those added up to between $7 billion and $8 billion a year. If you then add other technical applications, you get close to double that number. That market will grow at least 8% per year. It's a significant growth market.
What makes 8% growth significant? Bechtolsheim: Considering that unit costs per memory, per processors that go into servers, are declining, the fact that this market is growing is significant. People are buying more and more units, building new data centers. I was surprised by the number of customers looking to buy 1,000 server nodes. That used to be truly the exotic, very, very high end of the market. But as the costs have come down, suddenly 1,000 nodes is only a few million dollars.
What's driving this market's growth? Fowler: People have done improvements to applications used for design, such as Pro/ENGINEER and Catia. Compared to 5 or 10 years ago, the applications are a lot more sophisticated, so you can proceed much farther in your design, using just computer-based simulation. It's changing the way companies spend money. They are spending a lot less money drilling wells for fun, to see if they strike oil, as in the old days, and more of their time simulating it from data.
Bechtolsheim: Also, the impetus for people to buy high-performance now is that they can get better performance for the money. The cost performance of the Intel chips didn't improve as quickly as AMD's. People are shortening the replacement cycles of their installations, because chip improvement has accelerated.
Are you determined to stick with AMD instead of Intel for a while? Bechtolsheim: Our goal in life is to deliver the best capabilities to the customer. So if Intel had the fastest chip, we would use an Intel chip. But [it doesn't].
Today, AMD's chip is almost twice as fast as Intel's. There's really no comparison. And it's more energy efficient. We're obviously familiar with AMD's product road map, which we can't discuss in public, but we like what we see.
You recently began shipping your new Galaxy servers. What's their role in the high-performance computing space? Fowler: We ship with the fastest AMD processors available. In fact, we ship versions of AMD processors that no one else has been technically able to ship, because we designed Galaxy around those. So compared with other vendors using Opteron, we are at least 15% faster. We've also designed for greater power efficiency.
Where do you see Intel's Itanium processor in the context of high-performance computing? Bechtolsheim: If you look at the market reality, the shipments of Itanium are so low, anybody would ask why they're even putting it out. Actual shipment rates last quarter were, like, 6,000 systems. That compares to AMD Opteron, shipping close to 10 million shipments a year. Itanium was always more expensive, and it wasn't faster. The only place Itanium is left today is Hewlett-Packard's (HPQ) systems, which are very expensive.
Fowler: One of Itanium's problems is software support. Intel's operating systems and application programs had to be [modified] to work on Itanium, and that adds millions -- if not billions -- of dollars in expense. But AMD took an existing architecture, and it still runs all the existing applications. For the customers, the choice is easy. At the end of the day, every customer is going to choose a system that runs the greatest available software, and that's going to help drive volume.
When we decided to extend our line beyond SPARC, we evaluated Intel's Itanium, Xeon, and Opteron. Itanium was never considered seriously, because we figured we could never make any kind of market of that [because of] the software ecosystem and cost.
You mentioned software. What effect are you seeing from your decision to open-source Solaris earlier this year? Fowler: We have over 1,000 software partners now that are on Solaris. There are two companies that have already sprung up that are taking the source code and producing alternative distributions, like you would see in the world of Linux. And there are a number of people pursuing different business models around embedding Solaris into various devices.
Why are they interested in that when there's already Linux? Fowler: First, Solaris has a 20-year history of features and reliability and security that Linux can't match. Second, you can modify Solaris, and you're not required to expose those modifications to the world. With Linux, you have to give these modifications back. Plus, we grant patents and patent protections with our license; [with] Linux, you're not protected in case of any patent transgressions.
Finally, we are a very well-funded company that stands behind the product. We have already given out 3 million licenses on Solaris 10 operating system. We see an enormous uptake of people using Solaris.
And what are the benefits of that to Sun? Fowler: This enabled us to move away from the perception that we're a proprietary-systems company. The radical changes that we made completely alter how you look at Sun. You can no longer look at Sun as proprietary-source SPARC. Our systems run Windows, Linux, you name it.