Jouni Osmala (josmala@cc.hut.fi) on 9/18/09 wrote:
---------------------------
>>>For trivial levels of performance, yes. For anything high performance, not really
>>>(where are the high performance, low power in-order CPUs?).
>>
>>Indeed - in-order becomes very power consuming at higher frequencies, mainly due
>>to the much deeper pipeline and resulting complex forwarding/interlocking/replay
>>logic. Compare Atom or Cortex-A8 with Cortex-A9 - which is the clear winner in terms
>>of performance, power consumption AND frequency?
>
>Well they claim Cortex A8 gets <0.5mw/mhz @0.65u and Cortex A9 500mw@0.45u. with 800mhz desing.
>The overall result is 400mw@0.65 vs 500mw@0.45

The Cortex-A9 is a dual core, so just 250mW per core. Depending on how well the process scales, a 45nm A8 should be in the 200-300mW range.

> and atleast texas instruments claims
>that they get 40‰ of the power consumption in their 0.45u process AND 30‰ higher
>performance at same time compared to 0.65u.

TI can make a big improvement in 45nm OMAP3 because the current 65nm version consumes around twice as much power as expected (they use 1.35V rather than a low voltage process). Since OMAP4 will be 45nm, it will be possible to compare them directly on the same process, including at the same voltage, frequency and temperature.

>Performance and frequency no claim there. It was the high performance risc design.
>
>>>And no. You cannot compare x86 & ARM
>>>>and say they are most equivalent, since arm is the thing you use in sub 0.1W designs,
>>>>and all the way designed for low power consumption. And x86 is the thing with most
>>>>complex instruction decoders in the market today ;)
>>>
>>>Decoders that take up a whopping 4% of an Athlon core (probably less for Phenom I/II).
>>
>>It's more on the Atom though. Saying you can't compare x86 and ARM is disingenious
>>- Atom is meant to be low power and goes after the same markets as Cortex-A8/A9.
>
>What I wanted to say with original comment was that comparing ATOM power consumption
>to ARM power consumption and claiming because ARM has feature X and atom doesn't
>it means feature X doesn't cost power just because they happen to consume same amount of power thats disingenious.

Sure, they are not identical. But being able to design an OoO core that is significantly faster while using half the power of an Atom is pretty amazing in my book. It will force Intel to introduce competitive designs in this space unless they prefer ARM taking over their low-end.

>I probably should stop reading these posts. Cannot resist stopping to write, and
>that costs too much time, I'm never gonna get my PHD if I don't make cuts on my spending on times on other things.

:-)

Wilco