For power users, It all boils down to specs, when smartphones are put to fight head-to-head. Even though Benchmarks are synthetic, they can reveal a lot of phone’s performance in real world.
Update: We now have handson Complete iPhone 5 Benchmarks: CPU, GPU Graphics, Browser Performance vs. Android vs. iOS
iPhone 5 Benchmarks
iPhone 5 scores a score of 1601 on Geekbench. The iPhone 5 uses a custom Apple A6 ARMv7 processor which is NOT A15 as previously estimated. Also appears to be clocked higher than the A5 at 1GHz (the iPhone 4S A5 ran at 800Mhz). A6 has on-chip 1GB DDR RAM, which is half of Galaxy S3, but double of iPhone 4S. Here are the details of iPhone 5 floating point, integer, memory and bandwidth performance:
iOS Benchmarks (iPad 3, iPad 2, iPhone 4S, iPhone 4)
Galaxy SIII and other Android Benchmarks
iPhone 5 vs. Galaxy SIII Performance
iPhone 5 with Dual-core A6 scores 1601 and Galaxy SIII with quad-core Exynos scores 1628, which is only slightly higher to quad-core Tegra 3 Nexus 7 Tablet.
Update: As per another user-submitted benchmark, S3 scores 1894, which is way higher than anything in its competition. (Benchmarks source)
What’s wrong with these Benchmarks
First of all benchmarks by industry definition are supposed to be done under same or similar conditions: This essentially means: similar OS architecture, underlying platform. Since Android & iOS are totally two different platforms (Android apps run in Dalvik VM that translates to native calls vs. native on iOS), its hard to compare hardware performance against each other with good amount of accuracy. Benchmarks are designed to mimic a particular type of workload on a component or system, they do not reveal real-world performance results.
Synthetic benchmarks do this by specially created programs that impose the workload on the component. Hardware benchmarks run real-world programs on the system. While application benchmarks usually give a much better measure of real-world performance on a given system, synthetic benchmarks are useful for testing individual components, like a hard disk or networking device.
Synthetic benchmarks certainly have their usefulness since the data and performance is capable of being replicated on several different machines. It is also useful since in the world of reviews the reader very rarely has a system identical to that in the review. In such cases the synthetic benchmark allows the reader to test their system against the numbers provided in the review to gain an idea once again of relative performance.
Secondly, Geekbench, although written to be compatible with multi-processor architecture, truth be told, Geekbench 2 doesn’t use Four cores on the quadcore Android smartphones like Galaxy S III. Truth be told, only few apps on Android use all 4 cores. Most of the time, they are idle.
Other than that Samsung Stock Android ROM may not be tuned for performance. With few modifications to OS, performance can be increased 1.5x fold, easily. So simply, when you use Geekbench to compare iPhone 5 vs. Galaxy SIII, the results would be misleading.
The performance obtained from Synthetic benchmarks are essentially reflection of OS and environment performance than hardware alone.
e.g. Here is my Galaxy Note N7000 (Dual core Exynos 1.4Ghz running tuned ParanoidAndroid ROM) It clearly offers synthetic performance of HTC One X’s Quad-core Tegra 3 1.5Ghz. In real world, this doesn’t mean that my phone’s dual core is faster than quad core.
If that isn’t enough to convince you, here comes Galaxy Note N7000 (overclocked to 1.6Ghz) benchmarked on GeekBench 2.
Clearly, thats equivalent to performance of Asus Transformer Prime with quad-core CPU, and nearly 1.7x times faster than iPad 3.
Synthetic benchmarks are synthetic. We need to wait a bit longer till we can see real-life games and compute performance. For now, synthetic benchmarks declare Galaxy SIII as the winner followed by Nexus 7, and iPhone 5 taking the third place.