Re: lower power PCs

["Robert Green" <ROBERT_GREEN1963@xxxxxxxxx>]

"Marc_F_Hult" <MFHult@xxxxxxxxxxxxxxxxxxxxx> wrote in message
news:nmnd44pcp63rj4fj5dnt2vh7tf5tu82k7p@xxxxxxxxxx
> On Wed, 4 Jun 2008 10:33:10 -0400, "Robert Green"
> <ROBERT_GREEN1963@xxxxxxxxx> wrote in message
> <_bOdndS9wdtbOtvVnZ2dnUVZ_h7inZ2d@xxxxxxx>:
>
>
> >
> >Chip makers are finally seeing the light, and Via's new Nano processors
> >allegedly are drop in replacements (for the mostly soldered!) C7 chips
that
> >can handle Blu-ray decoding.  Via's previous line of chips were not known
> >for their graphical prowess.  The Nano chips all appear to use less than
one
> >watt - some as little as 100mW!
>
> BobbyG,
>
> You may have confused the Intel's Atom with Via's Nano (codename Isaiah)
or
> idle performance with actual use.

My bad.  You're correct.  The words "when idling" are missing from my
statement.  That's still quite an achievement when one considers the cost of
a server is mostly keeping it idling and ready for requests.   I believe
idling currents for the some of the newer Intel multicore CPUs is at or
beyond the "full tilt boogie" rating for the new Nano chips.  I found this
table which I assume to somewhat reflect the idle state power consumption
(TDP = Thermal Design Power & EHP = Extended Halt Power which loosely means
idling power, as far as I can tell):

Thermal Design Power

Model            Clock          TDP        EHP
QX9650         3.00 GHz     130 W     16 W
QX6850         3.00 GHz     130 W     37 W
QX6800         2.93 GHz     130 W     37 W
QX6700         2.66 GHz     130 W     50 W

> From http://www.via.com.tw/en/products/processors/eden-n/
>
> "VIA Eden-N Processor Power Consumption" is:
>
> 2.5W @ 533MHz
> 5W @ 800MHz
> 7W @ 1GHz.
>
> At higher speeds it's power use continues to increase non-linearly with
clock
> speed.
>
>
http://benchmarkreviews.com/index.php?option=com_content&task=view&id=180&It
emid=1&limit=1&limitstart=1
>
> VIA Nano L2200 processor
>  1.6GHz  17W
>
> VIA Nano L2100 processor
>  1.8GHz  25W

Yes - the table that Arstechnica supplies gives identical figures about its
power use:

http://arstechnica.com/news.ars/post/20080529-via-takes-the-wraps-off-isaiah
-meet-the-nano.html

or ( http://tinyurl.com/3p5jtj )

Model         Clock     FSB      TDP(W)   Idle power (mW)
U2300         1000       800       5.00           100
U2500         1200       800       6.80           100
U2400         1300       800       8.00           100
L2100         1800        800     25.00           500
L2200         1600        800     17.00           100

As you noted, the TDP numbers vary non-linearly.  Doubling the CPU speed
raises the watts consumed nearly 5 times. Complicating things in the real
world is that CPU power states vary, as you know, with C1 (aka Halt) which
is a "light sleep" where the processor is idling but ready to rock and roll
in milliseconds or less.  There are just too many variables beyond just CPU
power consumption involved in calculating how much juice a server will use
over a year.  That's too bad because it would be nice to figure out how soon
a new PC will pay for itself.  Part of the savings one would realize are
based on how long they want to wait for their PC to "spin back up" to full
operational readiness.  The latency of all the items involved in that
process has generally been getting shorter and shorter, making deep green
settings more practical than they used to be, at least for home office
servers.

Part of the complication comes from the many states of "idling" available on
modern PC's  It's hard to say how much power a CPU and PC are actually going
to use for the tasks intended if left on 24 by 7.  Complicating things
further, there are more "C state" subspecies, like Enhanced C1 state (C1E),
C2 (alias Stop-Clock), C3 (aka Sleep) where the processor does not maintain
its cache data, but maintains other state information.  Even further
taxonomic divisions occur with some processors having multiple C3 sleep
modes like Deep Sleep, Deeper Sleep, etc.  Each one differs in how long it
takes to awaken from the sleep state.

As I am sure you know, older machines often handle these wake ups quite
poorly, depending on what peripherals are installed, how well the firmware
was written, how well the motherboard was designed and how well the drivers
for all the added components were written.  I recall being pleasantly
surprised when a 1.7GHz AMD PC I built actually woke up from a very deep
sleep without apparent brain damage.  That lasted until I installed a ATI
video card with a video tuner built in.  It would wake up seemingly OK, but
in a few minutes, the dread blue screen would appear.  The new Asus P5K-E I
am building is a far more "green machine" than any of its predecessors
(spec-wise, anyway) because their engineers did all they could to maximize
power efficiency.   How it works in the real world remains to be seen.

> (also
>
http://laptopcom.blogspot.com/2008/05/isaiah-gets-official-with-new-name-via
.html)
> That's not much different than the current VIA processors at comparable
> speeds (although Nano is more powerful than the C7 at any given speed).

That same Arstechnica article says: "The rumors that Nano would be 2x-4x
faster than C7 in certain scenarios appear to have been true."  That's a
nice boost - using nearly the same amount of electrical power for well over
double the performance.  With the price of energy going through the roof,
low-power solutions are poised to really take off.

One interesting sign of the times is that some of the recent changes to
Intel CPUs Core 2 Duo processors allow them to tolerate 73° Celsius measured
at the surface of the processor.  That's a 12 degree rise, up from 61°C.
Even my much slower 2GHz Compaq often gets way too hot to sit on my lap.  I
can only imagine what's coming in terms of lap heat if the solution path is
to cool chips less because they can tolerate heat better.

Some day, it will seem incredibly bizarre that CPU's got hot enough to fry
eggs and memory was trusted to spinning metal platters and tiny bug-sized
read heads flying over them.  It already seems hard to believe the first
CPU's used tubes and consumed thousands of watts.   I welcome the trend
toward cooler CPU's for a reason other than power savings:  Since cooling
fans seems to be the number one cause for computer service issues, at least
in my neck of the woods, an era of fanless PC's that can actually DO
something like encode hi-res video stream will be very welcome.  I am
certain that the fewer the fans, the fewer times the machine will need
servicing.

--
Bobby G.