> Speaks to a fundamental misunderstanding of what clock rates are, what ARM extensions > are, and why benchmarks are of precisely fuck-all relevance to how an actual > application that relies almost purely on single-threaded integer performance is going > to bench on something that is scarcely 23% faster in reality than whatever artificial > benchmarks you've ginned up.
In fairness, the original Pi was ARMv6 ISA, so v7 does gain you some perf at the same clock vs. v6. That said, phones/tablets are on either v8 (Apple A8/A8X, Nvidia's "Denver") or v7 with instructions tuned to use fewer cycles (Apple A7, recent Qualcomm Snapdragons) so RPi 2 is now where phones/tablets were 2 years ago, much as the original RPi was when it was introduced.
A 1.3 GHz ARMv7 (original Nexus 7, Nvidia Tegra 3 SoC) with modern MAME works decently for games up to around a Space Harrier level of complexity (dual 68ks + Z80 + FM sound), which is where things break up. CPS2 / NeoGeo are fine. But deduct 400 MHz and I suspect things will get a lot less fine.
And by the way, there's ultimately no escape from reality as preached by MAMEdev. Dolphin recently fixed their sound emulation once and for all. But as part of that, they had to make it break up when the emulation overall isn't running at the full 59.97 FPS. Just like in MAME.
(My real terror? The presence of Windows 10 and Office on the Pi2 means a *lot* of people with no business owning one will be buying them).
|