A Preliminary Performance Evaluation of the Quadrics Architecture with the DARPA Image Understanding Benchmark

摘要

This work presents a preliminary performance evaluation of the Quadrics SIMD massively parallel computer in the execution of a suite of image processing algorithms. The suite is extracted from the low-level and intermediate-level processing sections of the widely-used DARPA Image Understanding benchmark for parallel computers. Implementation issues are briefly described and data partitioning and mapping is addressed. Measurements carried out on two different machine configurations are reported and compared with the performance of other well-known parallel systems.