One can only approximate because:
- Different hardware performs differently on integer operations vs floating point operations. Floating point uses different CPU instructions that use different transistor circuits that are optimized differently (or don't exist in hardware at all, see 386SX).
- a FLOPS test has to be optimized to a GPU platform, without taking advantage of hardware specific features it may not get full performance possible,
- Hashing requires optimized programming too, you ask for FLOPS to hashes, but GPU miners have also nearly doubled in efficiency since the first releases from optimized coding.
Core 2 Q6600 = 11.0 Mhash/s & 5 GFLOPS
doesn't compare at all with
5870: 400Mhash/s: