Saturday, September 4, 2010

Benchmarks for Smartphone and Tablet (ARM Cortex vs Intel Atom)

I already published some CoreMark results for ARM processors found in high end smartphones (e.g. iPhone 3G/3GS and iPhone 4) as well as netbooks.

After the Apple iPad revolution a number of tablet clones is invading the market and there is a lot of confusion with the used processors. Very ofter you only hear something like 600MHz or 1GHz processor, this can be misleading since there is a huge difference between different cores even though they are running at the same clock!

ARM Cortex A8 and A9 are way more powerful than ARM9 and ARM11 cores!
Below CoreMark benchmark results for ARM9, ARM11, Cortex A8, Cortex A9 and Intel Atom processors.
I also created a summary table reporting which ARM core you find in different SoCs.

Click on images to enlarge!

1 comment:

Anonymous said...

one thing is clear today when your comparing ARM to x86 for speed you NEED a real life test and that means x264 encoding as per below for these low power devices to be comparable.

and OC keeping on mind x86 has more SIMD than ARM cortex at this time.

you might want to pop over to the #x264dev IRC channel and post some ARM SIMD patches to help out soon.

this real life Benchmark test came from the old x264-dev logs if anyone's interested in the real life Number's.

and StippenG's number's came from an older Quad A9/NEON developer board at his Uni apparently, so a current Marvell ARM v7 A9 quad /SIMD at 1.6 GHz for instance would produce a Far better result today OC.

640x360 at Ultrafast: 38.59 seems like a very good real life start for encoding on ARM cortex even without those extra SIMD patches being written yet

"2010-08-24 15:39:19 < StippenG> Some X264 Benchmarks (Rush Hour 640x360, preset=medium, crf=24): 4-core Cortex-A9 @ 400 MHz gives 5.55 fps, Beagleboard (A8 @ 720MHz) gives 1,65. Really nice speedup, considering the much higher frequency of the A8

2010-08-24 15:39:56 < Dark_Shikari> It'd go a lot faster if you used a faster preset.
2010-08-24 15:40:01 < Dark_Shikari> Or if you wrote some of the asm we're missing
2010-08-24 15:40:26 < Dark_Shikari> But yeah, that scales surprisingly well. about 3.5x faster

2010-08-24 15:40:27 < StippenG> Yes. Superfast gives 22.07. Ultrafast: 38.59
2010-08-24 15:41:23 < StippenG> Guess the out-of-order execution and shorter pipeline is really quite a bit better for performance

2010-08-24 15:41:35 < Dark_Shikari> Well, the A9 is known to be a lot faster