The system, to be unveiled at the ClusterWorld conference in San Jose, California, is a dual-processor, rack-mounted server measuring 1.75 inches thick, an IBM representative said. It is expected to ship in the second half of 2003. The machine, which IBM first discussed in April, is designed to be clustered in large numbers to form a single, powerful computational engine.
Clustered systems are increasingly popular with customers that have intense computing requirements, such as pharmaceutical designers, automakers and banks. Intel's Xeon processor is widely used for these systems, but rival Advanced Micro Devices is working to find a place for its new Opteron chip as well.
"For those segments that have already embraced AMD, the enthusiasm for Opteron is very, very high," said Dave Turek, leader of IBM's new "Deep Computing" team, pointing to government laboratories in particular. "We have substantial pull from customers."
AMD's Opteron processor runs the same software as do Intel's Xeon and Pentium products, but it also has 64-bit extensions that allow it to speed some types of calculations and to accommodate more memory. To take advantage of those extensions, software must be rebuilt for the chip--an Opteron version of Linux from SuSE is available, for example.
One difference between IBM's high-performance computing business and that of its competitors is its program to rent out its own supercomputing centres. This program enables customers to tap into the centres when they have more number-crunching to do than their in-house machines can handle. The first customer for the rental service is Petroleum Geo-Services. On Tuesday, IBM is expected to announce another customer: GX Technology of Houston.
Big Blue's supercomputing work has led the company to trail niche leader Hewlett-Packard by a single system on the "Top500" list of the world's 500 fastest machines. To accelerate its work, the company formed an expanded supercomputing team--named Deep Computing--in April and last week hired longtime IDC supercomputing analyst Debra Goldfarb to be vice president of strategy and products.
While the Top500 list highlights cutting-edge supercomputing designs, most of the market is for less-exotic systems, Goldfarb said in an interview Thursday.
Buyers spent less than US$1 billion last year on super-high-end "capability" systems built to accomplish specific goals such as modeling climate change or predicting nuclear weapons' physics, she said. In contrast, about $5.5 billion was spent on more standard systems, typically close relatives of servers sold for ordinary business use.
Cluster supercomputers typically use ordinary lower-end servers that run the Linux operating system. However, a Dell Computer cluster at the Cornell Theory Center that runs Microsoft's Windows OS has appeared on the Top500 list.
One cluster of Opteron-based machines, code-named Red Storm, is slated to be created by supercomputing specialist Cray for Sandia National Laboratories in the United States. On Monday, SuSE announced that Cray had selected its version of Linux to run on the machine. Under the deal, the German Linux seller will provide developer support as well as its operating system.
For its part, chipmaker Intel hopes to use the cluster market to further its comparatively new Itanium family of 64-bit processors, which compete with AMD's Opteron and established 64-bit chip lines from IBM and Sun Microsystems. On Monday, Intel gave details of a new version of its upcoming Itanium 2 6M "Madison" chip. This version will be tailored to dual-processor machines, which are well-suited to clusters.
Compared with the mainstream Madison chip--expected to debut on June 30--the dual-processor Madison will have less high-speed cache memory, according to company spokeswoman Barbara Grimes. The new chip brings to four the number of Madison varieties that Intel plans.
In addition to IBM, specialists in cluster computing such as RLX Technologies, TopSpin Communications and Platform Computing are expected to highlight advances in their own wares at ClusterWorld.
