New EEMBC benchmark scores for the 208-MHz LPC3180 microcontroller from NXP Semiconductors are the first to publicly demonstrate detailed performance and energy tradeoffs using the EnergyBench power/energy metric on a microcontroller, and the first for any device to directly show the effect of an integrated floating point co-processor and instruction cache on device performance and energy consumption.
For the certified scores published today, the ARM9-based LPC3180 MCU was pitted against performance benchmarks taken from the EEMBC AutoBench automotive/industrial suite. The suite, which addresses a wide range of general-purpose embedded applications, was being tested simultaneously for power and energy consumption using EnergyBench, a metric that EEMBC implements with a National Instruments data acquisition card and LabVIEW interface.
Four separate benchmark runs were performed to produce scores that isolate the contribution of the LPC3180′s floating point co-processor and instruction cache — together and individually — to device performance, average power consumption, and energy consumption.
“These NXP LPC3180 scores are highly significant not only for the novelty of the test conditions, but also insofar as they provide a clear validation of EEMBC’s approach to power and energy consumption measurement,” said EEMBC President Markus Levy.
As expected, when the floating point and instruction cache functions are enabled, the EEMBC score data for the LPC3180 microcontroller shows an increase in performance. Furthermore, the absolute energy required to execute the performance benchmarks decreases by as much as an order of magnitude under the same conditions. However, it was unexpected that the average power also decreased, despite enabling the floating point and instruction cache functions. In addition, average power also varied by as much as 11 percent based on the benchmark workload that was applied.
“The results for the LPC3180 microcontroller underscore that power and energy are two quite different concepts,” Levy added. “They also show that ‘typical power,’ a number often provided on CPU datasheets, is a poor indicator of how much energy a device will actually require to perform a specific workload. The take-away for design engineers is the importance of looking at the whole picture when considering tradeoffs between performance, power, and energy.”
NXP Semiconductors’ 32-bit LPC3180 is the industry’s first 90-nm ARM9-based microcontroller and the first such device to provide a vector floating-point co-processor, integrated USB On-The-Go, and operation down to 0.9 V. Its target applications include point-of-sale (POS) equipment, medical and industrial devices, global positioning systems (GPS), and robotics. Additional LPC3180 peripherals include 7 UARTs, SPI, I2C, a real-time clock with a separate power domain, NAND Flash, and DDR memory controllers.
“The EEMBC benchmarks allow us to credibly demonstrate the real value of the performance and energy-related features of our LPC3180 microcontroller,” said Geoff Lees, general manager of the Microcontroller Product Line at NXP Semiconductors. “Furthermore, as shown from the benchmark data that we’ve published, we believe our customers benefit from seeing a wide range of performance and energy values to enable them to make more educated decisions in their designs.”
The certified EEMBC performance and EnergyBench scores for the NXP LPC3180 microcontroller provide data on performance and energy consumption for 11 benchmark kernels from the AutoBench suite. The wide range of applications in the suite include angle-to-time conversion, basic floating point, bit manipulation, CAN, infinite impulse response filter, inverse discrete cosine transform, pointer chasing, pulse width modulation, road speed calculation, and table lookup and interpolation.
AutoBench and EnergyBench are trademarks and EEMBC is a registered trademark of the Embedded Microprocessor Benchmark Consortium.