The boom in Linux clusters is quickly making high-performance computing a do-it-yourself affair, forcing chipmakers to take a more active role in keeping their customer base happy. No longer able to rely solely on hardware vendors as the middleman, Intel and Advanced Micro Devices are taking steps to deal directly with end users of their chips, and each has singled out life sciences as a key focus area in this effort.
The companies are following slightly different strategies: While Intel launched a formal life science group within its high-performance computing business unit just under two months ago, AMD has a more casual, customer-driven approach. However, the two companies hold one thing very much in common: the belief that commodity-based supercomputing is here to stay, particularly in the life science market.
“The opportunity for Intel and for us is quite good there,” said Dave Rich, director of high-performance computing marketing at AMD. “Unlike some other markets where they might have traditionally bought Sun, or HP, or other RISC workstations, a lot of the new biology work is going onto Linux clusters, where we do very well.”
Tim Mattson, worldwide life sciences manager at Intel, agreed that cluster computing has driven chip companies to consider themselves true players in the high-performance computing market: “Two years ago, we had fewer than three systems on the Top500 list and now we have 55. …So I think you’re seeing the front end of an explosion of Linux clusters on this list, and we are the dominant processor in those clusters.” [See p. 6 for a closer look at the recent Top500 rankings.]
But end users buy boxes, not processors. Why target them at all? Why not just sit back and let the Linux cluster wave sell your chips for you? Both Intel and AMD noted that in order to foster the anticipated growth, a better understanding of end-user needs will be crucial.
“You always need to work with the end users to understand who’s interested and why they’re interested. What they need may be a little bit different from what you have,” said Rich. “There’s always more we can do to our products,” added Mattson. Even though Intel is way ahead of its competitors, “I want to make sure that we hang on to this market and we do that by making sure that our customers are delighted, by engaging them and working with them side by side.”
Building an Ecosystem
Rick Hermann, Intel’s manager for worldwide high-performance computing, said that the company made a conscious decision to move away from its passive role in the life science computing market earlier this year. Not content with letting its chips fall where they may, “We made the decision that we needed to become more active in this space, we needed to have collaborations with the thought leadership accounts, we needed to focus on the ISVs, we needed to work with our channel [partners] and fellow travelers to really enable an ecosystem around Intel architecture for life sciences,” he said.
Hermann’s vision for this “ecosystem” is modeled after similar efforts the company has in place to address other vertical markets, such as finance and oil and gas. Intel takes a four-pronged approach that first involves identifying and delivering the processors, interconnects, and other “building blocks” appropriate for the market. The second piece involves collaborating with influential HPC players, such as CERN, NCSA, and the Cornell Theory Center. Third, Intel partners with domain-specific independent software vendors. “In the life sciences we may work with folks like Accelrys and TurboWorx as well as some of the major labs on public domain codes like Blast and Charmm and Amber and Gaussian,” Hermann said. Finally, Intel works closely with its sales channels, from big IT vendors to small systems integrators, to help them design their platforms around its processors.
Hermann credited the recent growth of Linux clusters as a key driver for the company’s success in high-performance computing, but said that the real potential for Intel going forward will be in grid computing. Intel’s HPC group has spent the last two years focusing on clustering, he said, “but as we move forward, it’s about grid, and there’s a natural intersection with the surge that’s occurring around life sciences right now.”
Mattson, who has a background in computational chemistry as well as high-performance computing, has been charged with identifying — and recruiting — key partners in the life science market. Mattson said he expects to have some partners on board by January 2003.
According to Mattson, Intel’s historical strength in floating-point performance presents a key opportunity in the life sciences. “Floating point tends to help you with modeling and that’s extremely important to this market,” he said. In addition, he said, “the problems faced in the life sciences are increasingly depending on addressing large amounts of memory, and that’s where Itanium 2 shines.”
Data storage and, more specifically, rapid transfer of large amounts of data between systems, sets the life science market well apart from other verticals, Mattson pointed out. While other sectors deal with a lot of data, “it’s nowhere near the size and it’s not growing as fast,” he noted.
“We can handle the loads today, but we’re going to have to do some very clever engineering to handle them tomorrow,” Mattson said, a challenge that helped drive the company’s decision to reach out to its user base and ensure its solutions stay in step with their needs.
The New Guard
One advantage for commodity chipmakers in the life sciences, according to AMD’s Rich, is that “it’s not so old as to be completely stuck in its ways — how they do certain things is still forming. So that’s another reason why it’s important to talk to them early to see how they’re doing.”
While AMD expects that the anticipated growth in life science IT will drive its products somewhat, it is really banking on the sector’s tendency to opt for lower-cost computing alternatives. “It’s an evolving market, it’s a growing market, and for products like ours, it’s growing both because the market overall is growing and because of the way that people are doing the work — they’re changing and moving more towards our platform,” Rich said.
AMD has taken a more casual approach to courting the life science market than Intel, however. Rather than a formal business group, Rich said the company’s HPC staff simply keeps its ear to the ground, listening to customer feedback and requests.
“What we’ve been finding in the life sciences is that users need an x86 processor for most of their jobs, and for some jobs they need a bigger memory space to run their applications, Rich said, noting that the company’s upcoming Opteron processor will meet that demand: “It’s an x86 processor that can run all their current applications and can also run in 64-bit mode for the 10 percent that need more memory,” he said.
Rich noted that the company’s recent deal with Cray, which will use 10,000 Opteron processors in the Red Storm supercomputer it is building for Sandia National Laboratory, has helped AMD achieve “good acceptance in the scientific market.”
But don’t look for an AMD account rep to come knocking at your door. The company will continue to gather most of its feedback on the sector through its vendors. “We’ll do more traditionally what a chip vendor would do, which is to run benchmarks and understand the performance of the applications to see if there’s anything that would be helpful,” Rich said. One key to AMD’s strategy, he added, is partnering with top-tier cluster vendors. “Making clusters work is very often in the details, as opposed to the big things, and you want somebody who can bring 100-node systems into a customer and get them running reliably with no hassles to the customer.”
However, that doesn’t mean AMD is turning a blind eye to the finer points of biology. Rich said he’s attended several life science-related conferences to bone up on the sector’s demands. “So far we haven’t discovered anything massive that we need to do besides deliver a good platform,” he said, “but if we do discover something, we will jump right on it.”