[Beowulf] Demo-cluster ideas

[Lux, Jim (337C)]

What I find interesting about this is that there's only a 3:1 difference between high and low.

That's a pretty compelling argument that if you need a 10x speedup, you're not going to get it by "buy a faster computer", and that, rather, parallelism is your friend.

It would be interesting to look at this on a ops/$ or ops/Watt basis.. My guess will be that there's not a huge power difference (in terms of wall plug power) between the various options (certainly not 3:1), and there's probably not a 3:1 difference in price, either.

I was looking at benchmarks for substantially lower end machines recently (ARMs, Atoms, bottom of the line i5s) for a signal processing problem and there's not a huge difference there, either.  To get a 10x difference, you have to be going WAY different technologies (like 8 bit microcontrollers or similar).



Jim

-----Original Message-----
From: Beowulf [mailto:beowulf-bounces at beowulf.org] On Behalf Of Douglas Eadline
Sent: Wednesday, February 03, 2016 6:58 AM
To: Olli-Pekka Lehto <olli-pekka.lehto at csc.fi>
Cc: beowulf at beowulf.org
Subject: Re: [Beowulf] Demo-cluster ideas


To help gauge your performance, you can find HPL results for my Limulus machines here (Sandybridge to Skylake):

  http://limulus.basement-supercomputing.com/wiki/LimulusBenchmarks

I have NAS parallel results as well, I'll be posting them on Cluster Monkey real soon.