To compare performance of the boards I used the nbench benchmark tool. This is an old tool that was originally created to measure the performance of early Pentium-class computers. When nbench runs it performs a series of tests which are meant to mimic real world workloads, like compressing data or training a neural network. Each test result is combined to build a score for the memory, integer, and floating point performance of each system.

For this test I compiled nbench for each board using GCC 4.7. I chose a minimum of optimization settings as the goal was to compare each board and not necessarily produce the fastest results.

Looking at the results, the Beaglebone Black has the strongest memory and integer performance. However the Beaglebone Black's floating point performance is slightly behind the Raspberry Pi. This can be explained because the ARM Cortex-A8 processor on the Beaglebone Black has a 'VFPLite' floating point unit which isn't as fast as other ARM FPUs. If you only care about performance and don't have a floating point heavy workload, the Beaglebone Black is a good board to consider.

Another interesting comparison is the Arduino Yun and Intel Galileo. Both boards run at the same 400mhz clock speed, but it's apparent the MIPS architecture of the Yun has slightly more performance than the Intel architecture of the Galileo. Floating point performance on the Yun is very low because it doesn't have a hardware floating point unit and must run those operations in software.

Finally, both the Raspberry Pi model A and model B are almost identical in performance. This result is expected because there is no difference between the processor on either board. The model B board only has more memory and peripherals compared to the model A. From a performance per cost standpoint, the model A board at $25 is quite impressive.