Multicore Chips To Improve Your PC's Speed
Multicore chips are said to be faster than single-core chips. They have the ability to split up computational tasks and run them on several cores at once.
Computer chips are typically slow! They don't seem to be of any solution to get faster processing. This is the reason why chipmakers feel the need to configure additional cores on a single platform in order to avoid congestion in supplying instructions and data.
A typical chip has six or eight cores which communicate over a single bundle of wires, bus. The problem emanates from the fact that only one pair of cores can talk at a time which becomes a serious limitation when we talk of chips with hundreds or even thousands of cores.
Li-Shiuan Peh, associate professor, electrical engineering and computer science, MIT desires to bundle the information carried by these cores into â€śpacketsâ€ť so that they can communicate the same way the networked computers do.
Supposedly, each core would have its own router responsible for sending the packets over the network. Multiple cores are faster as they can run multiple instructions at the same time, increasing overall speed for programs.
The core count on commercial chips has been low leading to extra communication pressure on a 'single bus'. But a scenario change is being hinted by Peh. "Buses have hit a limit. They typically scale to about eight cores," Peh said. Though the 10-core chips which are found in high-end servers frequently add a second bus, but that approach won't work for chips with hundreds of cores.
A technique called 'Virtual Bypassing' has been developed to address these concerns. It will facilitate the router to send an advance signal to the next which would help it to preset its switch, and hence increase the speed of the packet.
â€śVirtual bypassing allowed a very close approach to the maximum data-transmission rates predicted by theoretical analysis,â€ť Peh added.
These findings would soon be presented in Design Automation Conference in the US.