Intel unveils mysterious Larrabee architecture

Larrabee will scale to trillions of floating point operations per second

Intel has entered the product planning phase for a new chip architecture codenamed Larrabee, the company said at its Intel Developer Forum conference in Beijing. 

Larrabee products will offer highly parallel chips based on existing Intel architecture and will support existing software tools. It will be designed to scale to trillions of floating point operations per second.

The chip will include enhancements to accelerate applications such as scientific computing, recognition, mining, synthesis, visualisation, financial analysis and healthcare applications.

The announcement was notably short on technical details, however. Intel declined to say when it expects to ship the first products, where Larrabee fits in its product roadmap or what products it replaces.

"We are not giving details of specific products," Sean Maloney, general manager at Intel's sales and marketing group, said in a briefing with reporters on Monday.

"We are announcing the product planning phase. And we are saying that we are using the technology to address older applications in new ways."

The Larrabee codename has circulated on hardware enthusiast websites for the past 11 months, and has been identified as a high-end graphics processor.

vnunet.com's sister publication The Inquirer has described the chip as a processor with multiple x86 'mini-cores' and called it a "central graphic processor unit". 

The notion of using a graphics processor to accelerate floating point intensive applications in areas such as scientific computing is not new.

AMD started shipping its Stream processor last November dedicated to the scientific computing market.