The benefits of the perception processor architecture are tested on 10 benchmarks that were chosen both for their perceived importance in future embedded systems as well as for their algorithmic variety. In order to compare the approach to the the competition, four different implementations of benchmarks are considered:
- Software running on a 400 MHz Intel XScale processor. The XScale represents an energy efficient embedded processor.
- Software running on a 2.4 GHz Intel Pentium 4 processor. The Pentium 4 is optimized for performance rather than energy efficiency since more efficient processors can not currently support real-time perception tasks such as speech recognition.
- A microcode implementation running on the perception processor.
- Four of the benchmarks have been implemented as custom ASICs since ASICs represent a high level of performance and energy efficiency that general purpose processors are seldom able to match.
- 10.1 Benchmarks
- 10.2 Metrics
- 10.3 Experimental Method
- 10.4 Results
- 10.4.1 Instruction Level Parallelism
- 10.4.2 Power Consumption
- 10.4.3 Throughput
- 10.4.4 Energy Consumption
- 10.4.5 Energy Delay Product
- 10.4.6 Energy Delay Squared Product
- 10.4.7 Clock Gating
- 10.4.8 The Cost of Generality
- 10.5 Summary