Home > Hardware

Computerworld - . . . halting its supercomputing slide. Five years ago, Sun Microsystems Inc. had 92 systems among the 500 fastest supercomputers worldwide, and its speediest machine was ranked 44th, according to the annual Top500 rankings. This year, Sun managed

$7B HPC market in 2004, says IDC

to place just four machines on the list, with its fastest cluster topping out in 141st place. Even little-known Atipa Technologies in Lawrence, Kan., was able to best Sun this year by getting five of its supercomputers on the prestigious list. John Fowler, executive vice president of Sun's network systems group, has been tasked with reversing the company's plunging fortunes in the high-performance computing market. One of Fowler's first projects was this month's unveiling of a supercomputing lab called the Sun HPC Solutions Center in Hillsboro, Ore. Fowler says that with about 10 days' notice, the center, which has its own 8.6-megawatt power plant to handle the massive electrical needs of supercomputers, can put 1 trillion floating-point operations per second (TFLOPS) of computing power at the fingertips of customers who want to test their applications. As almost any supercomputer techie will tell you, configuring an HPC cluster isn't trivial, but Fowler claims that his lab will turn it into "a routine activity, which will be dramatic in this space." The lab can currently handle the testing needs for applications requiring performance of up to 10 TFLOPS, and Fowler says 40 TFLOPS capacity isn't far into the future. Interestingly, he acknowledges that most of the test demands are coming from users who want Sun to deliver Linux clusters - not Solaris ones.
Speech-recognition apps behave . . .
. . . properly when listening. That's the trick Amit Desai says he has taught voice-recognition software from Voxify Inc. in Alameda, Calif. Desai, the company's chief technology officer, admits that heretofore, "speech recognition has talked a good game but not delivered." However, he claims that Voxify's Automated Agents technology is well schooled, not just in quickly matching sounds to words but in exhibiting good listening behavior as well. Desai says complex human conversations include cultural "affirmative behavior"

Desai teaches voice-recognition software to behave better.

traits, such as unconsciously muttering "uh-huh" in response to a statement. Traditional speech-recognition technology gets confused by such chatter, which can gum up human-to-machine interactions, he adds. So if voice technology is going to get an expanded role in self-service business applications, it has to adapt to what people utter. Desai says the Conversation Engine software used by Voxify's agents groks people's behavior and knows