I’m trying Stan on a ppc64le (aka POWER8) system and looking for some compilation tips: I see ~3x slowdown against an i7, only half of which explained by CPU freq difference.
I don’t know anything about PPC, but there are a lot of moving parts.
The CPU clock speed is only one component of the CPU. You also need to know how much internal parallelism it supports and how, how it deals with integer and floating point, how it does branch prediction and other optimistic pipelining, and number of registers and cache sizes. Then there’s memory/bus speed itself. And the compilers, which can be optimized heavily for common hardware.