I have some algorithmic complexity estimates in Giga Operations Per Second (GOPS) and I would like to compare those with the capabilities of state-of-the-art processors. However, the processor performance is commonly measured in FLOPS.
According to this (Eq.4) FLOPS_core = OPS * FLOPs/operation
Processors without built-in floating point arithmetic usually require say 50 non-floating point operations to implement a floating point operations.
Processors with built-in floating point arithmetic usually have about equivalent or slightly higher non-floating point capabilities, but slightly lower (say five times lower) fixed point, non-integer capabilities.
You should also look at memory bandwidth which can be much more important. And every bit of code is different, so just plugging in numbers into formulas is not very informative. It may give you enough information to decide "it may be worthwhile to try to implement this algorithm on a 1999 supercomputer on my 2021 iPhone" or not.
But you really, really, really need to look at your particular problem. You can't solve this with tables, you solve it by measuring.
If you don't care too much about accuracy, FLOPS and OPS are roughly equivalent.
If you do care, then neither FLOPS nor OPS are precise enough.
