Parallella GFLOPS compared to other ARM boards

[openbluefish]

I am interesting in buying a parallella but I'm not sure how powerful and useful it is. Now if I was just interested in solving PDEs I would go out and get a GTX Titan Z for the most raw power. I understand ARM boards are not the most powerful but use very little electricity compared to others. I'm reading that the $150 board has 24 GFLOPS. The RaspberryPi's GPU alone has 24 GFLOPS. However that seems to be a nightmare to program for because there is no debugger for the Broadcom VPU. The Mali-T628 MP6 found in the $100 ODROID XU3 lite has 102 GFLOPS with OpenCL 1.1 full profile support in Linux. The $192 nVidia Jetson TK1 has 192 CUDA cores with about 300 GFLOPS with CUDA support.

My main question is what does the parallella provide that is different that other ARM boards? It seems more open than the other options. Also is seems like a great tool to learn on. But still it seems to be lacking power compared to the OpenCL and CUDA options. Is the parallella something much different than GPGPU making it more unique? Is part of the appeal that in the future the Epiphany will have many more cores? I'm interesting in it just leerious about its computational power. Any input would be greatly appreciated.

[aolofsson]

GFLOPS is only ONE of many important metrics...

Are you interested in programming or in a computing work horse?
Are you comparing assembly optimized hardware vendor provided libraries or code you have written yourself?
What programming language are you interested in?
What really matters is wall time on a complete specific benchmark. Which ones?

[xilman]