PassMark single threaded test is a very simple synthetic benchmark according to their docs:
> The single threaded test is an aggregate of the floating point, string sorting and data compression tests.
It’s an impressive score, but there’s more to performance in real apps than these three simple tests. This is a reflection of the high burst clock speed of the chip and the great job they did keeping it fed, but PassMark single threaded is about the simplest measurement you can do.
Modern CPUs aren’t really optimized for peak floating point throughput because any GPU will do a much better job at that. It’s also not clear if they actually use SIMD features of the chips, where desktop and server class parts should be able to walk away from the Apple phone chip due to their higher power limit alone.
> This is a reflection of the high burst clock speed of the chip
Apple's in house silicon strategy has long been to create wide cores that perform more instructions per clock cycle and then run the cores at a conservative clock speed for power savings.
Running up the clock speed for performance no matter what effect that has on power consumption and heat has been what we've seen from Qualcomm/Intel/AMD/Nvidia.
In Geekbench multicore:
> the A19 Pro delivered a powerful multi-core score of 11,054... while using a remarkably low 12.1W of power... To put that into perspective, other flagship chips from the Android camp had to push their power consumption much higher to even come close to its scores. The Snapdragon 8 Elite, for instance, had to use 17W of power to complete the same benchmark. Meanwhile, the Dimensity 9400 consumed a staggering 18.4W.
> The single threaded test is an aggregate of the floating point, string sorting and data compression tests.
It’s an impressive score, but there’s more to performance in real apps than these three simple tests. This is a reflection of the high burst clock speed of the chip and the great job they did keeping it fed, but PassMark single threaded is about the simplest measurement you can do.
Modern CPUs aren’t really optimized for peak floating point throughput because any GPU will do a much better job at that. It’s also not clear if they actually use SIMD features of the chips, where desktop and server class parts should be able to walk away from the Apple phone chip due to their higher power limit alone.