none (none@none.com) on 9/22/09 wrote:
---------------------------
>? (0xe2.0x9a.0x9b@gmail.com) on 9/22/09 wrote:
>---------------------------
>>Right?
>
>No, as IntelUser2000 pointed out, you are off by a factor
>of 2 for A9:
>
>
>

Score	Processor
23044 (5761/core)	4 >cores on Cortex-A9 @2GHz(?)
2*4058 = 8116 (2029/core)	hypothetical >4 cores on two Atoms @1.6GHz
24828 (6207/core)	4 cores on PhenomII @2.6Ghz
39201 >(9800/core)	4 cores on Core2 Quad 6700 2.667GHz

>
>>So, how exactly is Cortex-A9 "competitive with the latest x86 cores" like PhenomII?
>>
>>If Atom is overclocked to 2Ghz, the Cortex-A9 at (supposedly) 2GHz still beats
>>it - but *not* by a wide margin. And it even has the advantage of OoO execution
>>against the "supposedly poorly designed" in-order Atom.
>>
>>Bottom line: If you think A9 will be pulling miracles as if designed by an advanced
>>alien race from outer space, then you are plain wrong.
>
>I guess will have to update your conclusion based on the
>above fix :-)
>
>Also to be taken into consideration: only a dual core A9 @2GHz
>was announced. And the Atom score was with HT, so perhaps
>one should compare a dual core A9 vs a dual core including
>HT. So that'd be 11522 for A9 vs 10145 for Atom.

That is with a dual core Atom overclocked to 2GHz with HT enabled (without HT it would probably be around 7000-8000). There is no such Atom today, the dual core hyperthreaded ones max out at 1.6GHz, and use 8W at that speed.

>Anyway that benchmark leaves at least two questions
>unanswered: what about the memory system performance? And
>what about coherency performance?

Historically EEMBC benchmarks have been hovering between awfully bad and ridiculously bad, so it's no wonder there are so few published scores. This new benchmark is supposed to replace Dhrystone, and that's quite a challenge. I haven't looked at it yet, so I am keeping an open mind. However it is a micro benchmark, it will not stress the memory system besides the L1. That's why the scores scale well.

Wilco