John,

I have seen here many people seem to be interested in the ARM architecture. As it's power tends to reach low-end intel products I believe many people will start using it for cloud environments.
For each new technology the problem of a fair comparison arises and I believe the need for benchmarking will grow for ARM architectures.

I have been benchmarking this architecture for some time now and I'd like to point out the different Linux ABIs available for ARM and the different instruction sets make a great difference in the benchmark results.

Typically you will have 3 kinds of ARM processors :
- armv5 based : these are low end ARM processors with typically no FPU, will use armel Linux ABI
- armv7 based : these are high end ARM processors (Cortex-A8/A9/A15) with "almost mandatory" FPU, will use armhf Linux ABI
- armv8 based : these are next gen, 64bits, not yet production ready

When compiling Geekbench for ARM I'd suggest to build 2 different releases :
- armel toolchain : -march=armv5te (the low end processors usually use thumb (16 bits) instructions because of their low memory bandwidth, hence the 't')
- armhf toolchain : -march=armv7 -mfpu=neon

This would reflect the use case of these processors.

What do you think about this ?

Cheers,