(05-09-2018, 02:51 AM)JonnyH Wrote: Remember that geekbench doesn't test *exactly* the same things as the dolphin JIT generates (which is why I dislike pc "enthusiasts" bandying around "IPC" as a single magic number to compare - IPC doing *what* is important - there can be big changes in relative performance doing different things and running different instruction streams. To be honest, the entire goal of using a single number to compare will always be nonsense, as different designs will be better/worse at different things).
And the dolphin JIT happens to like really wide frontends and reorder buffers, which means that the big x86_64 CPU designs (like and post-p4 Intel and AMD's ryzen) tend to overperform what you'd expect from just comparing the geekbench scores (compared to most arm designs at least)
Would that mean that Samsung’s M3 core is a significant improvement above and beyond what the numbers suggest? Anandtech’s analysis revealed it to be an extremely wide design, very Apple-esque custom cores. Obviously it’s more complicated in real world, but anandtech’s further testing revealed that the chip itself is speedy when the DVFS is fixed.
