Robert David Graham (robert_david_graham.delete@this.yahoo.com) on January 21, 2017 5:34 pm wrote:
> Maynard Handley (name99.delete@this.name99.org) on January 20, 2017 8:55 pm wrote:
> >
> > Moreover, as I have pointed out before, the world consists of many compute tasks, some of which look very
> > different from dense linear algebra. For example #81, Monty
> > Pi-thon, on the Graph 500 is built out of a number
> > of Raspberry Pi's, and there are a few other such weird machines (some based on ARM) on the list.
> >
> > http://www.graph500.org/results_nov_2016
>
>
> Uh, I think you are confused as to what this "Graph 500" is measuring. The
> "Monty Pi-thon" contains 32 Raspberry Pis, which are less powerful (while
> using more electricity) than a single quad-code Intel desktop computer.

Thank you for pointing that out. I think that illustrates that the Graph500 is really just a joke of a data set if that is a winning entry.

Benchmarks are a valuable tool to understand the world...but only if you understand what they measure. In this case, the Graph500 appears to measure 261 computers that have admins with enough spare time and cycles to run the Graph500.

I'm filing that away under: "Things I will ignore or mock" for the future.

> I think the issue is less dumb nationalism than the dumb belief that ARM
> will deliver better performance/watt for supercomputer applications.

I am pretty sure there's a healthy dose of the former. Especially in the EU and China. The latter has a variety of efforts using different ISAs (Alpha, MIPS, ARM, Power). Clearly China could care less and just wants to develop the expertise and local capability. That potentially hurts Intel or IBM, but it's not obvious who that helps. RISC-V has got to be pretty attractive once it is complete.

David