/. US DOE gets a $24.5 Million Linux Supercomputer

[Greg Lindahl]

On Wed, Apr 17, 2002 at 02:31:31PM -0500, Richard Walsh wrote:

> 8.3 TFLOPS (if they are quoting peak) with 1400 processors 
> would mean they are getting chips with 1.5 GHz clocks (peak
> performance would be 6 GFLOPS per chip [4 ops per clock]). 

This bid is for an install in the future, and it involves a
combination of McKinley and Madison parts. I don't believe that Intel
has made Madison's specs available, nor has HP made the specs of the
chipset they'll be using available.

It's likely that they aren't quoting peak; PNL prefers figures like
the actual speed of matrix-matrix multiple (DGEMM). Now the Itanium is
reasonably good at delivering a nice % of peak for DGEMM, but it's not
the same as peak. It's a lot more fair number to use than peak, and
gives you a good idea of what the Top500 Linpack score will be.

> this is $70 per MFLOPS sustained 
> for the Mckinley (again using triad) ... or more than the CRAY SV2 
> ($65), EV6($55), EV7 ($50), Pentium 4 ($30).

And Cray SV2 figures aren't publically available either. Cough.

> but you would think that QCHEM stuff 
> they run at PNNL (G98) will be mostly memory bound and therefore 
> the stream triad sustained performance is not too far off.

As you might guess, the bid required that you benchmark PNL's actual
codes at PNL's actual data sizes. I don't believe that your analysis
is correct.

Alas, I can no longer say that FSL was the only time a Linux cluster
won a traditional supercomputing bid.

greg