Spec 2000 had the autopar and .art problem, 2006 has libquantum. Other than that they are fairly comprehensive gauges of general purpose CPU performance. In Spec 2000 you took specIn_rate and divided with number of physical cores (and discounted Art scores if they looked dodgy), in 2006 you discount libquantum.
One can argue that Spec isn't representative of modern day client workloads where C spaghetti has been replaced by jitted Java/Objective-C/Swift/C#/JS spaghetti, but it is still 100x times better than Geekbench.
The only integer subtests in Geekbench that aren't microbenchmarks are Dijkstra and LUA. Yet everything is weighted evenly !! 40% of a modern integer workload apparently consists of doing crypto hashes !!
Using Dhrystones would be better (but not good) and more honest.
Cheers
One can argue that Spec isn't representative of modern day client workloads where C spaghetti has been replaced by jitted Java/Objective-C/Swift/C#/JS spaghetti, but it is still 100x times better than Geekbench.
The only integer subtests in Geekbench that aren't microbenchmarks are Dijkstra and LUA. Yet everything is weighted evenly !! 40% of a modern integer workload apparently consists of doing crypto hashes !!
Using Dhrystones would be better (but not good) and more honest.
Cheers