Ryan Whitwam over at Extremetech.com has just published an interesting article comparing the system-level power-management approaches taken by the Nvidia Tegra 3 and Qualcomm Snapdragon 4 ARM-based multicore SoCs. Both of those products have been discussed here at the EDA360 Insider. (See “Processor Wars: NVIDIA reveals a phantom fifth ARM Cortex-A9 processor core in Kal-El mobile processor IC. Guess why it’s there?” and “Qualcomm reveals more Snapdragon 4 SoC details in a White Paper. Want to know what’s inside?”)
Qualcomm and Nvidia are taking different approaches to saving power under varying loads. The Nvidia Tegra 3 (Kal-El) has four main ARM Cortex-A9 processor cores that it can turn on and off and a fifth, low-power ARM Cortex-A9 core that’s implemented with transistors optimized for low-power operation instead of speed. So the Tegra 3 SoC will run the lone, low-power core for low processing loads and will then turn on one to four of the faster processors as the processing load increases. The low-power core is never run in parallel with the four fast cores. Nvidia calls this approach Variable Symmetrical Multi-Processing (vSMP).
Qualcomm’s Snapdragon 4 SoC takes a different approach. The SoC incorporates two Qualcomm-designed Krait processor cores based on the ARM 32-bit architecture. Each Krait processor can be run independently at varying clock frequency and supply voltage—generally called “Dynamic Voltage and Frequency Scaling” (DVFS)—to adjust the performance and power of the processor pair according to the processing load. Qualcomm calls this approach aSMP (asynchronous Symmetric Multi-Processor).
As Whitwam points out in his Extremetech.com article, the current versions of the Android OS assume that available processors are equally able when it makes task assignments. That’s not going to sync up well with Qualcomm’s aSMP approach so I’m guessing that changes to the OS will be forthcoming to accommodate the power-saving approach used by the Snapdragon 4 SoC.
You can read Whitwam’s article here.