[HPGMG Forum] Kiviat diagrams
mfadams at lbl.gov
Tue May 27 12:31:33 UTC 2014
cc'ing the forum.
On Tue, May 27, 2014 at 12:44 AM, John Shalf <jshalf at lbl.gov> wrote:
> The Kiviat diagram is not intended to be a standalone figure to assess the
> value of the benchmark. All it tells you is whether the benchmark
> exercises different aspects of the hardware in equal measure (it does not
> measure algorithmic efficiency).
> If you are concerned about the Kiviat not being sufficient as a standalone
> visual reference, then add your favorite FOM to the Kiviat.
I am not but Sam seems to be and perhaps I should reconsider using it.
Bert and Ian have application data and I was intending to integrate our
data with very little work required of them. I don't want to let the great
be the enemy of the good and not do anything because we don't have the
resources to do it well. Part of what I want to convey is that a critical
part of this process is fitting metrics of the benchmark to those of the
solver, to be a good benchmark. This is the only data that I can show that
is an example of this. I want data. I love the Bloomberg line "In god we
trust; everyone else bring data".
> However, I think this is counterproductive since our point is to come up
> with a mathematically rigorous algorithm that exercises the
"challenging" parts of the hardware architecture. Having the most
> efficient DOF/s for this particular solver is orthogonal. You want to
> focus on mathematical rigor (in the same way that Linpack has a
> mathematically rigorous definition) and exercising the hardware aspects
One of my narratives (see my slides) is that we are staying faithful to the
HPL model by: 1) solving a hard problem (fully connected global problem
with known mathematical complexity), 2) using the best method to solve it,
3) using the best computer science to implement it. I would avoid the term
"mathematical rigor", because it has baggage (the 5% of the audience with
math Ph.D.s would have a panic attack :)
I think it is very good to work out some of these ideas on the forum as we
are not colocated and I would like to build consensus on these basic
philosophical issues where possible.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the HPGMG-Forum