November 16, 2003 12:01 PM PST
Intel moves deeper into supercomputing
The Santa Clara, Calif.-based company will be a visible presence this week at the SC2003 supercomputing conference taking place in Phoenix.
Along with the announcement of the research effort, called the Advanced Computing Program, Intel will discuss how Lawrence Livermore National Laboratory is erecting a cluster, code-named Thunder, that will contain 3,840 1.4GHz, 4MB Itanium 2 processors and is expected to be one of the fastest computers in the world. It is slated to be completed in December.
At 20 teraflops, or 20 trillion floating-point calculations per second, Thunder would be the second-fastest supercomputer in the world if it were running now, following NEC's Earth Simulator, said Rick Herrmann, high-performance computing program office manager at Intel.
Intel-based computers account for four of the top 10 computers on the latest Top500 supercomputing list, which was released Sunday, and for 189 computers total on the list. A year ago, only 56 computers on the list ran on Intel chips.
Call it the democratization of the supercomputer. Years ago, the market was a province of huge multinationals such as IBM or specialists such as Cray because supercomputers were largely customized beasts that often required years to build.
Now, because of improved clustering, which allows researchers to stitch hundreds or thousands of commercially available servers into a functional supercomputer, and ever-increasing processor performance, a variety of companies are now making supercomputer-class machines.
The basic building block of the Thunder cluster, for instance, is a four-processor, Itanium 2 server from California Digital. The machine is expected to take only five months to build from concept to final installation. A cluster of Apple Computer desktops put together by Virginia Polytechnic Institute and State University ranks as the third-fastest computer on the most recent list.
"We don't think it is an and/or equation. There is a place for customized solutions, but there is also a need for off-the-shelf systems," Herrmann said. "The performance of computers with off-the-shelf components intersected with specialized machines in the 2001-2002 time frame."
To continue to increase the popularity of these types of systems, Intel will use the $36 million in the Advanced Computing Program to iron out some of the lingering technological limitations found in large computers fashioned from off-the-shelf parts.
The research effort, which will involve server makers and universities, will revolve around such issues as increasing parallelism in applications, reducing energy consumption and improving system management.
"We're looking at three to five years from now (to see) what needs to be solved," Herrmann said.
In other Intel server news, the company this week said momentum for its Itanium chip continues to increase. "We?ve gone from a handful of apps to 1,000," said Ajay Malhotra, general manager of enterprise marketing and planning at the company. Around 1,500 applications will be available, or in development, by the middle of next year.
Malhotra also added that Montecito, a dual-core Itanium expected in 2005, will have 24MB of level 3 cache and contain multithreading, a feature now found on Pentium 4 and Xeon chips.