Variable latency and bandwidth communication pathways

Abstract

A system and method for choosing communication pathways for data transfers on a computer chip based on desired latency and bandwidth characteristics. On a computer chip including a network of resources, those resources are allocated based upon the needs of the various components of the computer chip. Typical resources on the computer chip include a first bus with a plurality of data lines and control lines and having first bandwidth and latency characteristics, a second bus with a plurality of data lines and control lines having second bandwidth and latency characteristics, and a plurality of devices coupled to the first bus and second bus. Each device includes interface logic for accessing and performing transfers on the first and second buses. Each device is operable to select either the first or second bus depending on desired bandwidth and latency characteristics. Normally the first bandwidth is greater than the second bandwidth. Each device selects the first bus for higher speed transfers or the second bus for lower speed transfers. When the first latency is shorter than the second latency, each of the devices select the first bus for lower latency transfers and the second bus for higher latency transfers. Other characteristics which may be varies by each device according to the transmission needs of the particular device include clock rate, block size, and bus protocol depending upon desired bandwidth and latency characteristics. For highest possible bandwidth transfers, a multiple bus transfer may be requested by any device.