Saturday, May 1, 2010

Benchmarks Atom vs iPad A4 vs iPhone 3GS ARM Cortex and much more...

See update here

With the iPad, Apple is creating a new type of device that got some similarities with the actual netbooks. It is known that the iPad got an Apple A4 processor clocked at 1GHz, it isn't clear yet which type of ARM core is really used: Cortex A8, A9, a customized version? It is very interesting to understand how this new processor compare to other ARM processors (ARM11 in iPhone 3G and Cortex A8 in iPhone 3GS) and to the Intel Atom processors.
It is really tough to compare performance of CPU with different architecture, running different operating systems and especially targeting very different applications.
Since years ARM claims superior performance for the Cortex A8 and A9 compared to Intel Atom. Now I could not resist, especially because the benchmarking race started and I finally got a critical mass of benchmarking data on Atom vs ARM performance.

FBenchmark iPad vs iPhone vs Atom netbook CoreMark. Cortex A8 vs Cortex A9 vs Apple A4 vs Intel Atom vs Nvidia Tegra 2irst of all a clarification: here we talk about benchmarking of CPU cores, it has little to do with comparing performance of the iPhone vs iPad or iPad vs netbooks. If you want to compare two devices you have to find first a common use case and metrics to measure it, for example Anandtech published browsing benchmarks showing that ARM Cortex cores in iPhone 3GS and iPad are much slower than Atom in one of the most important use cases: Internet browsing.

The benchmarking ARM vs Atom race started and I finally got a critical mass of benchmarking data: ARM, the Linley group and the german magazine C't published CoreMark benchmarks for many ARM cores and Intel Atoms.
EEMBC CoreMark is a good metrics of the pure processing power of the CPU core, the algorithm is pretty small and fits in level 1 cache. CoreMark basically replaces the old Million Instruction Per Second (also called as MIPS, not to be exanged with the MIPS company ain direct competition with ARM...)

I created a chart with normalized CoreMark/MHz for each of the result I got.
The result for Atom processors seems stable around 2,5-2,8 CoreMark/MHz, I don't have a clear bottom line for the Cortex processors. The best results for Cortex A8 and A9 probably derive from tests done in best case conditions in development boards (e.g. for TI OMAP, Freescale i.MX515 and Samsung S5PC110) while real life products (such as iPhone and iPad) got much lower results.
Until the test conditions are clarified is not possible to state who really wins!!


Anonymous said...

Only the Atom N280 figures are source: Linley Group.

Joe Byrne

djsm said...

Joe, with source Linley group I mean the Linley blog (