The new chip design presents a new approach to communication between a computer’s cores, or processors.
“This approach, called the core-to-core communication acceleration framework, improves communication performance by two to 12 times,” said Yan Solihin, an N.C. State professor and co-author of the study, in a press release. “In other words, the execution times — from start to finish — are twice as fast or faster.”
The study’s researchers said when it comes to dealing with large amounts of computational data, the traditional way of communicating between cores is inefficient.
Michael Kowolenko, a researcher at N.C. State, said computers are often slowed down by the amount of information they need to process and the chip can improve processing speeds.
“So if you can increase the efficiency of how you put information in and out of a chip by decreasing the path length, or opening up the freeway so to speak, the machine can go much faster,” he said.
The chip’s design — which includes new built-in hardware — replaces a slower software-based method for communication between cores.
The new hardware will be more efficient in communicating information between cores, which will increase processing speeds for the computer.
At the center of the chip’s design is a queue management device that performs the task of managing communication between cores without the need for time-consuming software instructions.
The queue management device can be used to collect data from multiple cores, which can speed up basic computational functions by as much as 15 percent.
Intel Corporation selected N.C. State for this project because a Ph.D. student from the university joined Intel Labs as an intern last year and worked on core-to-core communication, Stephanie Matthew, a spokesperson for Intel Corporation, said in an email.
She said the new chip design presents exciting opportunities for future research.
Solihin said he is looking into developing other on-chip devices that would accelerate multi-core functions.
Kowolenko said he is excited about the researchers’ progress and what it means for his field of study.
“I think it’s really addressing one of the fundamental problems we’re having in trying to design an infrastructure that can support the data manipulations they want to do,” Kowolenko said.