[Beowulf] The Walmart Compute Node?
jimlux at jpl.nasa.gov
jimlux at jpl.nasa.gov
Fri Nov 9 06:16:38 PST 2007
Quoting "Robert G. Brown" <rgb at phy.duke.edu>, on Fri 09 Nov 2007
04:49:13 AM PST:
> On Thu, 8 Nov 2007, Jim Lux wrote:
> <snip of my deathless prose... no really, the internet archive will
> preserve ALL our words, useful or not, long after we shuffle off
> this mortal coil>
>> from zero to one or vice versa). So, in general, a 2N GHz
>> processor consumes less than twice the power of a N GHz processor.
<snip of RGB's deathless prose>
> 100 watts seems like a pretty solid upper bound for what anybody
> releases a processor at, IIRC from over the various times I've looked at
> per-processor draw (and I don't remember seeing any that spec'd out over
> something in the low to middle 90's). I'm guessing that this is a
> physical limitation in terms of how much heat you can lose with a big
> heat sink and fan, recalling that the E-Z Bake oven used to bake
> brownies and cookies on the power output from a single 100 W bulb. I
> don't know if current multicores violate this rule -- I'd be a bit
> surprised if they did, though, especially if it is a heat sink issue.
Here's the things that drive that 100W number:
1) the difficulty of getting the heat out of the chip (the physical
package is only so big, the die is only so big, you can only
practically get a certain thermal flux out and keep the die
temperature reasonable with the "next temperature in the chain" being
- physical size limits power dissipation
2) if you radically changed the power consumption, all the power
distribution design heritage for motherboards and power supplies would
change. Power supplies today are pretty much like power supplies of
yesterdecade, perhaps with a 3.3V output and different connectors.
3) The airflow and fan configuration in the package would need to be
changed. Note that we're still mostly buying boxes that don't look too
different from the original IBM PC or PC/AT. Over the years, the mobo
mfrs have figured out how to place the components on the mobo so that
when installed in a conventional tower or desktop case, the
temperatures are reasonable. Good thermal design is expensive.
Commodity computer manufacturers tend to not want to spend any money
on new designs.
4) There's an ultimate limit set by the 7Amp rating of the usual IEC
power cord and receptacle. Recall that historically, one would plug
one's (power consuming) CRT monitor into the back of the PC, so all
the juice flowed through the cord. 300W for the monitor + 300W for
the PC = 600W.. just a bit of margin below the 700-800W possible for a
7 Amp cordset. 300W for PC -> 50-60% efficient power supply ->
150-200W for the electronics, half of which goes to CPU, half to
everything else (disk drive, memory, blinky lights on panel, etc.)
More information about the Beowulf