next up previous [pdf]

Next: Acknowledgements Up: Clapp et al.: Hardware Previous: Discussion

Conclusions

The variations in approaches for implementing an optimized version of RTM on CPU, GPGPU, and FPGA archictectures make direct comparison infeasible. Finding the platform that provides the optimal RTM performance involves balancing price, compute and memory potential, programmability, and the set of algorithmic approaches that are feasible on a given platform. Understanding the variation in underlying architecture and the resulting optimal programming approach can lead to new insights on how to produce optimal code.




2009-10-16