Geekbench is a bit of a bogus benchmark. I really doubt that the Rockchip board would have even quarter the actual performance on "real" programs (e.g. x264, imagemagick, povray etc.)
...none of which are CPU-bound, instead relying on either hardware accelerators or the presence of specific SIMD instructions.
Geekbench is pretty good at measuring how well a chip would perform / scale as a web-app server. (Ignoring IO bandwidth, that is. A benchmark that incorporated both would be brilliant.)
x264 is an h264/AVC encoder that uses the CPU to extract maximum quality for a given bitrate. It does not support any fixed-function hardware on GPUs and the like since that goes contrary to its goals (gpu-accelerated encodes tend to have worse quality for a given bitrate and the fixed-function hardware doesn't support profiles like Hi10p that can improve quality).
Pov-ray is a raytracer. It only runs know he CPU (it has no codepaths to use the GPU).
imagemagick is an command line image manipulation suite that by default uses the CPU to perform operations in images. There is opencl support but it requires either a build option or configuration option for it to be enabled.
My own experience, comparing LAME and flac encoder speeds on an Android device (binaries invoked via a terminal emulator) to a desktop CPU and comparing the actual performance to the geekbench 4 scores showed that gb overrated the Android device by almost 125%.
