Re: [gentoo-embedded] personal compile-farm ?

[Christopher Friedt]

Hi Janusz,

On Sat, Sep 5, 2009 at 11:31 PM,  <jsyrytczyk@×××××××××.pl> wrote:
> Hi, I got Atom 330 with this  motherboard sice  six months or so.
>
> http://www.intel.com/products/desktop/motherboards/D945GCLF2-D945GCLF2D/D945GCLF2-D945GCLF2D-overview.htm
>
> The machine runs all right, but the compilation times are similar to *very*
> outdated Celeron 1.7 GHz (which is one core only). The Atom goes very snappy
> on every other task it performs (software mirror, backup, rsync, remote NX
> station, email, iscsi, puppet, pulseaudio server etc.),
>
> but
>
> compilation times *are* slow.

Your reply has been the most relevent to date, having been the only
person who has an Atom 330 device for reference.

After reading about your compilation speeds, I did a bit of googling
for "intel core 2 duo vs intel atom 330", and found [1] which reveals
quite a lot. It would seem that all of the Atom 330 "slowdowns" are
caused by memory latency.

It seems, that the Atom (both the 230 and 330) were not designed to
use the blazingly fast FSB frequencies that all other modern Intel
processors use, which is likely the primary reason for (sub-par)
performance, and probably also the reason for their low power
consumption. Most of the 45nm Core-2 processors support FSB
frequencies around 1 or 1.3 GHz. On the other hand, the Atom 330 only
supports FSB frequencies of 533 MHz. In terms of silicon / FETs, high
clock speeds == high power leakage.

So essentially, if the clock speeds of a Duo and Atom core were the
same, then the Atom would require twice as much time as the Duo for
the the same amount of "work" (i.e. memory reads / writes). Assuming
that the Atom in-use power is about 1/2 of the duo, then both systems
consume the same amount of power for a "task", but the Atom takes
twice as long.

In reality, the Atom consumes over half of the Duo in-use power.
Therefore, power-efficiency ironically favours the Intel Core-2 Duo
rather than the Atom for computation-intensive applications. For
multimedia, I would say that the Atom is slightly more
power-efficient.

In my estimation, the lower in-use power of the Atom would be lost if
it used a 1.3 GHz FSB controller. Does anyone disagree?

Conclusion:

The main bottleneck on the Atom 330 is not the CPU frequency, but
rather the FSB frequency. Therefore, for a dedicated HTPC and / or NAS
device, the an Intel Atom 330 device is a good choice. For a
dedicated, low-power build machine, the Atom 330 is a bad choice for
performance, but a good choice if only moderate performance is
required.

For a box that is intended to be used for HTPC / NAS and also a
dedicated low-power build machine, an Atom 330 device is still a
decent choice, because at least it performs efficiently for 2 out of 3
functions, and it's unlikely (physically impossible?) that one will
find a comparable dual-core, low-power, fanless (and not liquid
cooled) device with a 1.3 GHz FSB.

So ... yea, I think I'll probably grab one of these ZOTAC boards
anyway, at least for having an HTPC. Using it as a dedicated build
machine would still be useful, even if the performance isn't
particularly great. In any event, it won't be building packages
constantly for my purposes, but only periodically.



Cheers,

Chris


[1] http://www.tomshardware.com/reviews/dual-core-atom-330,2141-6.html
[2] http://www.intel.com/products/processor/core2duo/specifications.htm
[3] http://processorfinder.intel.com/details.aspx?sSpec=SLG9Y