Paper showing Linpack scalability of mainstream clusters

Carlos O'Donell Jr. carlos at
Thu Sep 27 18:24:18 PDT 2001

> > So for the I-cluster that would have been
> >
> > 75 Gflops/(210 * 50 W/1000) = 7.14GFlops/kW
> >
> > Roger, out of curiosity could you do a similar calculation?
> I checked our UPS before and after powering up the first 128 nodes (256
> PIII 1GHz, 1GB, 20GB IDE), and came up with around 20kVA of power draw.
> I haven't converted that to Wattage (they are running at 208V,
> if anyone wants to do the math).

kVA's are simply apparent power rather than average power or Watt's.

They should be almost the same thing if the power factor of the PC supply
is relatively "good."

	P = Vrms*Irms*cos(theta_V - theta_I)

Where (theta_V - theta_I) is the phase difference between voltage and curent.

The cosine is dimensionless and thus it really is still watt's. 
And if the supply is _good_enough_ then it will approach 1 and give
you the worst case GFlops/kW ;)

So XX GFlops/(20,000W/1000) = XX GFlops / kW


More information about the Beowulf mailing list