Can CPUMark values be compared across different platforms? Ie. does a value of say, 1000 CPU Mark on Windows, Android, and iOS mean that the three cpus are has about the same performance?
Our original plan was to have the CPUMark directly comparable across multiple platforms.
But it is harder than it sounds.
There are differences in programming languages, compilers, how much RAM is available, how multi-threading is done, what functions are available in each different operating system and what 3rd party libraries are available. For example we saw huge differences in encryption performance depending on if hardware accelerated AES was available and exposed via the O/S API.
For example on Android, all the code is written in Java, which tends to be slower than the C++ and assembler code used in Windows. There are also compiler differences and some differences forced upon us in the actual test code. For example on Windows we have access to much more RAM than on iOS and Android. In Windows we also have access to the faster SIMD instructions.
In additional we are about to release PerformanceTest V8 on Windows. And some the CPU tests have changed in PT8.
So as a result we aren't really recommending direct comparison of the CPUMark value between Windows and mobile platforms although some of the individual values are broadly comparable (see below).
We have had a couple of people ask similar questions, so here are a couple of threads with related information.
Comparing the CPUMark of Phones and Desktop CPUs in 2011
Comparison of mobile CPUs and Desktop CPUs in 2014