Amdahl’s Law is descriptive, not exculpatory. It explains why a workload with a large serial/contended fraction won’t scale. It does not prove that the worklo

Amdahl’s Law is descriptive, not exculpatory. It explains why a workload with a large serial/contended fraction won’t scale. It does not prove that the workload’s serial fraction is representative of the category it claims to stand in for. So when a benchmark’s “text processing” test over ~190 files barely gets past ~1.3x on 8 cores, that’s not some profound demonstration that CPUs can’t parallelize text work. It’s mostly a demonstration that this benchmark’s implementation has a very large serial bottleneck. That would be fine if people treated GB6 multicore as a narrow benchmark of specific shared-task client workloads. The problem is that it is labelled as a general multicore CPU metric, and is used as such, including for 18-core vs 96-core comparisons. That’s the misuse being criticized. TL;DR: Amdahl’s Law explains the ceiling; it does not justify treating an avoidably low ceiling as a general measure of multicore CPU capability. EDIT: Also, submitter, I'm not sure why parent is upset that you submitted. Thanks for sharing. I've been wondering for years why GeekBench was obviously broken on multicore. (comes up a lot in Apple discussions, as you know)