[PragmaticPulp]

Some synthetic benchmarks are so simple that they almost reduce to a measure of CPU clock speed and instruction parallelism. Find a benchmark that uses a unique instruction combination on one CPU and it will heavily disadvantage the other CPU.

Add a real world workload to the mix with heavy memory access and mixed compute workloads and the chips will diverge significantly in performance.

I work with some cross-platform code that has to run on mobile devices and desktop platforms. The advances Apple has made in low power performance are incredible, but the idea that their iPhone chips are as fast as desktop computers is still far from the truth unless you’re measuring specific, heavily optimized workloads.

I’m still excited to see what Apple can do with a full desktop level power budget though.

[saagarjha]

I suspect the "benchmarks are stupid" comments will die down the moment people start running desktop software on these chips, because it will cease to be a convenient excuse. While benchmarks may occasionally be slightly misleading, they are usually a good indicator of performance if done well–and if you've ever actually run desktop-class software on one of these chips you'll see that the divergence is just not there.