Benchmarking ARM boards on WEP-M+2

[Dirk Broer]

I've started a new chapter in benchmarking: I bought myself a power consumption meter, to be able to have the exact wattage that a system runs on.

A Brennenstuhl PM 231 E

So far I had the exact values for runtime, CPU time, and credits and was able to get the exact Per day/proc score. The Credit per KWh however was an estimate based partly on the wattage, which in practice meant the TDP value of the CPU -if known, such as e.g. with ARM boards.
But in the future I will now be able to report the exact wattage of e.g. an ARM system when running Wanless -or any other project/BOINC platform.

The first competition will be between my best 32-bit boards: The Odroid-C1, the Odroid-XU4, and the ASUS Tinkerboard S on one side versus my worst 64-bit boards on the other side: the Raspberry Pi 3B's (both original version, not 3B+). All losers will be retired in favour of additional better performing 64-bit boards. Pretty expensive buggers though: the 8GB nVidia Jetson Orin Nano devkit and the 16GB Radxa Rock 5B are the most likely candidates.

[Dirk Broer]

The somewhat surprising winner of the 32-bit part of this contest is the venerable old Odroid-C1:

Model	SOC	Speed	Credits/day WEP-M+2	Watts needed	Credit per kWH
Odroid-C1	Amlogic-S805 (Quad-core Cortex-A5)	1500 MHz	600	3.8	6,500
ASUS Tinker Board S	Rockchip RK3288 (Quad-core Cortex-A17)	1800 MHz	960	9.3	4,300
Odroid-XU4	Samsung Exynos 5422 (Octa-core Cortex-A15/Cortex-A7)	2000/1400 MHz	1,670	13.3	5,300

And these three were the top-performing 32-bit boards. The C1 really clobbered the Cortex-A7 equipped Raspberry Pi-2's and Banana Pi M2, and also was better than the Cortex-A9 equipped Hummingboard, that need more power to achieve lower scores.