Future Chip Multiprocessors won't just have to contend with the of long latency of
main memory access and with keeping the power budget of the processor in check,
but will also face bandwidth limitations because pin-bandwidth and the number of
pins won't be able to keep up with the growth in the number of processor cores.
Optical interconnects have been proposed as a method to ease this future bandwidth
bottleneck. Here we calculate the bandwidth offered by future electrical and optical
interconnects based on the SIA roadmap and conservative extrapolation of the ca-
pabilities offered by array's of VCSEL's in the given timeframe 2008-2014. We run
programs from the Spec2000 and Mediabench suites on an execution-driven simu-
lator showing that programs which are strongly dependent on memory bandwidth
can take advantage of an optical interconnect and see increases in performance of
up to 340%.