[Arm-netbook] Selecting the right Soc for out ARM Notebook project.

[Gordan Bobic]

On 06/05/2012 18:46, Lauri Kasanen wrote:
> On Sun, 06 May 2012 15:20:34 +0100
> Gordan Bobic<gordan at bobich.net>  wrote:
>
>> Yes, but most of those are synthetic benchmarks that are rather FP
>> heavy, such as rendering, AV encoding, etc.
>
> Well, I do rendering monthly, encoding about weekly and compiling daily. So I wouldn't quite discount those benchmarks.

Maybe so, but transcoding and rendering is something that has always 
required (and always will require) lots of CPU power, so it isn't 
exactly the sort of thing that Atom and ARMs are aimed at.

>> In Apachebench it shows rough equivalence clock-for-clock (which puts
>> the A9 massively ahead in performance per watt), and 7z test shows them
>> to be about the same which puts A9 considerably ahead of the Atom
>> clock-for-clock.
>
> Apache is a multiprocess test, so the A9 only got half the perf per clock?
 >
> 1753/1600 = 1.09 hits/Mhz
> 1362/1200/2 = 0.56 hits/Mhz (dual core)

That does surprise me. Hmm... Memory/network I/O? Was the test done over 
loopback? I can certainly believe that the Atom has a LOT more memory 
I/O than an A9 (double the memory bus width for a start).

>> It is also interesting that the Atom got soundly beaten by the A9 in the
>> OpenSSL RSA test. I wonder if this test is multi-threaded which would
>> clearly put the A9 at an advantage since N270 only has a single core. If
>> it isn't then that is quite embarrasing for the Atom.
>
> It's single-threaded, IIRC.

That is quite shocking poor for the Atom, then.

Gordan